WAVELETS TUTORIAL Peer-reviewed

An introduction to wavelet transforms: a tutorial approach

C M Leavey, M N James, J Summerscales and R Sutton Paper submitted 8 November 2002

Non-destructive testing (NDT) and condition monitoring aspects of the wavelet transform have led to growing interest in
techniques are among the most rapidly developing the technique by engineers involved in non-destructive testing
engineering disciplines. In the past decade methods have (NDT). NDT applications are varied, ranging from de-noising(12)
been sought to enable real-time or on-line analysis of to the measurement of dispersion curves for multimode Lamb
degradation in structures(1-3). Methods developed in the waves(13 and 14). These rely on the ability of the wavelet transform to
1980s based on modal analysis(4) have resurfaced, partly decompose a signal into spatially distributed frequency components
due to new developments in signal processing. Many of the that can then be selectively filtered according to the requirements
on-line NDT techniques currently under research utilise of the application.
wavelet transforms(5-9) to decompose spatial signals into the Lamb wave propagation in plates is generally found to be a
time/frequency domain. The aim of this paper therefore, is to depth-related phenomenon, the depth affecting the dispersion
provide a friendly tutorial to the two most commonly used properties of the waves. Neithammer and Jacobs(13) have
forms of the wavelet transform. It is hoped that this article demonstrated that the use of wavelets for crack detection in plates
will provide interested readers with sufficient knowledge of produces superior results to the more commonly used short-time-
the subject to allow productive usage of the various software Fourier-transform (STFT). A similar technique has been applied to
packages offering wavelet technologies. Being aware of the analysis of dynamic strain data(15), where the decomposition of
the building knowledge provided both by literature and the signals taken from beams allowed identification of phase velocities
internet, a list of useful references and world-wide-web links and hence improved accuracy in the determination of crack lengths
has been provided at the end of the paper. and location. Wavelets are particularly good in the analysis of
noisy data through decomposition of the frequency components
1. Introduction contained within a signal. Many applications exist where the
wavelet transform has been used to remove unwanted noise from
The wavelet transform is a relatively new tool to be taken up by a signal allowing for improved damage identification. Sasi et al(16)
the engineering community. Particular strengths lie in its noise applied the wavelet transform to analysis of eddy-current data
and data reduction abilities, which have been exploited in a taken from stainless steel cladding tubes. In this instance a discrete
number of practical applications(10 and 11). These, and many other version of the wavelet transform was used to improve the signal-
to-noise ratio.
Mr Christopher Leavey, BEng(Hons), MPhil, GMIMarEST, MInstP is a Common applications of the discrete versions of the wavelet
research student in the Department of Mechanical and Marine Engineering transform are in data reduction and feature extraction. Due to the
at the University of Plymouth and a member of the Marine and Industrial physical significance of the wavelet transform in the analysis of
Dynamic Analysis Research Group. His main research interests lie in smart oscillating signals, feature extraction and data reduction can often
structures particularly autonomous condition/health monitoring and life be considered as being the same thing. A wavelet packet approach
prediction of carbon fibre reinforced plastics. was adopted by Yang(17) to improve classification of damage in
Professor M Neil James, BSc(Eng), PhD, Hon DTech is currently Professor structures. This work found that wavelet packets were particularly
of Mechanical Engineering and Dean of the Faculty of Technology at the good at identifying characteristics relating to damage. For a number
University of Plymouth. He is a Chartered Engineer and a Fellow of the of damage scenarios, wavelets provided very high detection rates.
Institute of Materials, Minerals & Mining. Wavelets are simply mathematical functions exhibiting some
Dr John Summerscales, IEng, FIMMM, FInstNDT, FIAQP is Reader in kind of oscillatory behaviour. Unlike sinusoidal functions, where
Composites Engineering in the Department of Mechanical and Marine the oscillation dominates the entire signal, wavelets show only
Engineering at the University of Plymouth. He is a member of the Advanced localised oscillation. Wavelet analysis involves breaking a signal
Composites Manufacturing Centre with research interests in composites into different frequency components by comparing the signal to a
manufacturing, process control, total quality management, non-destructive number of differently sized wavelet functions. Wavelets are closely
testing, microscopical analysis, natural fibre reinforcements, smart allied to Fourier analysis, but have the advantage of being able
materials and intelligent structures. Contact address: ACMC/DMME,
University of Plymouth, Plymouth PL4 8AA, UK. Tel/Fax: +44 1752 23
to overcome some of their commonly associated constraints. To
2650; e-mail: [email protected] understand how this is achieved, a review of the issues follows.
Signal processing is the term given to the process of
Professor Robert Sutton, BEng(Tech), MEng, PhD, CