Acoustic Emission Properties of

Partial Discharges in the

time-domain and their applications

PATRICK JANUS

Degree project in
Electromagnetic Engineering
Masters' Degree Project
Stockholm, Sweden 2012

XR-EE-ETK 2012:004

Acoustic Emission Properties of Partial Discharges

in the time-domain and their applications

PATRICK JANUS

Stockholm 2012
Electromagnetic Engineering
School of Electrical Engineering
Kungliga Tekniska Hgskolan

XR-EE-ETK 2012:004

ABSTRACT
A partial discharge (PD) is an unwanted phenomenon in electrical equipment. Therefore it is
of great importance to identify different types of PD and assess their severity. This thesis
investigates the acoustic emissions associated with PD from different types of sources in the
time-domain and based on these results a low-cost detection system is designed, built and
verified against much more expensive commercially available sensor platforms. The results
are promising both for the detection of different types of PD and their classification. The lowcost detection system shows equal usefulness as commercially available systems and the
system is also compared against other systems for detection. Also some investigation is
carried out concerning the nature of the ultrasounds emitted by PD. The results from these
experiments indicate that a description with the classical wave equation is inadequate. In
order to explain the frequency content of acoustic PD signals at different distances away from
the source, a more complex description in the form of nonlinear acoustic equations may be
more accurate.

-1-

-2-

ACKNOWLEDGEMENTS

I would like to express my deepest gratitude towards Associate Professor Hans Edin for
believing in me and giving me the opportunity to conduct this thesis. I would also like to
thank him for showing great patience even towards the most craziest ideas and for his
confidence in my abilities.
I would also like to thank Msc, PhD student Mohamad Niasar Ghaffarian for his knowledge
and help in the laboratory, no matter at what time. His selfless help and support aided me in
the darkest hours of this thesis for which I am forever grateful.
I wish to extend my gratitude towards the Innoenergy projects KIC-ASS and CIPOWER,
whose financial aid has been invaluable for me as to the purchase of new sensors and
equipment required for the thesis work.
To the entire team working in the high voltage lab I wish to thank for the friendly and
stimulating environment that they create.
I wish to thank my friend, John Laury for his support and aid throughout all of this time that
we have spent together, both in- and outside of school.
Finally, but not least, I thank my loving wife Natalia for her outstanding patience with me
working through the nights, late phone calls, and so on.

-3-

-4-

TABLE OF CONTENTS
1. INTRODUCTION ................................................................................................................ 7
1.1 PD in a historical perspective ........................................................................................ 7
1.2 Classification of PD ........................................................................................................ 7
1.3 PD detection methods..................................................................................................... 8
1.3.1 Electrical Detection ................................................................................................. 8
1.3.2 Chemical detection .................................................................................................. 9
1.3.3 Acoustic detection .................................................................................................... 9
1.4 Low-cost ultrasonic detectors ...................................................................................... 10
1.5 Data acquisition methods............................................................................................. 10
1.6 Aims ............................................................................................................................... 11
1.7 Disposition ..................................................................................................................... 11
2. ULTRASOUND GENERATED BY PD ........................................................................... 13
2.1 Partial discharges and ultrasound .............................................................................. 13
2.2 The nature of ultrasound / nonlinear acoustics ......................................................... 13
2.2.1 Fubinis solution .................................................................................................... 14
2.2.2. The Burgers equation........................................................................................... 15
2.2.3. The KZK (Khokhlov-Zabolotskaya-Kuznetsov) equation ............................... 15
2.3 Detection of ultrasound ................................................................................................ 16
3. TRANSDUCERS ................................................................................................................ 18
3.1 Transducers used in thesis ........................................................................................... 18
3.2 Expected ultrasonic and electric parameters ............................................................. 19
4. AMPLIFIERS AND DETECTORS ................................................................................. 20
4.1 Overview of needed amplifiers and detectors ............................................................ 20
4.2 Band-pass detection system ......................................................................................... 20
4.2.1 Design aspects ........................................................................................................ 20
4.2.2 Modelling................................................................................................................ 23
4.2.3 Implementation...................................................................................................... 24
4.2.4 Verification ............................................................................................................ 25
4.3 Broadband amplifier .................................................................................................... 26
4.3.1 Design aspects ........................................................................................................ 27
4.3.2 Modelling................................................................................................................ 28
4.3.3 Implementation...................................................................................................... 29
5. LABORATORY SETUP FOR PD DETECTION USING ULTRASOUND ................ 31
5.1 Experimental setup ...................................................................................................... 31
5.2 Phase resolved PD measurement equipment ............................................................. 32
5.3 Test objects .................................................................................................................... 33
5.3.1. Test object for Corona in air ............................................................................... 33
5.3.2. Canonical Test objects for PD in transformers ................................................. 33

-5-

6. Results ................................................................................................................................. 35
6.1 PD patterns for different situations ............................................................................ 35
6.1.1 Corona in air .......................................................................................................... 35
6.1.2 Corona in oil .......................................................................................................... 36
6.1.3 Surface discharge .................................................................................................. 37
6.1.4 Void discharge ....................................................................................................... 38
6.2 Frequency dependence in PD signals as function of the distance ............................ 38
6.3 Comparison of PD Patterns for different situations ................................................. 39
6.4 Comparison of different PD detection methods ........................................................ 40
6.4.1 Corona in oil .......................................................................................................... 40
6.4.2 Surface discharge in oil ......................................................................................... 42
7. CONCLUSIONS................................................................................................................. 44
8. FUTURE WORK ............................................................................................................... 45
9. REFERENCES ................................................................................................................... 46
APPENDIX A: DESIGN PROCESS OF A TRANSISTOR AMPLIFIER ....................... 48
APPENDIX B: DESIGN OF NARROWBAND AMPLIFIER .......................................... 51
APPENDIX C: DATASHEETS FOR KNOWLES ACOUSTICS SPM0204UD5 ........... 56
APPENDIX D: DATASHEETS FOR MURATA MA40MF14-0B .................................... 57
APPENDIX E: DATASHEETS FOR PHYSICAL ACOUSTICS SENSORS .................. 58

-6-

1. INTRODUCTION
1.1 PD in a historical perspective
Even though electrical discharges have been known to man since the dawn of time in the form
of lightnings, the study of partial discharges is a relatively new branch on the evolutional tree
of electrical engineering. One milestone in the technical area came in 1963 with the release of
the draft of VDE 0433/63, part 4, Guidelines for measuring set-ups for corona insulation
tests and part 5 Guidelines for corona tests on power system equipments with alternating
voltages up to 500 Hz. In conjunction with this release a technical conference with the theme
Early recognition of defects in the insulation of windings was held in Germany. This
conference mainly addressed itself to the questions and problems of partial discharges in
electrical apparatus, especially their measurement and evaluation. This meeting proved to be
of great importance since it provided a forum for discussing questions which went well
beyond the knowledge and experience of old-school power engineers. The greatest
breakthrough of this conference was that it introduced the concept of non-destructive testing
techniques that up until then had only been known to a certain group of technical persons was
now introduced to a broader audience. The introduction of these testing techniques, for
instance, lead to a drastic reduction of failures in epoxy insulated instrument transformers,
which were till then very high and in some cases not at all understandable. [1]
Since then a lot of work has been conducted to refine the techniques used for testing and also
a great effort has been made to further understand the underlying mechanisms of discharges,
but a lot of work still remains until all secrets of partial discharges will be revealed
1.2 Classification of PD
The basis for the nomenclature on partial discharges (PD) is the IEC standard 60270 - HighVoltage Test Techniques Partial Discharge Measurements, which contains among other
things the basic definitions in context of partial discharges.
The definition of a PD is according to IEC standard 60270, a localised dielectric breakdown
of a small portion of a solid or fluid electrical insulation system under high voltage stress,
which does not bridge the space between two conductors.
The partial discharges are then divided into two major sub-groups, internal and external PD,
depending on their occurrence in a particular set-up, as seen in fig. 1.

Fig 1: Terminology

-7-

These two sub-groups are internal and external discharges. Each of these sub-groups can be
represented by a number of different experimental set-ups which can be treated as typical
cases which can be seen in fig. 2. These typical examples are of great significance when
conducting laboratory experiments in order to simulate different situations in which PD may
occur.

Fig 2: Idealized configurations for internal and external partial discharges used in laboratories [1]

1.3 PD detection methods

Over time, electrical insulation subjected to high electrical fields begins to breakdown due to
mechanical, thermal and electrical stress. Because partial discharges are both symptomatic of
insulation breakdown and a mechanism for further insulation damage, PD detection is used to
evaluate the condition of and diagnose problems with the insulation [2].
Over the past fifty years, several methods have been developed to detect PD. These methods
fall into one of four categories: electrical, chemical, acoustical and optical. The optical
detection method is not used to any known extent and is therefore omitted in the following
description of the methods.
1.3.1 Electrical Detection
Electrical detection is based on the electrical pulse created by the current streamer in the void.
These single pulses last in the order of nanoseconds and have measurable frequency
components in excess of 1 MHz [3]. The pulse shape, its relative phase location within the
AC cycle, and the signal intensity all lead to information about the type of PD fault and the
severity of the insulation damage.
Electrical measurements are grouped into two categories, direct probing and RF emission
testing. The direct probing method requires that capacitive couplers be connected to the phase
terminals of the transformer. The second group, RF emission testing, is conducted by using
antennas inside mainly transformers.
The main problem with electrical detection is the equipment susceptibility to noise, which can
lead to false detection of PD. Usually the equipment subjected to testing also needs to be
taken off-line, energized from a high-voltage source and then tested, this requires a lot of time
and equipment[4]. This in turn results in high cost for the owner [3]. Despite these drawbacks, this type of detection system is currently the most accurate and provides equipment
managers with valuable information about the condition of equipment [5].

-8-

1.3.2 Chemical detection

Partial discharges can be detected chemically since current streamers break down the
surrounding materials into different chemical components. The two primary chemical tests
that are in use today are dissolved gas analysis (DGA) and high performance liquid
chromatography (HPLC). It is worth mentioning that these tests only can be performed on oilfilled components, mainly transformers, which is a limitation in their application.
The DGA test identifies gases dissolved in the oil produced by the breakdown of mineral oil.
The test is performed by taking an oil sample and determining the levels of different dissolved
gases, manly hydrogen but also acetylene, methane, carbon oxide, carbon dioxide, and
ethylene [5]. This test indicates the presence of PD but also provides additional information
because different levels of each of the gases can be correlated to a specific type of fault.
HPLC test measure the by-products of transformer wall insulation breakdown, i.e. cellulose
and its by-products. The disadvantage of this test is that glucose levels in the oil are very
small and the degraded forms of glucose are not very stable. In addition, this test suffers the
same uncertainty as DGA because there are no standard values for glucose concentration and
its correlation to faults [5].
The limitations of chemical testing are that these methods do not provide any information
about the position or extent of the PD and/or insulation damage. An interesting development
in this field is currently being conducted by adding certain chemical tags to specific
portions of the insulation. These tags are released in the oil when the insulation breaks down
and can be detected. The problem with this method is that it has a fair amount of uncertainty
regarding the severity of the PD activity, also that this method doesnt provide instantaneous
information about the state of the component being evaluated. [6]
1.3.3 Acoustic detection
Acoustic detection is based on the phenomenon that PD emits sound. This sound, audible or
not, is created when the current streamer is formed and the material surrounding the streamer
is vaporized. This vaporization causes a rapid release of mechanical energy, which propagates
in the form of a pressure field [7]. The best analogy which is easily understood is that of the
formation of thunder after a lightning strike.
Acoustic detection systems can be broken into two categories: external and internal systems.
External acoustic detection systems, currently more utilized in power systems, employ sensors
mounted to the outside of power equipment to detect the PD acoustic signal. Internal systems
on the other hand use sensors placed inside of power equipment to directly measure the
pressure wave.
The primary advantage of acoustic detection over other methods is that positioning
information is available by using sensors at multiple locations. This position information can
help to identify the type of PD and the severity of an insulation fault. Another advantage of
acoustic detection over electrical is its immunity to electromagnetic interference (EMI). This
immunity to EMI makes acoustic detection ideal for online PD detection because a better
signal to noise ratio (SNR) for the acoustic signal will lead to fewer false alarms. [8][9][10]
But immunity to EMI does not mean that there is no acoustic noise in the system.

-9-

Mechanical vibrations in transformer cores are the primary source of acoustic noise, but the
frequency content of these vibrations are sufficiently lower than the PD acoustic signal.
Acoustic detection also has its limitations. The main problem with acoustic detection is the
complex nature of the acoustic propagation pathways, which is outlined later. There is also
the problem that received acoustic signals have very low intensity so sensors must be very
responsive to small changes in signal amplitude in order to detect a PD [5].
One other very important issue which currently limits the usefulness of acoustic PD detection
equipment is the high price for these units. Therefore the question arises if such detection
systems cannot be made cheaper by using mass produced components like commercially
available piezoelectric transducers and other off-the-shelf components.
1.4 Low-cost ultrasonic detectors
As a part of this thesis a low-cost but efficient PD detector was constructed and evaluated in
conjunction with the commercially available systems in order to investigate if there are ways
of making these sensor platforms interesting enough for commercial introduction. The
benefits of cost-efficient systems are that grid and power company owners will have the
opportunity to install them in places where up until now a surveillance system was too
expensive to install. This will hopefully lead to a more reliable power system with fewer
outages even in smaller components thanks to increased surveillance.
As previously mentioned, this amplifier will be constructed so that the per-unit production
cost is kept as low as possible. Therefore all of the components used for this sensor should be
available in large quantities on the market at affordable prices.
The detector should be able to detect both internal and external PD, both from AC and DC
sources; it should have both electrical and acoustical means of detection. If a PD is detected
an audible signal should be heard through a pair of headphones and a measurable electrical
signal should be provided to an oscilloscope or similar device and the sensitivity should
match that of commercially available sensor systems.
1.5 Data acquisition methods
Signals from both electrical and acoustical measurements are in one way or another digitally
processed, and here a brief explanation is given on the function of one such processing
method which is common to all commercially available equipment for PD analysis.

Fig. 3: Typical emission signal characteristics

Fig. 4: Phase-resolved PD pattern for a surface discharge

in air on pressboard .

-10-

The method in question, the hit principle, is outlined in fig. 3. The basis for this measurement
is that the system reacts when the signal crosses a threshold and then it measures the
following parameters: amplitude, rise time and duration. In order for this system to be
accurate a good SNR (signal-to-noise) ratio is necessary. The better this ratio is, the more
sensitive the apparatus gets. [11]
By grouping these parameters according to their specific parameters and their relative position
against the supply voltage, a chart can be produced, and from patterns within these charts it is
possible to draw conclusions as to the type of PD, and the severity of such discharges. Such a
phase-resolved partial discharge pattern (PRPD) is presented in fig. 4.
1.6 Aims
The main goal of this thesis is to investigate the following aspects of partial discharge
detection using acoustic emission techniques
1) Distinguish different types of PD using acoustic emission techniques according to the
time-domain behaviour of individual pulses.
2) Design and implement a cost-effective detector for acoustic detection of PD using offthe-shelf components and evaluate its usefulness compared to commercially available
systems.
3) For the commercially available transducers a preamplifier with an adjustable gain of
20, 40 and 60 dB is needed. Since these preamplifiers cost several hundred Euros and
the budget for the thesis is limited a selectable-gain amplifier will be constructed
based on the technical specifications of the commercially available amplifiers.
The first issue is of interest since few efforts have been made to explain the time-domain
behaviour of an acoustic PD signal and the type of PD associated with it.
The second issue is of interest since commercially available sensors are so expensive that it
isnt cost-effective to mount these in remote places and in small installations. If a sensor could
be developed cheap enough with adequate performance it could help increasing the safety of
power system by alerting of possible faults where they now only are discovered when
something really breaks and causes outages and/or disruptions.
The third issue is connected to the second, since the cost of commercially available
components is very high, and this thesis consists in part to lower the cost of such devices and
to demonstrate how this can be achieved.
1.7 Disposition
This thesis is organized into 9 chapters as follows.
Chapter 1: Introduction and general information which covers the basics concerning PD and
motivation for the thesis work.
Chapter 2: This chapter contains a more theoretical description of ultrasonic phenomena in
connection to partial discharges and how these are detected.

-11-

Chapter 3: Describes the transducers used for this thesis work and their expected working
parameters.
Chapter 4: Describes the amplifiers and detectors which were designed and built during the
course of this master thesis. The first part gives a short outline and motivation for the different
types which is then explained in more detail
Chapter 5: First the current lab environment with all the necessary auxiliary equipment is
and then the experimental set-ups used for experiments within this thesis are described.
Chapter 6: Results obtained from the experiments are presented, starting with an analysis of
different types of PD and then continuing with a comparison of different patterns and
detection methods.
Chapter 7: A discussion of the work conducted for this thesis is carried out and the results
are summarized and reviewed.
Chapter 8: This chapter contains identified areas of future research which will help
improving the understanding of acoustic emissions from PD.
Chapter 9: Contains all scientifical references used during the duration of the master thesis
work.

-12-

2. ULTRASOUND GENERATED BY PD
2.1 Partial discharges and ultrasound
As stated earlier there is a relationship between ultrasound and PD[7]. Usually the
propagation of these ultrasonic waves away from the PD source is described by combining
three basic equations describing continuity, conservation of momentum and elasticity of the
medium which form the general differential equation of acoustic wave motion:
2 p
= c 2 2 p
2
t

Eq. 1

Where p is the acoustic pressure and c is the speed of sound. However there have been several
simplifications when this equation was derived, and also there has been an assumption made
that there are no non-linear effects occurring and not even that there is an attenuation of the
wave as it travels through the medium. The last statement would for example imply that
someone standing in Kiruna could talk to someone in Malm, which is approximately
thousands of kilometres away. This is obviously not true.
The previously explained simplifications also create problems when certain experimental
results obtained in experiments are being explained. These results mainly considers the
frequencies associated with PD activity, as different papers report completely different
frequency spectrums of PD activity, ranging from 20 kHz up to 1 MHz [4][10][12]. This
cannot be explained by the classical wave equation in which frequency is treated as constant.
There have been certain attempts as to explain this phenomenon by assuming that the
insulating materials features a low-pass character for ultrasonic waves [13], but up until now
no satisfactory explanation has been given.
However an interesting discovery has been made by acoustic researchers dealing with wave
propagation in oil, which is an identical situation to a PD in a power transformer. These
discoveries might account for the large spreading in frequency spectra which cannot be
explained by classical phenomena like scattering, diffraction, etc.
2.2 The nature of ultrasound / nonlinear acoustics
First of all a brief intuitive explanation of what is going on is given, then the formal
expressions and equations are presented that better model the wave propagation of
ultrasounds emitted by PD.
It should be noted that the phenomena that is called sound is actually a pressure wave
propagating through a material as a localized pressure change. Next, from basic physics, it
should be realised that increasing pressure of a gas increases its temperature and that the
speed of sound in a compressible material increases with temperature.

-13-

A result of the previous statements is that waves travel faster during the high pressure phase
of the oscillation than during the low pressure phase. This affects the wave's frequency
structure; for example, in a plane sinusoidal wave of a single frequency, the peaks of the wave
travel faster than the valleys, and the signal becomes more like a sawtooth wave. In doing so,
other frequency components are introduced, which can be expanded into Fourier Series. This
phenomenon is characteristic of a non-linear system, in comparison to a linear acoustic
system which only responds to the driving frequency.[14]
Additionally, waves of different amplitudes will generate different pressure gradients,
contributing to the non-linear effect.[14]
The pressure changes within a medium cause the wave energy to transfer to higher harmonics.
Since attenuation generally increases with frequency, a counter effect exists that changes the
nature of the nonlinear effect over distance. To describe their level of nonlinearity, materials
can be given a nonlinearity parameter, B A . The values of A and B are the coefficients of the
first and second order terms of the Taylor series expansion of the equation relating the
material's pressure to its density. [14]
In order to account for these type of phenomena a lot of research has been carried out in the
area of non-linear acoustics and here three fundamental equations from this research field are
presented: The Fubinis solution, the Burgers equation and the KZK equation.
2.2.1 Fubinis solution
The Fubini solution solves the problem of nonlinear propagation of monofrequency sources
only before shock formation, basically meaning that it is valid to a distance where the wave
develops a vertical tangent, i.e. the beginning of becoming a shock wave. The pressure
distribution is expressed by an infinite sum of weighted Bessel functions:

2
J n (n ) sin( nt ) Eq. 2
n =1 n
Where t = t x / c0 is the retarded time, p 0 is the initial amplitude of the sinusoidal wave,
J n is the Bessel function of order n and = x/x where x is the chock formation distance,
meaning that the equation is valid as long as < 1 . Also it might be worth noticing that
t represents a reference frame moving at speed c0 i.e. at the speed of sound.
p ( , t ) = p0

The Fubinis solution is easily calculated but has its limitations. Nevertheless it gives a fair
understanding of the demodulation phenomena previously described. As an example a
simulation has been performed and shown in fig. 5 of a decaying sine wave calculated with
Fubinis solution.

-14-

Fig 5: Decay of a sine-wave calculated with Fubinis solution

2.2.2. The Burgers equation

The Burgers equation is an approximation to a more exact second-order equation which can
be solved exactly. It properly describes the combined effects of nonlinearity and losses for
plane progressive waves. The equation has the following appearance:
p 2 p
p 2
3
=
x 2c0 t 2 2 0 c03 t

Eq. 3

Where is the sound diffusivity, t = t x / c0 is the retarded time, is a nonlinearity

coefficient of the medium, c0 is the speed of sound and 0 is the ambient density of fluid. It is
also worth noting that the above equation can be extended the generalized Burgers equation
which can handle convergence (or divergence) of both progressive spherical and cylindrical
waves.
2.2.3. The KZK (Khokhlov-Zabolotskaya-Kuznetsov) equation
The large spectrum in which PD occur might have an explanation in the above mentioned
mechanisms. As stated earlier the classical wave propagation model cannot account for
different frequencies obtained with basically the same type of experimental set-up, but in the
late 1990s a new model which describes the propagation of ultrasonic waves was introduced
which is an augmentation compared to the Burgers equation. This model is the KZK
(Khokhlov-Zabolotskaya-Kuznetsov) nonlinear parabolic wave equation with the following
appearance:
2 p c0 2
3 p
2 p2
Eq. 4
= r p + 3
+
zt 2
2c0 t 3 2 0 c03 t 2
2
1
where 2r =
Eq. 5
+
r 2 r r

( )

In the equation p denotes the sound pressure, z is the coordinate along the axis of the beam, r
is the transverse radial coordinate (the soundbeam is assumed to be axisymmetric),

-15-

t ' = t z / c0 is the retarded time and c0 is the sound speed. The strength of this equation is that
it accounts for diffraction (first term of right hand side), thermoviscous attenuation (second
term on right-hand side, is the diffusivity of sound) and the third term accounts for
quadratic nonlinearity of the fluid ( is the coefficient of nonlinearity and p0 is the ambient
density of fluid). [15]
The most interesting part of this equation is that not only does it account for diffraction and
attentuation, but the third term also accounts for quadratic non-linearity of the material in
which the wave travels.
The interesting influence of this third term is that it demodulates waves travelling in a nonlinear medium. This means that a wave travelling in the medium will shift its frequency
spectrum to lower frequencies as it propagates away from the source. This is illustrated in fig.
6, where experimental results are compared to theoretical models of such a demodulation
process for pressure waves travelling in oil. [15]

Fig 6:Experimental and theoretical results of the demodulation processes occurring for a pulse travelling in a
thermoviscous fluid, in this case oil [15]

Since an in-depth analysis of the KZK-equation is far beyond the scope of this thesis only a
simple experiment will be carried out to evaluate if there is some evidence as to claim that
ultrasonic signals emitted from a PD source behaves as suggested by nonlinear acoustics.
2.3 Detection of ultrasound
The problem with ultrasonic detection of PD is the very small sound pressures generated. In
some reports the measured rms sound pressure is reported to be 0.2 Pa at 100 millimeter from
a 1-pC discharge and a meter away it will be 0.02 Pa in oil[16].
These very small pressures are usually measured by piezo-electric transducers (illustrated in
fig. 7). Their principle of operation is that they convert mechanical energy (movement,
pressure) into electrical energy, using piezo-electric crystals, which is then measured.

-16-

Fig 7: Inside of a transducer[17]

Fig 8: Inside of a transducer[17]

What is needed to know about the construction of this type of transducer is that the
piezoelectric element is cut to 1/2 the desired wavelength. To get as much energy out of the
transducer as possible, an impedance matching is placed between the active element and the
face of the transducer. Optimal impedance matching is achieved by sizing the matching layer
so that its thickness is 1/4 of the desired wavelength. This keeps waves that were reflected
within the matching layer in phase when they exit the layer (as illustrated in fig. 8). For
contact transducers, the matching layer is made from a material that has an acoustic
impedance between the active element and steel. Immersion transducers have a matching
layer with an acoustic impedance between the active element and water. Contact transducers
also incorporate a wear plate to protect the matching layer and active element from scratching.
The backing material supporting the crystal has a great influence on the damping
characteristics of a transducer. Using a backing material with an impedance similar to that of
the active element will produce the most effective damping. Such a transducer will have a
wider bandwidth resulting in higher sensitivity. As the mismatch in impedance between the
active element and the backing material increases, material penetration increases but
transducer sensitivity is reduced. [17]

-17-

3. TRANSDUCERS
3.1 Transducers used in thesis
For this thesis four transducers are used, two of them from Physical Acoustics, the first is a
R15 Sensor with an operating frequency range from 50 to 200 kHz, the second, WS
sensor with an operating frequency range of 100 to 1000 kHz.
For this master thesis one other design has to be introduced. Since the thesis include the
construction of a low-cost PD detector, a sensor had to be found since the above mentioned
transducers have a price range of 300-800 Euro. The solution was a sensor which has only
recently been introduced to the market. This is a so-called MEMS (MicroElectroMechanical
System) sensor, which means that it is constructed with methods usually found in
semiconductor production like deposition, pattering and etching. This also means that the
sensor will be much cheaper than a conventional transducer, which makes them interesting for
the design of a low-cost detector. The sensor selected for this thesis is the Knowles Acoustics
SPM0204UD5
Also a fourth kind of transducer was used, namely a Murata MA40MF14-0B ultrasonic
sensor. This is a narrow-band type contact sensor with a center frequency around 40 kHz and
-87 dB sensitivity. The basic construction is very similar to that of the sensors from Physical
Acoustics, but with a narrow working range. These sensors will be used as contact sensors
for the cost-efficient detector.
SENSOR
MODEL

MANUFACTURER

FREQUENCY
RANGE

COST

3 610 SEK

R15 alpha

Physical acoustics

50-200 kHz

WS alpha

Physical acoustics

100-1000 kHz

4 300 SEK

SPM0204UD5

Knowles Acoustics

10-65 kHz

95,60 SEK

MA40MF14-0B

Murata

40 kHz

153 SEK

Table 1: Features of all sensors used for the experiments

-18-

APPEARANCE

3.2 Expected ultrasonic and electric parameters

First of all the sensitivity of the transducers is recalculated so that it is given in mV/Pa. As the
sensitivity for the R15 - sensor is -63 dB (in V / bar ) and -62 dB (in V / bar ) for the
WS - sensor, and 1Pa = 1 10 5 bar , gives that the sensitivity for the R15 is
10

63

20

7.08 10 4 V / bar = 7.08mV / Pa , and 10

62

20

7.94 10 4 V / bar = 7.94mV / Pa for

the WS .
For the Knowles Acoustics SPM0204UD5 the sensitivity will be approx. 10
87
And finally for the Murata MA40MF14-0B it will be 10 20 10 0.446mV / Pa .

47

20

4.47 mV / Pa

As mentioned earlier, the measured rms sound pressure is reported to be 0.2 Pa at 100
millimetres from a 1-pC discharge and a meter away it will be 0.02 Pa in oil. Unfortunately
these reports seem somewhat questionable since the reference article [13] that is supposed to
contain this information doesnt support these claims. However there is an article [18] by the
same author and with a similar name, but there the electrical charge is related to relative
acoustic signal measured in mV, unfortunately denoted as p, which might have lead to a
misinterpretation since p usually denotes pressure.
Without more precise information about the sound pressure generated it is assumed that the
sound generated at a distance of 100 mm from the source generate sound pressures of around
1 Pa. The desired output of the electrical pre-amplifier which is used in between the sensors
and the oscilloscope used to record the readings, is in the range of volts. This means that the
amplifier for the Knowles SPM0204UD5 needs amplification of 10 3 times, and for the Murata
MA40MF14-0B of 10 4 times.
For the Physical Acoustics transducers there is a preamplifier commercially available with a
adjustable gain of 20, 40 and 60 dB. Since these cost several hundred Euros a selectable-gain
amplifier was constructed based on the technical specifications of the commercially available
amplifiers, but at a significantly lower cost.

-19-

4. AMPLIFIERS AND DETECTORS

4.1 Overview of needed amplifiers and detectors
This chapter describes the different types of amplifiers and detectors that were constructed
within the master thesis. First the tuned bandpass detector with a high Q-value for the lowcost sensor system is described. This detector/amplifier was used together with the Murata
and Knowles sensors.
The Second one is the selective-gain broadband amplifier used with the Physical acoustics
transducers which was built since the commercially available amplifiers for these transducers
come at a very high price and it was decided that a cheaper solution had to be found, therefore
one was built based on the technical specifications of the amplifier intended for use together
with these transducers.
4.2 Band-pass detection system
The amplifier used for the low-cost detector design will be designed as a narrow-band
amplifier with a centre frequency of 40 kHz. The choice of 40 kHz as the centre frequency is
made for two reasons. The first is that almost all low-cost transducers have a gain peak at this
frequency and this frequency is also relatively free of ambient noise. The second reason was
that there is strong evidence that acoustic PD signals occur at this frequency [4][10][12].
As stated earlier this circuit should also have the ability to present detected acoustic emissions
in the form of audible sound, therefore there should be some sort of demodulation circuit
which brings down the high frequencies of the ultrasound to audible frequencies. According
to these specifications the layout of fig. 9 was devised and then implemented.

Fig 9: Principle of detector operation

4.2.1 Design aspects

Ultrasonic transducer
For this detector the transducers Knowles Acoustics SPM0204UD5 and Murata MA40MF140B will be used. The first is constructed to detect a wide range of acoustic signals propagating
through air. The second, designed as a surface mounted sensor will be used for detecting
signals propagating in solids or liquids.
Preamplifier
The preamplifier must be able to amplify a very small signal, in the order of a few microvolts,
a couple of hundred times without adding interference. In this step no filtering can be applied
-20-

since the signal is so weak that there is a risk that any filtering will diminish the desired signal
beyond measurability.
Another aspect of this part that needs to be considered is the capacitive nature of the sensors
used. A low input impedance is therefore desired on the preamplifier so that ringing in the
capacitive circuit can be eliminated faster.
In order to fulfil these specifications a CE-step was chosen as preamplifier since it possesses
all the features required: adjustable input impedance, high amplification and low distortions.
After performing the design process outlined in appendix A, the circuit shown in fig. 10 was
obtained.

Fig. 10 The preamplifier, a CE-step with a low input impedance

and an amplification of about 100 times

For a detailed circuit analysis and design method outline the reader is referred to the
appendix, but the most important aspects of this design is that the input impedance is
approximately 300 , which is rather low for an amplifier, and the amplification is around
100 times.
Bandpass amplifier
The amplifier consists of four coupled operational amplifier steps, each based on a modified
Deliyannis filter [18]. The filters are based on the TL074 operational amplifier, which
together with a couple of passive components form a second order narrow bandpass filter. In
this application the Q factor and gain are set to 10. The reason for this is that a small
amplification is desired in every step and the operational amplifiers only have a certain
margin of amplification that cannot be passed (in this case approximately 75 times
amplification), there are also certain stability issues involved in higher gains and Q-factors.
After that the design process has been performed as outlined in appendix B the circuit shown
in fig. 11 is obtained.

-21-

Fig. 11 One part of the narrowband amplifier based on a Deliyannis second

order bandpass filter

Each step is also separated from the next one in terms of DC-component by adding a blocking
68 nF capacitor at the input of each amplifier step.
It should also be mentioned that there was a deliberate choice of not including any
inductances in the circuit for two reasons, first to avoid any coupling to the high electrical and
magnetic fields present in the vicinity of the amplifier, and also because inductors are hard to
produce with a given inductance, and therefore are expensive.
Multiplier
The multiplier is a circuit that multiplies two signals with each other. The concept which is
utilized in this case is called heterodyning. The basic idea is to demodulate the high frequency
ultrasonic sounds picked up by the transducer to the audible range. This is done by
multiplying the amplifier output signal with a reference signal. In the case of two sinusoidal
input signals to the multiplier the following trigonometric identity can be used:
1
1
cos( ) cos( + ) Eq. 6
2
2
or
1
1
sin (2f1t ) sin (2f 2t ) = cos[2 ( f1 f 2 )t ] cos[2 ( f1 + f 2 )t ] Eq. 7
2
2
sin sin =

This states that two sine wave signals multiplied with each other produce two signals, one
with the frequency f1 f 2 , and the second with the frequency f1 + f 2 . In this case we know
that the amplified transducer signal will be at f1 = 40 kHz, and then we choose the external
clock signal to f 2 = 38 kHz, we will get two signals at the output. The first at 2 kHz, which is
in the audible range, and the second at 78 kHz, which is unwanted and can be filtered away by
a simple RC low-pass filter at the multiplier output. As a note it might be mentioned that the
external clock signal in this case was provided by a function generator outside the circuit for
easy tuning purposes, but it can easily be implemented on the circuit itself with a Wien-bridge
oscillator for a sine wave signal or a NE555-timer for a square-wave signal.
For use as a multiplier in this application the AD633JNZ produced by Analog devices was
chosen mainly for its comparatively low cost and large availability on the market.

-22-

Audio amplifier
The audio amplifier selected in this design is the LM386, a low-cost, low-voltage audio power
amplifier which is sufficient to power a pair of headphones. It also requires a very small
number of components in order to operate.
Power supply
The entire circuit was fed from a split power supply consisting of two standard 9V batteries,
except for the Knowles SPM0204UD5, which requires a voltage source of 1,5-3 volts to
operate. This was first realized by a voltage regulator, but since this introduced a lot of ripple
in the supply voltages thus affecting the amplifier a separate power source consisting of two
1,5 V batteries was used.
4.2.2 Modelling
To simulate the circuit behaviour the preamplifier and narrowband amplifier were simulated
in the Orcad Pspice simulator. The simulation of the circuits yielded the following gain and
phase margin curves for the CE-step used as preamplifier:

Fig 12: Gain curve of preamplifier

Fig 13: Phase curve of the preamplifier

The gain curve in fig. 12 show that the amplifier gain starts at around -60dB, crosses unity
gain at 200 Hz and stabilizes at 40 dB around 10 kHz. The phase curve in fig. 13 indicates
that the phase margin is acceptable, and that the circuit is closed-loop stable.

-23-

As to the narrowband amplifier, the simulation results were as follows:

Fig 14: Gain curve for each amplification step of the narrowband amplifier

Fig 15: Zoomed in gain curve for each amplification step of narrowband amplifier

Fig 16: Phase curve of one amplifier step

From the gain curve it is easy to determine that the amplifier indeed will feature a band-pass
character with a relatively high Q-factor. Due to the narrow band-pass character the amplifier
is expected to have a small phase margin, which is also verified in the simulation. Due to this,
the amplifier will be prone to oscillations when operating. This also gives an indication of that
the amplifier is operating on its stability margin which was mentioned earlier.

4.2.3 Implementation
-24-

The biggest issue discovered during implementation which didnt occur during simulation
was ripple found in the supply voltages. This ripple in turn caused stability issues in the
narrowband amplifier, therefore the transducer received a separate power supply and coupling
capacitors were introduced to the power rails of the operational amplifiers in order to reduce
the ripple in supply voltage, which eventually solved the problem.
Another issue was the audio output of the detector. The problem was to find suitable
parameters for the reference clock, which was done by tuning the signal generator to different
frequencies and simultaneously adjusting the voltage. Finally it was decided that the best
audibility was achieved at 37 kHz with a V pp = 200 mV on the reference clock.
There also were some problems with the narrow-band amplifier, since the components used
(i.e. resistors and capacitors) have a certain tolerance. This caused the peak frequencies of
each step to be shifted from the ideal of 40 kHz.
4.2.4 Verification
The verification was made in such way that the part of the detector circuit tested was fed by a
well-defined input signal delivered by a function generator, and the output(s) were measured
with an oscilloscope. This procedure yielded the following results shown in fig. 17-18.

Fig. 17: Gain curve of preamplifier

-25-

Fig. 18: Gain curve of narrowband amplifier

From the above curves it is noted that both the preamplifier and the narrowband amplifier are
in good agreement with the simulated results. Also as expected the narrowband amplifier has
the peaks of each amplifier step at different frequencies which deviate from the simulated
results. From the measurements it was assessed that the narrowband amplifier had an overall
peak response for 37 kHz, and not 40 kHz as designed.
4.3 Broadband amplifier
The broadband amplifier, intended for use together with the transducers from Physical
Acoustics, is a fairly straightforward construction with a flat gain in the frequency range from
10 kHz to 1 MHz, a high input impedance, low noise and a Common-mode rejection ratio
(CMRR) of 42 dB. The gain also has to be adjustable in the range 20, 40 and 60 dB. The
overall design concept is presented in fig. 24.
The Common-mode rejection ratio (CMRR) of the amplifier is a very useful feature which
usually isnt considered when operational amplifiers are being used, but in certain
applications it needs to be considered. In this case it is of interest since there will be large
voltage swings in the last amplifier and therefore it might be easier for that amplifier to
saturate, which is an undesired situation. First of all the equation which considers the
common-mode has the following appearance:
A
Vout = Ad (V+ V ) + cm (V+ + V ) Eq. 8
2

-26-

In this equation the desired differential gain is denoted Ad and the undesired common-mode
gain is denoted Acm . Intuitively this equation can be understood as that the output of an
amplifier is the difference on the input multiplied with differential gain plus the sum of the
inputs times the common-mode gain. Usually the common-mode gain is much lower than the
differential gain, and from this relation the common-mode rejection ratio is defined as:
A
CMRR = 20 log10 d
Acm

Eq. 9

4.3.1 Design aspects

The implementation of this amplifier requires a better operational amplifier than the
previously used TL074, since the amplifier needs to produce a gain in a higher frequency
range. Therefore the AD817 operational amplifier is chosen as the main component since it
features all the necessary specifications. It has a CMRR of 100 dB, the unity gain bandwidth
is 50 MHz, and the input impedance is 300 k .

Fig 19: Non-inverting amplifier

The amplifiers will be connected in a non-inverting configuration, presented in fig. 19, and
then series connected, as shown in fig. 20, with tapping of the signal at each of the amplifier
output terminals. The gain is set by

R
Vout = 1 + 2 Vin Eq. 10
R1

Since the gain is supposed to be 20dB (10 times) per step, the resistors are chosen as
R1 = 18 k and R2 = 180 k . This gives a gain of 11. This little higher gain is chosen so that
the effect of the DC blocking capacitors in between every step is compensated.
A low-pass network is added on the output terminal in order to limit the bandwidth of the
amplifier to approximately 1 MHz. This network consists of an RC-link in series between the
output terminal of the operational amplifier and ground, as seen in fig. 21. This network will
have a cut-off frequency equal to f c = 1 2RC = 1 /( 2 10 22 10 9 ) = 723.4 kHz .
Another very important aspect of this amplifier is that the positive input of the operational
amplifier has a resistance of 18 k to ground, which is equal to the ground resistance of the
negative input. The reason for this seemingly unnecessary resistance is that it ensures that the
bias currents of the inputs are equal, thus ensuring that the offset voltage on the output is kept
minimal.

-27-

Fig 20: Principle of amplifier operation

4.3.2 Modelling
The amplifier is modelled using the Orcad Pspice simulator which gives the results presented
in fig. 21-23.

Fig 21: schematic view of the cascaded broadband amplifier

Fig 22: Gain curve of broadband amplifier

-28-

Fig 23: Phase curve of broadband amplifier

The simulation shows that the gain is as expected in the desired interval. From the phase
curve the amplifiers are judged to be very stable, and therefore with little or no oscillations
due to instability.
4.3.3 Implementation
The main difference between the simulation and implementation was that decoupling
capacitors were added on the power rails in order to eliminate ripple in the supply voltages.
One problem which was discovered in the implementation was that whenever one of the
operational amplifiers saturated there was a significant amount of ripple in the supply
voltages. The solution to this problem was to add another switch which allowed the last
amplifier to be disconnected, which solved the problem with saturation and power rail ripple.
Another issue was that the low-pass links were removed since it was realised that the cut-off
at higher frequencies was unnecessary to implement, since it had no impact on the desired
measurements.
Another major issue which was discovered during the implementation was that there was
large capacitive coupling present in the circuit which demonstrated itself as lower gains at
increasing frequencies, unfortunately this problem was never resolved mainly because it was
judged that the impact on the overall performance was negligible.
4.3.4 Verification

Fig. 24 Gain curve of broadband amplifier

-29-

The verification shows that the amplifier is working almost as expected. However compared
to the simulations the gain in lower frequencies is approximately 57 dB, which is slightly
lower than the expected value, as shown in fig. 24.
Another interesting aspect is the decrease in gain at frequencies above 500 kHz for the last
amplifier. This is caused by stray capacitances within the circuit, which were introduced when
the last amplifier was connected to a switch and long connection leads were required in order
to achieve that. This is a perfect example of a type of phenomenon that cannot be simulated
with Spice.
Despite of the shortcomings in the last amplifier step the design is still useful. It provides
sufficient amplification up to a frequency where most of the signal is expected to be. In order
to cure this problem, methods that eliminate the stray capacitances should be applied. These
methods include using as short interconnecting leads as possible together with a ground plane.
However since time was limited, the decision was made that the amplifier is good enough to
use.

-30-

5. LABORATORY SETUP FOR PD DETECTION USING ULTRASOUND

5.1 Experimental setup
For the practical experiments two separate voltage sources were used to generate the high
voltages necessary, the first source is a TREK 30/30 High Voltage amplifier, controllable
both in voltage and frequency, which can deliver up to 30 kV. This source was used for
corona in air experiments, since this type of discharge doesnt require voltages above 30 kV
due to the lower PD inception field strength of air compared to oil. The second source is a
100 kV transformer which is controllable in voltage. The main detection system used in the
experimental setup is an ICM system provided by Power Diagnostix, seen in fig. 31. The
ICM system is decoupled with a H.V. capacitor in order to protect the system in case of a full
breakdown in the test specimen. This capacitor can bee seen as the red column standing on
the right in fig. 27, also the H.V. transformer can be seen in the picture appearing as a blue
barrel standing on the left.
The test chamber where the experimental setups were exposed to high voltages consisted of a
transformer oil-filled 33x33x33 cm grounded steel tank with circular observation windows
mounted on two sides. The contact transducers used for the experiment were attached to
these observation windows with duct tape (fig. 27) and a viscous gel was applied to the
sensors to ensure proper acoustical coupling.
The Knowles sensor used to detect PD through air was placed at a distance of approx. 1
meter away from the test chamber (fig. 28), and a funnel belonging to another ultrasonic
measurement instrument was placed as a waveguide in front of the sensor.
For the experiment where the Knowles sensor was used to detect corona in air (fig. 29) there
were two reference instruments used to detect PD, first as in the previously mentioned
experiment the ICM system was connected but also a second instrument which detects
ultrasonic waves was used as a reference. This instrument, provided by Tettex (fig. 30), gave
some reference as to how the audible detection was working. Also waveguides belonging to
this instrument was used to increase the gain of the ultrasounds emitted for both instruments.

Fig 25: Experimental setup in laboratory [19]

Fig 26: Experimental setup used for thesis

-31-

Fig 27: Sensor attached to steel tank

Fig 28: Air sensor used together with attached sensor

Fig 29: Test setup for testing corona in air. ICM system and voltage source also visible

Fig 30: Tettex ultrasonic detector

Fig 31: ICM system

5.2 Phase resolved PD measurement equipment

The ICM system from Power Diagnostix as seen in fig. 31 is used as a reference in the
laboratory experiments. It is a digital partial discharge detector which detects the current
pulses that are generated by a PD. This instrument is widely used in both academical and
commercial applications.

-32-

This instrument is designed in such a way that PDs are logged and categorized accordingly
to its apparent charge and the occurrence relative the phase of the supply voltage. This is then
displayed as a phase-resolved PD pattern as can be seen in fig. 4.
5.3 Test objects
For the experiments a couple of different test setups were used in order to simulate different
situations in which PD occur and these are listed below.
5.3.1. Test object for Corona in air
The test object which was used to simulate corona discharges in air is a modified point-plane
setup with a steel hemisphere used as ground plane and a needle connected to high voltage
placed in the center of the hemisphere which can bee seen in fig. 32. This set-up has an
apparent charge for negative PD of 200 pC.

Fig. 32: Air-corona source

5.3.2. Canonical Test objects for PD in transformers

The other test objects used were immersed in the steel tank in order to investigate void,
surface and corona discharges in oil. For simulation of void discharges a stack of 15
transformer insulation papers with air bubbles trapped between the papers was mounted
between two brass electrodes and then immersed in oil. For surface discharges two brass
electrodes were used where a pressboard placed in between the electrodes guaranteed that no
full breakdown of the gap could occur. The last setup used consisted of a needle-plane setup
for simulating corona discharges. Once again a pressboard was used to prevent a full
breakdown of the gap.

Fig 33: Test setups for Surface, Void and Corona discharges, with expected PD activity marked in red

5.3.3. Test object for investigation of distance dependence of frequency spectra

For the experiment where the dependence between frequency and distance was tested, the
experimental setup, schematically drawn in fig. 34, consisted of a movable PD source
immersed in a transformer-oil filled Plexiglas pipe with a diameter of 10 cm and energized by
a 100 kV transformer. At the bottom a WS and a R15 transducer from Physical Acoustics

-33-

connected in parallel to the amplifier was mounted on the surface of the lid. The reason for
using multiple sensors is their respective bandwidth, since if the two sensors are combined the
detection system will have a bandwidth from 50 kHz to 1 MHz, which was done in order to
investigate as broad spectrum as possible. A proper acoustical coupling was guaranteed by
adding a viscous gel between the lid and the sensors.
The PD source itself consists of a PVC pipe with a High-voltage centred electrode surrounded
by a grounded copper shield, which is visible in fig. 35. This PD source will generate corona
discharges when exposed to high voltages.
The frequency spectrum emitted by the PD source will then be investigated, and if the
frequency spectrum tends to move to lower frequencies as the PD source is moved away from
the transducer this is a strong indication of non-linear demodulation processes. On the other
hand, if the PD follows classical wave propagation models, the received spectrum will be
similar independently of the distance from the source.

Fig 34: Schematic figure of experimental setup for testing

frequency dependence

Fig. 35 Cross-sectional view of the PD source itself.

Fig 36: Experimental setup for testing frequency dependence

-34-

6. Results
6.1 PD patterns for different situations
Here follows a presentation of the waveforms acquired during the experiments and their
comparison to the reference measurement instrumentation already in use where applicable.
6.1.1 Corona in air

Fig 37: background noise (electrical and acoustical)

1 meter from PD source

Fig 38: Detected PD compared to the supply line voltage

1,5 meters away from PD source

Fig 39: Detected PD under AC supply voltage

1 meter from PD source

Fig 40: Detected PD under DC supply voltage

1 meter from PD source

The experiments proved that a 200 pC PD could be clearly detected up to 1,5 meters away
from the actual discharge site (fig. 38). This was a better result than could be obtained with
the Tettex ultrasonic sensor which only detected the discharge at a range of 75 centimetres.
This also suggests that low-cost MEMS-based sensors are a viable alternative for detecting
PD in air. Unfortunately no waveforms from the Tettex instrument could be acquired since it
is only equipped with a pair of earphones through which the operator can hear the PD.
The assumption that an acoustic PD contains a 40 kHz component also was confirmed. Also
as it can be seen in the oscilloscope figures there was a sufficiently large signal to noise ratio
when discussing in terms of ambient noise, which can bee seen in fig. 37 compared to fig. 38.
It should be noted that no special precautions were made to eliminate background noise since
the conditions for testing should be as similar as possible to a field measurement situation.
One feature that should be noticed is that in fig. 38 there is almost a phase-shift of 90 degrees
between where the PD actually occurred (in the peak of the negative voltage) and the
measured acoustic PD signal. This difference is mainly caused by finite sound propagation
speed and a delay in the amplifier circuit. Unless care is taken when analyzing such signals

-35-

one can be mislead to think that the PD displayed in fig 38 is caused by a void discharge
which in this case is incorrect.
The experiment also shows that the Knowles sensor is useful both for AC and DC corona, the
latter demonstrated in fig 40, but if the noise level from fig. 37 is unknown, DC corona may
be dismissed as background noise, which would be incorrect.
6.1.2 Corona in oil

Fig 41: Negative corona, amplifier gain 60 dB, leading edge

Fig 42: Negative corona, amplifier gain 60 dB, full signal

Fig 43: Positive corona, amplifier gain 60 dB , full signal

Fig 44: Positive corona, amplifier gain 60 dB, leading edge

The results from the experiments where corona in oil occurred showed that both positive and
negative corona could be detected, the differences noticeable were that positive corona
generated higher signal peaks than negative corona, which was sometimes barely
distinguishable from the background noise. Also the signal from positive corona faded more
slowly than the signal from negative corona, as seen in fig. 43. Both of these results suggest
that positive corona contains more energy than negative, which already is known.
Another interesting result in these experiments is that the bulk of the detected signal is
preceded by a sharp spike occurring for both positive and negative corona, as seen in fig 41
and fig 44. The most probable explanation for this phenomenon is that it has something to do
with how the acoustic wave propagates. The suspected mechanism is that the first highfrequency spike has travelled through a medium which has a small acoustic nonlinearity and
high wave propagation speeds, in this case the setup frame and steel tank, and the second
burst with a considerably lower frequency has travelled through a medium with high acoustic
nonlinearity and lower wave propagation speeds, like the transformer oil.

-36-

The experiment also suggests that there is a difference in the acoustic emissions of positive
and negative corona since the positive corona demonstrates an exponential decay visible in
fig. 43 as the positive corona is more randomly distributed as in fig. 42.
6.1.3 Surface discharge

Fig 45: Suspected corona discharge, amplifier gain 60 dB

Fig 47: Surface discharge, amplifier gain 60 dB

Fig 46: Surface discharge, amplifier gain 60 dB with similar

discharges both for negative and positive half-cycle

Fig 48: Acoustically well shielded surface discharge, amplifier

gain 60 dB

In the experiment involving surface discharges a wide variety of different signals was
obtained. The most probable explanation is that other types of discharges than surface
discharges were present during the experiment, like in fig. 45 which shows a greater
resemblance to corona discharges than to a surface discharge in fig. 47. This is probably due
to the fact that there are corona discharges present at the edges of the brass electrodes due to
carbonization of the pressboard layer or possibly gas bubbles within the setup.
An interesting aspect visible in fig. 46 is that it appears that the acoustic energy contained in
both positive and negative surface discharges is similar, but the experiment also indicates that
there is a difference in discharge strength probably connected to where the discharge takes
place, deep inside the test setup or somewhere near the edge.
The acoustic PD signals from the surface discharges also suggest that there are two separate
bulks of the signal. The first larger in amplitude than the second, both with a somewhat
random waveform.

-37-

As it is shown in fig. 48, some discharges are more or less acoustically shielded, this in turn
affects the properties of the registered waveforms, sometimes occurring like large spikes in
the oscilloscope pictures.
6.1.4 Void discharge
This experimental setup yielded no result for the acoustical detection methods. There were
discharges present which could be recorded through the ICM system, but not on any of the
acoustical sensors. This is probably due to the screening effect of the insulation paper used,
which doesnt allow the pressure waves formed to exit the test setup. Basically this
experiment shows that acoustic detection of void discharges seem to be impractical. There has
been some research into this field [20] which suggests that the minimum value that can be
detected for void discharges at a range of 100 cm is approx. 1000 pC, or even up to 10 000 pC
if the discharge site is well shielded, which is a very powerful and completely unacceptable
discharge within any type of high voltage apparatus.
However the situation in the real world might be different from laboratory conditions, since in
a transformer where void discharges usually occur there might be a better acoustic coupling
due to the core or the windings being closer to the discharge site thus allowing acoustic waves
to pass through them and further on to the sensors by the transformer tank wall.
6.2 Frequency dependence in PD signals as function of the distance
The experiment was conducted so that the waveforms were acquired with two simultaneously
connected sensors from Physical acoustics for negative corona discharges with the Agilent
oscilloscope at a sample rate of 50 Msamples/sec, and then the FFT transform was applied in
a Matlab environment, which gave the following frequency spectras:

Fig 49: Time domain signal and frequency domain distribution for the short distance

-38-

Fig 50: Time domain signal and frequency domain distribution for the long distance

The two previous signals were chosen as representative for the test runs. From their
frequency distribution it is clear that the distance has an impact on the frequency distribution
of the signal. This impact is also as predicted by non-linear acoustics, which suggests that
demodulation processes in a non-linear medium will shift the frequency spectrum towards
lower frequencies with increasing distance. This is a clear indication that there are
demodulation processes present and that these should be considered when discussing
frequency spectras of PD, which up until now seems to have been unknown to the
community studying PD.
These tests should however be concerned as a strong indication of demodulation processes,
but still not treated as conclusive evidence. Further investigations are required in order to
eliminate other possible mechanisms that affect the test results.
It should also be mentioned that the test setup itself is much idealized in comparison to real
world situations where the wave propagation paths are usually very complex and passing
through several different types of materials like paper, copper aluminium and so on. Also the
oil itself was kept at the same temperature, but in real transformers there is always a
temperature distribution with hot oil at the top, and colder oil at the bottom, not to mention
that usually the oil also has a certain flow.
6.3 Comparison of PD Patterns for different situations
When comparing different acoustic PD patterns obtained during the experiments there are
some features that might allow different types of PD to be distinguished from each other.
In the experiments conducted for this thesis the corona discharges were clearly
distinguishable because the main signal was preceded by a sharp high-frequency spike.
However since only one type of source was used for corona there is a lack of evidence that

-39-

this is a feature characterizing all types of corona discharges or only the particular set-up used
in the experiments.
Surface discharges seem to be more oscillative in their nature, generating more random
signals than the exponentially decaying signals produced by corona discharges. The length of
the signal itself may also provide some information as to the type of PD generated, since
corona discharges seem to generate shorter bursts of acoustic energy than surface discharges.
Another source of information might lie in the energy contained within a PD signal. Corona
discharges clearly have a bigger acoustic energy than surface discharges because the PD event
is more acoustically exposed in the case of corona than the surface discharge.
Since no data is available for void discharges there is no basis to conduct an analysis.
However a more fundamental investigation should be conducted using many types of
different voids, both in size and material in order to fully discard the usefulness of acoustic
measurements of void discharges. Also as discussed earlier there might a somewhat different
situation in a real transformer which might facilitate void discharge measurements.
6.4 Comparison of different PD detection methods
A comparison of all available sensors was also made for the cases of corona in oil and surface
discharge in oil.
6.4.1 Corona in oil
For this experimental set-up a comparison was conducted between the ICM system (green
traces in the oscillograms), the Physical Acoustics 1MHz sensors (Red traces in the
oscillograms) and the Murata MA40MF14-0B (Blue traces in the oscillogram). These
experiments yielded the results presented in fig. 51-55:

Fig 51: Negative corona detected by three separate sensor arrays

Fig 52: Negative corona with large spikes on acoustic sensors

-40-

Fig 53: Void discharges detected by ICM system, but no

discharges within the test object

Fig 54: Negative corona detected by the broadband acoustic

sensor

Fig 55: Negative corona which is acoustically detectable, but not electrically

The results obtained showed that the acoustical methods are more sensitive than the electrical
measurements (fig. 51), and that discharges that are barely detectable electrically give rise to
high acoustic peaks (fig. 52 and 55), and sometimes the corona cannot be detected at all
electrically (fig 54-55). This is caused by the fact that the ICM system signal has a too high
noise level to distinguish these PDs and therefore doesnt react since the signal is buried in the
noise.
Another very interesting aspect which shows the strength of acoustic PD measurement are
the electrical spikes detected by the ICM system at zero crossings (fig. 53-54), these are
caused by a void somewhere in the setup, but the acoustic detection allows us to eliminate the
test setup as the source of these void discharges. This is a common problem which is often
encountered during electrical PD measurements on real equipment that PD pulses are
propagated within electrical systems and give false readings as to the tested object. This error
source can be eliminated with acoustic measurement since these measurements are measuring
the actual source and cannot be influenced by other components which might be located as far
as a couple kilometres away from the test object.
Another test that was performed on corona in oil was to investigate if it is possible to detect
corona discharges outside of the test tank at a certain distance with the air coupled sensor.
This was done by placing the Knowles SPM0204UD5 sensor at a distance of approximately
50 cm from the transformer tank and simultaneously read the Physical Acoustics sensor
placed on the tank wall with the Knowles sensor. This yielded the following result with green
traces belonging to the Physical Acoustics sensor and yellow traces belonging to the Knowles
sensor:

-41-

Fig 56: Negative corona, electrically detected

Fig 57: Negative corona, electrically detected

Fig 58: Positive corona, detected both

electrically and acoustically

Fig 59: Positive corona, detected both

electrically and acoustically

The results showed that negative corona could not be detected with the Knowles sensor (fig.
56-57). However, positive corona was detectable (fig. 58-59) which proves the concept that
measurement sensors dont have to be directly coupled to the test object in order to detect PD.
The benefit being that even equipment under high voltage can be monitored for PD activity
without using costly setups to allow the PD sensors to be mounted in areas with strong
electrical fields.
6.4.2 Surface discharge in oil
For this set-up a comparison was conducted for the ICM system (green traces in the
oscillograms), the Physical Acoustics 1MHz sensors (Red traces in the oscillograms) and the
Murata MA40MF14-0B (Blue traces in the oscillogram). These experiments yielded the
results shown in fig. 60-63.

Fig 60: void discharges detected by two arrays

Fig 61: positive and negative surface discharges

-42-

Fig 62: positive and negative surface discharge

Fig 63: Void discharges, propably not present within the test
setup

The results from this experiment indicates that all three sensors were able to detect PD under
both positive and negative voltage. What is even more interesting are the void discharges that
were present during this test run. These were formed after that the test object had been
stressed with high voltage for some time, and at first only the ICM system detected these
voids (fig. 63) but after a while there was acoustical signals present as well (fig. 60). Another
interesting aspect is the difference between fig. 61 and fig. 62, where the surface discharges
are at first detectable with both of the acoustical sensors and only 6 seconds later they are only
detectable on the tuned 40 kHz-sensor. If the assumption is made that the signal follows
nonlinear acoustic principles this suggests that the PD signal received in fig. 62 has travelled a
longer distance in the oil or through an acoustically more non-linear media. This insight
seems to be a completely new approach on how PD sources can be localized within a test
sample based on the frequency content of the received signal.
Up until now the most common method to position PD with the use of acoustic sensors is
based on triangulation schemes which requires three or more transducers and some signal
processing, but if the frequency content of the signal could be used for this purpose one sensor
and one signal could be sufficient to give a good estimation as to where in the test specimen
PD occurs.

-43-

7. CONCLUSIONS
The work conducted during this thesis fulfils the criterias that were set as goals. The
difference in the acoustic signature of different types of discharges has been investigated and
with a good certainty these results hold. Also a proof of concept has been mad as to the lowcost detector solutions and their usefulness has also been demonstrated within the work.
Two amplifiers have been constructed which meet the desired specifications and it has also
been verified accordingly. However there is still much work that can be done to the
narrowband amplifier in order to get a better signal-to-noise ratio and an overall better and
maybe cheaper design. Also field testing is required in order to verify the usefulness of these
amplifiers, since there still might be aspects which cannot be simulated in the laboratory but
rather needs to be field tested. This might include EMI and other disturbance sources, both
electrical and acoustical.
There is still a lot of work to be done in order for the amplifiers to be mass-produced, since
right now they are mounted on a lab experimental board with a lot of wiring which is
unacceptable in a finished product. Some consideration as to the choice of components also
has to be done since it might be easier and cheaper to produce amplifier boards with surface
mounted components which is a path that has not been pursued in this thesis.
Another path which couldnt be fully pursued is the issue of wave propagation within oil, but
the experiment conducted within the frame of this thesis showed that there is a correlation
between the distance from a PD source and the frequency spectrum from its acoustical
emissions, which suggest a nonlinear acoustic behaviour of acoustic PD signals
The conclusions that can be drawn from the work conducted is that acoustic emission
detection has been proved to be a powerful method for detection and classification of PD, but
from the work done the conclusion is also drawn that the more information from many
different sources that is available the easier it is to characterize the PD. Therefore acoustic
emission detection from PD sources can and should be used as one method of detecting PD,
but an effort to increase the knowledge of the underlying mechanisms should be performed.

-44-

8. FUTURE WORK
There still remains a lot of work to be done to fully understand the acoustical emissions of
partial discharges. A more thorough investigation of the propagation paths of these emissions
should also be carried out in order to develop the understanding of how different materials
affect the frequency content of signals.
Another interesting aspect is that most of the fundamental work on PD has been conducted
almost 40 years ago. Therefore there is an interest in repeating these old experiments but with
equipment which is available today, i.e. fast oscilloscopes, computers, better sensors, etc.

-45-

9. REFERENCES
[1]D. Knig, Y.N. Rao, Partial Discharges in Electrical Power Apparatus. Berlin;
Offenbach: VDE-Verlag, 1993, pp. 15-36
[2] Stone, G.C.; , "Partial discharge. VII. Practical techniques for measuring PD in operating
equipment," IEEE Electrical Insulation Magazine, vol.7, no.4, pp.9-19, July-Aug. 1991
[3] Boggs, S.A.; , "Partial discharge: overview and signal generation," IEEE Electrical
Insulation Magazine, , vol.6, no.4, pp.33-39, July-Aug. 1990
[4] Howells, E.; Norton, E. T.; , "Detection of Partial Discharges in Transformers Using
Acoustic Emission Techniques," IEEE Transactions on Power Apparatus and Systems,
vol. PAS-97, no.5, pp.1538-1549, Sept. 1978
[5] Kemp, I.J.; , "Partial discharge plant-monitoring technology: present and future
developments," IEEE Proceedings - Science, Measurement and Technology, vol.142,
no.1, pp.4-10, Jan 1995
[6] Duval, M.; Dukarm, J.; , "Improving the reliability of transformer gas-in-oil diagnosis,"
IEEE Electrical Insulation Magazine, vol.21, no.4, pp.21-27, July-Aug. 2005
[7] Lundgaard, L.E.; , "Partial discharge. XIII. Acoustic partial discharge detectionFundamental considerations," IEEE Electrical Insulation Magazine, vol.8, pp.25-31,
Sept.-Oct. 1992
[8] Anderson, J. G.; , "Ultrasonic Detection and Location of Electric Discharges in Insulating
Structures," Power Apparatus and Systems, Part III. Transactions of the American
Institute of Electrical Engineers , vol.75, no.3, pp.1193-1198, Jan. 1956
[9] Lundgaard, L.E.; , "Partial discharge. XIV. Acoustic partial discharge detection-practical
application," IEEE Electrical Insulation Magazine, vol.8, no.5, pp.34-43, Sept.-Oct.
1992
[10] Markalous, S.; Tenbohlen, S.; Feser, K.; , "Detection and location of partial discharges in
power transformers using acoustic and electromagnetic signals," IEEE Transactions on
Dielectrics and Electrical Insulation, , vol.15, no.6, pp.1576-1583, December 2008
[11]Sarathi, R.; Singh, P.D.; Danikas, M.G.;, Characterization of partial discharges in
transformer oil insulation under AC and DC voltage using acoustic emission technique
Journal of electrical engineering, vol.58, no.2, pp.91-97, 2007
[12] Harrold, R.T.; , "Ultrasonic Spectrum Signatures of Under-Oil Corona Sources,"
IEEE Transactions on Electrical Insulation, vol.EI-10, no.4, pp.109-112, Dec. 1975
[13]Beyer, M.; Borsi, H.; Hartje, M.;, Some aspects about possibilities and limitations of
acoustical partial discharge (PD) measurements in Insulation fluids, 5th international
symposium on High Voltage Engineering, Braunschweig, Germany, 24-28. August 1987
[14] Internet: http://en.wikipedia.org/wiki/Nonlinear_acoustics, march 9, 2012 [Apr. 27,
2012]
[15] Averkiou, M.A.; Lee, Y-S.; Hamilton, M.F.;, Self-demodulation of amplitude- and
frequency-modulated pulses in a thermoviscous fluid Journal of Acoustical society of
America, pp. 2876-2883, 1993
[16] Mark MacAlpine, Zhao Zhiqiang, M. Suleyman Demokan, Development of a
fiber-optic sensor for partial discharges in oil-filled power transformers, Electric
Power System Research, vol. 63, pp. 27-36, 2002
[17] S. Joy. about instrumentation; A blog for Instrumentation engineering Students
Internet: http://aboutinstrumentation.blogspot.se/2012/01/piezoelectric-transducers.html,
Apr. 14, 2012 [Apr. 14, 2012]
[18] Sedra, A.; Ghorab, M.; Martin, K.; , "Optimum configurations for single-amplifier
biquadratic filters," IEEE Transactions on Circuits and Systems, vol.27, no.12, pp. 11551163, Dec 1980

-46-

[19] Niasar, M.G.; Edin, H;, Corona in Oil as a Function of Geometry, Temperature and
Humidity, KTH Electrical engineering,
[20] Harrold, R.T.; , "The Relationship Between Ultrasonic and Electrical Measurements of
Under-Oil Corona Sources," IEEE Transactions on Electrical Insulation, vol.EI-11,
no.1, pp.8-11, March 1976

-47-

APPENDIX A: DESIGN PROCESS OF A TRANSISTOR AMPLIFIER

In this appendix the outline for designing a Common Emitter amplifier is given. The design
process of the amplifier used in the thesis is presented. The descriptions in this appendix are
kept very short and in order to fully understand the concepts the reader is referred to a
textbook on the subject [1].
The Common-Emitter amplifier basic outline is that of fig. 1 when using a NPN transistor.
What has to be done is to choose suitable values of R1 , R2 , RL , RE , C1 and C2 . In the amplifier
used for the PD detector the following design criteria are given:
-

Input impedance of 300

A signal gain of 100

Fig. A1: A CE-amplifier

Fig. A2: A CE-amplifier with the nomenclature used

below

First of all the AC properties of the amplifier are investigated. For that purpose the small
signal model of the transistor is applied

Fig A3: Small signal model of Amplifier

After applying Kirchhoffs Voltage and Current laws the following result is obtained:
Avs =

Rc h fe
Vout
Eq. A1
=
Vin (1 + Rc hoe )hie

Then the values for the transistor used, 2N2222A, are obtained from the datasheet assuming a
collector current of 10 mA:
hie = 750 h fe = 225 hoe = 113S

-48-

This in turn gives that

( Avs hie )
Rc =

h fe

( Avs hie hoe )

1
h fe

= 346.3803 360 Eq. A2

Now in order to meet the desired AC input resistance at Rin = 300 we conclude that
Rin = R1 // R2 // hie , let Req = R1 // R2 then:
Rin =

hie Req

hie + Req

Req =

hie Rin
= 500 Eq. A3
hie Rin

The desired AC properties are now met, and therefore it is time to turn the attention to finding
a suitable DC operating point for the amplifier. In order to do this the Thevenin-equivalent to
the left of the x:es in fig 4 is applied to the circuit. This gives the resulting schematic as in fig.
5.

Fig A4: DC-biasing of the

CE-amplifier

Fig 5: Thevenin-equivalent of the amplifier input

The values of Veq and Req are given by

hie Rin
= 500 Eq. A4
hie Rin
R2
Eq. A5
Veq = Vcc
( R1 + R2 )
Applying Kirchoffs Voltage law around the input and output loops with the approximation
that I c I e and Vbe = 0.65V gives the following equations:
Req =

Vcc = Rc I c + Vce + I c Re (1)

Eq. A6-A7
Veq = I b Req + 0.65 + I c Re (2)
Since I c = 10mA and I c = h fe I b I b =

Ic
44 A Eq. A8
h fe
-49-

Vcc 9V
=
= 4.5 V , this allows
2
2
the input signal to operate within the linear region of the amplifier. The assumption made here
is that the amplifier is fed from a 9 Volt DC source, and therefore Vcc = 9 V . If this value of

If Vce now is chosen in the middle of the operating range, Vce =

Vce is then inserted into equation (1) together with all other known values it gives that
Re = 103.6197 100 (Resistor value adapted to E12-series). With this resistor calculated
the value of Veq is calculated, which gives the value Veq = 1.7084 V . This is then used to
calculate R1 and R2 according to:
Veq Req

+ Req
V V
s
eq

R1 =
= 2.6340k 2.7 k Eq. A9
Veq
Vs Veq
R2 = R1

Veq
Vs Veq

= 617.1501 620 Eq. A10

The values that are still left to decide are the ones of C1 and C2 . The first capacitor, C1 , is
used to block DC voltages from the input of the amplifier. It should be chosen in such a way
that it doesnt block signals at the desired frequencies. In this case the desired frequency is
exactly 40 kHz. In this case the capacitor C1 will form a high-pass filter together with the
input resistance Rin = 300 . We wish frequencies below 40 kHz to be blocked, therefore the
1
capacitor value is first calculated as f c =
C1 13 nF . In order to have even
2Rin C1
stronger signals passing through the capacitor, it is given the value of 68 nF in the final
circuit, this is due to the definition of f c , since that is the frequency at which half of the input
power is transmitted to the output of the RC high-pass filter . The second capacitor is used to
give as large amplification as possible in the amplifier. The emitter resistor stabilizes the
transistor's bias point for DC conditions, but it does this by introducing negative feedback at
all frequencies. The capacitor "shorts out" the emitter resistor for AC signals above some
frequency, and restores the lost AC gain above the "shorts out" frequency. In this circuit a
value of 100 F is chosen as a suitable value.
This concludes the calculations necessary to implement the CE-amplifier. The values obtained
are:

R1 = 2.7 k Re = 100
R2 = 620 C1 = 68 nF
Rc = 360 C2 = 100 F
REFERENCES
[1] A.S. Sedra, K.C. Smith, Microelectronic circuits. Oxford; Oxford: Oxford university
press, 2004
-50-

APPENDIX B: DESIGN OF NARROWBAND AMPLIFIER

Introduction
First of all it is recognized that the amplifier needs to have a high Q-value, a bandpass
frequency peak at 49 kHz and be realized with as few and as cheap components as possible.
From the different families of amplifier topologies the Deliyannis bandpass circuit [2] was
chosen since it meets the specifications required. The Deliyannis topology belongs to a family
called Single-amplifier biquadratic active filters[1][2].
These filters of the second order were chosen for their very economic use of operational
amplifiers, since they only require one amplifier per second-order section. This is certainly a
benefit in battery-operated system where the power-supply current needs to be conserved.
The drawback of this second-order realization is that it suffers from a great dependence on the
limited gain and bandwidth of the op-amp and can also be more sensitive to the unavoidable
tolerances in the values of resistors and capacitors.
Single-amplifier biquadratic active filter
The first step of the process is to synthesise a feedback loop for the amplifier. Since the
objective is to realize a pair of complex-conjugate poles, a bridged-T network is used since it
is the simplest RC network that can realize this. In fig. 1 such a network is shown together
with the transfer function from b to a, with a open-circuited. Another important aspect which
is easily shown is that the filter poles are identical to the zeros of the RC network.

1
1 1
1
+
s 2 + s +
C1 C2 R3 R3 R4C1C2
Eq. B1
t (s) =

1
1
1
1
+
s 2 + s
+
+
R3C1 R3C2 R4C1 R3 R4C1C2
Fig B1: bridged-T network with its transfer function

From the transfer function above and the fact that the pole polynomial of the active-filter
circuit will be equal to the numerator polynomial of the bridged-T network which gives that:

s2 + s

1
1 1
1
+
+ 02 = s 2 + s +
Eq. B2
Q
C1 C2 R3 R3 R4C1C2

If the above equation is remodelled then 0 and Q can be obtained as:

0 =

1
R3 R4 C1C 2

R3 R4 C1C 2
Q=
R3

1
1

+
C
C
2

-51-

Eq. B3-B4

Since 0 and Q are given, and for easy component selection C1 = C2 = C is chosen and the
resistors are denoted R3 = R and R4 = R / m , the above equations can be reformulated as:
m = 4Q 2
2Q Eq. B5-B6
RC =

0
We conclude that if the Q-value is given the first of the above equations can be used to
determine the ratio between the resistors. Then the values of 0 and Q can be substituted in
the second of the above equations to determine the time constant RC . Now the values of C
and R can be chosen arbitrarily.

Fig. B2: Feedback loop with signal injection path

When the feedback loop has been designed, the input signal needs to be connected. For this
reason the resistance R4 is divided into two parts as seen in fig. 2. These parts are
R4 and R4 (1 ) . Note that the parallel equivalent of these two is R4 . The reason for
choosing R4 as where the signal is injected is that any circuit node connected to ground can
instead be connected to an input voltage source without causing the poles in the transfer
function to change.

Fig B3: Analysis of circuit to determine the voltage transfer function

A nodal analysis conducted in a manner as described in fig. 3, in the order of the encircled
numbers where the final step, number 9, consists of writing the node equation at X and
substituting for Vx by the value determined in step 5 yields the following transfer function:

-52-

V0
=
Vi

s ( R4C1 )
Eq. B7

1
1
1
1
+
s 2 + s +
C1 C2 R3 R3 R4C1C2

The above function is recognized as a bandpass function whose center frequency gain can be
controlled by the value of . As expected the denominator polynomial is identical to the
numerator polynomial of t (s ) .
Dimensioning the components the easier way
Since the Single-amplifier biquadratic active filters have been used for quite some time more
or less useful guides have been developed to help facilitate the design of such filters. One of
the better examples of such guides is a short paper entitled Filter design in thirty seconds
written by Bruce Carter of Texas Instruments [3]. In this guide the following design procedure
is introduced for a narrow bandpass filter:

Fig B4: Schematic of filter as recommended in Filter design in thirty seconds

Design Procedure:
1) Pick C1 = C 2
1
2 C1 Frequency
3) Calculate R3 = 19 R1
2) Calculate R1 = R 4

4) Calculate R 2 =

R1
19

This is a great simplification compared to the analysis and design procedure as described
above. However this circuit has a smaller freedom than the one described previously since it
has a Q which is set to 10. With this design approach there is also a relation between the gain
and the Q factor which is:
R3 + R 4
= Q = Gain Eq. B8
2 R1
By using this design procedure a filter can be quickly implemented with no lengthy
calculations but with the drawback that the gain and Q factor cannot be set independently.
Choice of operational amplifier
Up until now very little has been said about the operational amplifier itself, but a critical
remark has been made that there is a great dependence on the limited gain and bandwidth of
-53-

the operational amplifier. Here this aspect will be developed in terms of the choice of
operational amplifier.
The gain and bandwidth of an operational amplifier under the assumption that >> b are
coupled to each other by the approximate relation where A0 is the DC-gain of the amplifier
and b is the 3-dB frequency:
A( j ) =

A
A0
A
0 b A( j ) 0 b

1 + j b
j

Eq. B9

From this relation it can be seen that that the gain A reaches unity (0dB) at a frequency
denoted by t and given by t = A0b . The frequency f t = t 2 is usually specified in data
sheets of commercially available operational amplifiers and is known as the unity-gain
bandwidth.
With t = A0b the above equation can be rewritten as
A( j ) =

f
A( j ) t = t Eq. B10

f
j

This equation has a high significance for choosing the operational amplifier used in the filter
since a low unity-gain bandwidth wont allow to have the desired amplification in the circuit.
The TL074, chosen for this amplifier in the thesis, has a unity-gain bandwidth of 3 MHz,
which means that in this application the maximum amplification is:
A( j )

f t 3MHz
=
= 75 Eq. B11
f
40kHz

Simulation of narrowband circuit used in thesis

The circuit was also simulated in Matlab based on the last transfer function derived in the first
section of the appendix and the component values used in the simulation were:
V0
=
Vi

C1 = C2 = 1nF

s ( R4C1 )
Eq. B12
1
1 1
1
2
+
s + s +
C1 C2 R3 R3 R4C1C2

R3 = 76.8k = 0.0496453 R4 = 199.6

-54-

Fig. B5: Gain and amplitude curve of narrowband amplifier transfer function

The conclusion of these simulations is that there is a good agreement between the simulated
results both in Spice and Matlab, as well as with the practical measurements of the
implemented circuit.
REFERENCES
[1] A.S. Sedra, K.C. Smith, Microelectronic circuits. Oxford; Oxford: Oxford university
press, 2004
[2] Sedra, A.; Ghorab, M.; Martin, K.; , "Optimum configurations for single-amplifier
biquadratic filters," IEEE Transactions on Circuits and Systems, vol.27, no.12, pp. 11551163, Dec 1980
[3] Internet: www.ti.com/lit/an/sloa093/sloa093.pdf, december, 2001 [Apr. 27, 2012]

-55-

APPENDIX C: DATASHEETS FOR KNOWLES ACOUSTICS SPM0204UD5

-56-

SPM0204UD5

ULTRASONIC ACOUSTIC SENSOR

Knowles Acoustics,
a division of Knowles Electronics, LLC

Revision: PRELIMINARY
Page 1 of 9

SPM0204UD5

1. DESCRIPTION AND APPLICATION

1.1. Description
Surface Mount Wide-Band Ultrasonic Acoustic Sensor
1.2. Applications
Hand held telecommunication devices, Positioning Sensing, Pneumatic Flow Sensing
2. PART MARKING

Tracking Number Convention

S:
1-6:

Manufacturing Location
Job Identification Number

S123456

3. TEMPERATURE RANGE
3.1. Operating Temperature Range: -40C to +100C
3.2. Storage Temperature Range:

-40C to +100C

Knowles Acoustics,
a division of Knowles Electronics, LLC

Revision: PRELIMINARY
Page 2 of 9

SPM0204UD5

4. ACOUSTIC AND ELECTRICAL SPECIFICATIONS

Symbol

Condition

Directivity

Min.

Limits
Nom.

Max.

Unit

Omni-directional

Frequency Range

Sensitivity

@ 1kHz (0dB=1V/Pa)

-51

Output impedance

ZOUT

@ 1kHz (0dB=1V/Pa)

Current Consumption

IDSS

Signal to Noise Ratio

S/N

Supply Voltage

Vs

Typical Input Referred

Noise
Sensitivity Loss across
Voltage
Maximum Input Sound
Level

10k

65k

Hz

-47

-43

dB

n/a

n/a

300

across 1.5 to 3.6 volts

0.100

n/a

0.250

mA

@ 1kHz (0dB=1V/Pa)

55

59

n/a

dB

1.5

n/a

3.6

n/a

35

n/a

dBA SPL

ENL

A-weighted

Change in sensitivity over 3.6v to No Change Across Voltage

1.5v
Range
At 100dB SPL, THD < 1%
At 115dB SPL, THD = < 10%

dB
dB

5. FREQUENCY RESPONSE CURVE

Knowles Acoustic MEMS Ultrasonic Sensor
Frequency Response
(Nominal Sensativity 0 = -47dBv/Pa)

30

20

Magnitude (dB)

10

-10

-20

-30
10,000

20,000

30,000

40,000

50,000

60,000

Frequency(Hz)

Knowles Acoustics,
a division of Knowles Electronics, LLC

Revision: PRELIMINARY
Page 3 of 9

SPM0204UD5

6. MECHANICAL SPECIFICATIONS
Acoustic Port Mesh



 

 

 

 

1
4



 

2  

 

Item

Dim.

Tol. (+/-)

Units

(H) Height

1.40

0.10

mm

(L) Length

4.72

0.10

mm

(W) Width

3.76

0.10

mm

(AW) Width

2.95

mm

(AL) Length

3.12

mm

 

 

PIN Designation
Pin # Function
1
Output
2
Ground
3
Ground
4
Power

Acoustic Area

Note:

Weight

0.07

Coplanarity

<0.1

mm

Special care must be used so that the Acoustic Port Mesh area is not touched or
disturbed during application.
(Tolerance +/-0.15mm unless otherwise specified)

Knowles Acoustics,
a division of Knowles Electronics, LLC

Revision: PRELIMINARY
Page 4 of 9

SPM0204UD5

7. RECOMMENDED LAND PATTERN

 
 !
 "#

 $%

8. RECOMMENDED INTERFACE CIRCUIT

+ Term 4

Dotted Section
Represents
Ultrasonic Sensor

R
Term. 1

Term. 3
Term. 2

Vref

Knowles Acoustics,
a division of Knowles Electronics, LLC

External Gain (set by

customer) = -R1/R2

Revision: PRELIMINARY
Page 5 of 9

SPM0204UD5

9. PACKAGE DETAIL
0. //
-#. //

8 . //9/. /4

132 . 4/65 //. /. 2

7

2 . ;49/. 2

&' (&)&' &*

4. 4/9/. /4

28 . //9/. :

/. ;4

4. 28

=/. :/>?>

@ACB.

-#. 2<

12 . 4/

+' *,

Model Number

Suffix

Reel
Diameter

Quantity
per Reel

SPM0204UD5

-2

1,200

SPM0204UD5

-6

13

4,800

Knowles Acoustics,
a division of Knowles Electronics, LLC

Leader
Length

800mm or minimum of 100

empty pockets

Label

Label applied to external

package and direct to reel.
Per JEDEC.

Empty
Units

No consecutive empty
pockets; No more than 3
empty pockets per reel. (Does
not include empty pockets for
leader/follower)

Revision: PRELIMINARY
Page 6 of 9

SPM0204UD5

10. SOLDER REFLOW PROFILE

260C
230C

170180C

Pre-heat

Solder Melt

120 sec.

100 sec.

Stage

Temperature
Profile

Time
(maximum)

Pre-heat

170 ~ 180 C

120 sec.

Solder Melt

Above 230 C

100 sec.

Peak

260 C maximum

30 sec.

Notes:

Do not pull a vacuum over the port hole of the microphone.

Pulling a vacuum over the port hole can damage the
device.
Do not board wash after the reflow process. Board
washing and cleaning agents can damage the device. Do
not expose to ultrasonic processing or cleaning.
Number of Reflow = recommend no more than 3 cycles.

11. ADDITIONAL NOTES

(A) Packaging (reference SiSonic_Packaging_Spec.pdf)
(B) Shelf life: Twelve (12) months when devices are to be stored in factory supplied,
unopened ESD moisture sensitive bag under maximum environmental conditions of 30C,
70% R.H.
(C) Exposure: Devices should not be exposed to high humidity, high temperature
environment. MSL (moisture sensitivity level) Class 2.
(D) Out of bag: Maximum of 90 days out of ESD moisture sensitive bag, assuming maximum
conditions of 30C/70% R.H.

Knowles Acoustics,
a division of Knowles Electronics, LLC

Revision: PRELIMINARY
Page 7 of 9

SPM0204UD5

12. RELIABILITY SPECIFICATIONS

Note: After test conditions are performed, the sensitivity of the microphone shall not deviate
more than 3dB from its initial value.
Test

Description

Thermal Shock

Microphone unit must operate when exposed to air-to-air thermal shock 100 cycles,
from 40C to +125C. (IEC 68-2-4),

High Temperature Storage

Test

Microphone unit must maintain sensitivity after storage at +105C for 1,000 hours.
(IEC 68-2-2 Test Ba)

Low Temperature Storage

Test

Microphone unit must maintain sensitivity after storage at 40C for 1,000 hours.
(IEC 68-2-1 Test Aa)

High Temperature Operating Microphone unit must operate within sensitivity specifications for 1,000 hours at
Test
105C. (IEC 68-2-2 Test Ba)
Low Temperature Operating
Test

Microphone unit must operate within sensitivity specifications for 1,000 hours at
40C. (IEC 68-2-1 Test Aa)

Humidity Test

Tested under Bias at 85C/85% R.H. for 1,000 hours. (JESD22-A101A-B)

Vibration Test

Microphone unit must operate under test condition: 4 cycles, from 20 to 2,000 Hz in
each direction (x,y,z), 48 minutes, using peak acceleration of 20 G (+20%, -0%).
(MIL 883E, method 2007.2, A)

Electrostatic Discharge

Tested to 8kV direct contact discharge to the case and tested to 2kV direct contact
to I/O terminals.

Reflow

Microphone is tested to 5 passes through reflow oven, with microphone mounted

upside-down under conditions of 260C for 30 seconds maximum.

Mechanical Shock

Microphone must operate after exposure to shock test of 10,000 G per IEC 68-2-27,
Ea.

Knowles Acoustics,
a division of Knowles Electronics, LLC

Revision: PRELIMINARY
Page 8 of 9

SPM0204UD5

13. SPECIFICATION REVISIONS

Revision

Detailed Specification Changes

Date

The information contained in this literature is based on our experience to date and is believed to
be reliable and it is subject to change without notice. It is intended as a guide for use by
persons having technical skill at their own discretion and risk. We do not guarantee favorable
results or assume any liability in connection with its use. Dimensions contained herein are for
reference purposes only. For specific dimensional requirements consult factory. This publication
is not to be taken as a license to operate under, or recommendation to infringe any existing
patents. This supersedes and voids all previous literature.
Knowles Acoustics,
a division of Knowles Electronics, LLC

Revision: PRELIMINARY
Page 9 of 9

APPENDIX D: DATASHEETS FOR MURATA MA40MF14-0B

-57-

1/7

Product Specification for Reference Only

Issued Date: 19 April, 2011
Rev. :

Part Description:

Ultrasonic Sensor

MURATA Part No.: M A 4 0 M F 1 4 - 0 B

The product specification in this sheet is for reference only.

The content of this specification is subject to change.
You are requested to receive the latest specification and to return one
copy of the specification to us with your receipt signature before going
into mass production.

Product Engineering Section

Sensor Products Department
Sensor Products Division
Device Unit
Murata MFG. Co.,Ltd.

MURATA MANUFACTURING Co., LTD.

2/7

Specification of Ultrasonic Transducer

Type : MA40MF14-0B
1. Scope
This product specification is applied to the water proof type ultrasonic transducer used for
obstacle detection system around vehicle.
Please contact us when using this product for any other applications than described in the
above.

2. Customer Part Number

3. Murata Part Number
MA40MF14-0B
4. Dimension
as per Fig.1

5. Absolute Maximum Ratings

5-1
5-2
5-3

Items
Maximum Input Voltage
Operating temperature range
Storage temperature range

Specification
160Vp-p
-40 to +85 deg C
-40 to +85 deg C

Note
Do not apply D.C. voltage.

6. Specifications (* Temperature 253 deg C, 45 to 60 % R.H, unless otherwise noted

6-1

Items
Operation Frequency

Specification
40 kHz

6-2

Sound pressure Level

more than 101 dB

6-3

Sensitivity

more than 87 dB

6-4

Beam pattern (Typical)

6-5
6-6

Capacitance
Insulation Resistance

(Typical)
110 deg x 50 deg
2700 pF +/- 20%
100 Mohm min.

Note
at 40kHz
distance is 30cm, input voltage is 10Vrms
0 dB = 20 uPa
(per measuring circuit Fig.2)
at 40 kHz
0 dB = 10V/Pa
(per measuring circuit Fig.3)
6dB down angle of overall sensitivity
at 1kHz
at 100V D.C.

MURATA MANUFACTURING Co., LTD.

Drawing No.

P.

3/6

Reference Only
7. Environmental tests (Standard Test Condition: 25 3 deg C, 45 to 60 % R.H)
7.1 Shock Test
The variation of the Sound Pressure Level at 40 kHz is within 3dB compared with initial
figures at 25 deg C after following test conditions
Acceleration
:
sine 980 m/s2 (100G), 6ms
Direction
:
3 directions
Shock time
:
3 times / directions
7.2 Vibration Test
The variation of the S.P.L at 40 kHz is within 3dB compared with initial figures at 25 deg C
after following test conditions
Vibration frequency:
10 to 200 Hz
Sweep Period
:
15 min.
Acceleration
:
43.12 m/s2 (4.4G)
Directions
:
3 directions
Time
:
96 hours / direction
7.3 Drop Test
The variation of the S.P.L at 40 kHz is within 3dB compared with initial figures at 25 deg C
after following test conditions
Height
:
1 meter onto concrete floor
Times
:
10 times
7.4 Pull Strength
There should be no substantial damage after 2.45 N of force.
7.5 High Temperature Test
The variation of the S.P.L at 40 kHz is within 3dB compared with initial figures at 25 deg C in
24 hours after following test conditions
Temperature
:
+85 2 deg C
Time
:
1000 hours
7.6 Low Temperature Test
The variation of the S.P.L at 40 kHz is within 3dB compared with initial figures at 25 deg C in
24 hours after following test conditions
Temperature
:
-40 3 deg C
Time
:
1000 hours
7.7 Humidity Test
The variation of the S.P.L at 40 kHz is within 3dB compared with initial figures at 25 deg C in
24hours after following test conditions
Temperature
:
+65 2 deg C
Humidity
:
90 to 95 % R.H.
Time
:
1000 hours
7.8 Heat Cycle Test
The variation of the S.P.L at 40 kHz is within 3dB compared with initial figures at 25 deg C in
24hours after following test conditions
Temperature
:
+85 3 deg C, 30 min
-40 3 deg C, 30 min
Cycles
:
1000 cycles

MURATA MANUFACTURING Co., LTD.

Drawing No.

P.

4/6

Reference Only
8.

Caution

8-1 Limitation of Applications

Please dont use for the applications listed below which require especially high reliability for the pr
evention of defects which might directly cause damage to the third party's life, body or property.
1) Aircraft equipment
2) Aerospace equipment
3) Undersea equipment
4) Power plant control equipment
5) Medical equipment
6) Transportation equipment (trains, ships, etc.)
7) Traffic signal equipment
8) Disaster prevention / crime prevention equipment
9) Data-processing equipment
10) Application of similar complexity and/or reliability requirement to the applications
listed in the above
8-2 Fail-safe
Be sure to provide an appropriate fail-safe function on your product to prevent a second damage t
hat may be caused by the abnormal function or the failure of our product.

9. Caution in use
9-1. Notice in design and usage
1) The transducer may generate surge voltage by mechanical or thermal shock. Care should
be taken to protect from it in designing your application circuit.
2) Please do not apply an excessive stress to the transducer because the piezo electric
element of the transducer might be damaged or inner cable might be disconnected.
3) The piezo electric element of the transducer may be damaged by force pressure from
back of the transducer.
4) Please do not apply D.C.voltage to the transducer to avoid failure. Electrode of piezo
electric element might be shorted out with electronic migration.
5) Please do not use the transducer in water.
6) Please hold the transducer with soft material such as rubber. The direct holding with hard
material will be cause of vibration leakage from, or into the transducer. It will be
influenced to decay time and short distance detection.
7) Please avoid humid penetration at the rear of the transducer in order to avoid short circuit.
Please use the transducer with waterproof design.
8) Please do not exceed 95 deg C / 120min in the assembly process and painting process to
avoid malfunction.
9) The transducer is designed for dual use purpose. Please do not use the transducer only
as receiver.
10) Care should be taken when select the material to hold, or cover backside of transducer. If
it contains sulfur or sulfide, electrode on piezo-electric element may corroded and
because of malfunction.

9-2. Notice in storage

1) The products should not be used or stored in a corrosive atmosphere, especially where
chloride gas, sulfide gas, acid, alkali or the like are present. Store the products in the
room where is normal temperature and humidity, and avoid the sunlight, sudden changes
in temperature and humidity. It may cause of failure or malfunction in such conditions.
2) Store the products where the temperature and relative humidity do not exceed -10 to 40
deg C, and 30 to 80%RH.Please Use the products within 6 months after receiving.

9-3. Notice in soldering and mounting

MURATA MANUFACTURING Co., LTD.

Drawing No.

P.

5/6

Reference Only
1) Please do not clean the transducer with water or solvent.
2) Please do not solder the transducer with flow or reflow soldering. Do not exceed the
soldering iron temperature 350 deg C and 3 sec.

10.

Note
1)
2)
3)

Please make sure that your product has been evaluated in view of your specifications
with our product being mounted to your product.
You are requested not to use our product deviating from the agreed specifications.
We consider it not to appropriate to include any terms and conditions with regard to the
business transaction in the product specifications, drawings or other technical
documents. Therefore, if your technical documents as above include such terms and
conditions such as warranty clause, product liability clause, or intellectual property
infringement liability clause, they will be deemed to be invalid.

MURATA MANUFACTURING Co., LTD.

Drawing No.

P.

6/6

Reference Only
Dimensions

Aluminum case surface

NON Hexavalent Chrome for anti-corrosion

9 +/- 0.1

40 +/- 5

3 +/- 0.1

Epoxy primer and Black color paint.

Lead wire : Black, AWG 30

(Connected to Aluminum-case )

Lead wire : Red, AWG 30

# : EIAJ code
unit : mm
12 +/- 0.2
14 +/- 0.1
Fig. 1

MURATA MANUFACTURING Co., LTD.

Drawing No.

P.

7/6

Reference Only
S.P.L.Test circuit

U.S

Amp.

S.C.M
30cm

Voltmeter

OSC.

Freq. Counter
Fig. 2
OSC.

: Oscillator

(Brel & Kjr 1013)

U.S

: Ultrasonic Sensor

S.C.M

: Standard Condenser Microphone

Amp.

: Amplifier (Brel & Kjr 2610)

(Brel & Kjr 4135)

Sensitivity Test circuit

Am p .
SP.

OSC .

S.C .M .

30 cm

U .S.
Vo lt m eter

RL
F req . C o un te r

Fig. 3
O SC.

Oscilla tor

(Br el & Kj r 1 01 3 )

U .S.

Ultra son ic Se nso r

S.C.M

Sta nd ard C o nd en se r M icro p ho n e

Am p .

Am plifier

SP.

Twe e te r

RL

3.9 k oh m

(Br el & Kj r 4 13 5 )

(Br el & Kj r 2 61 0 )

MURATA MANUFACTURING Co., LTD.

APPENDIX E: DATASHEETS FOR PHYSICAL ACOUSTICS SENSORS

-58-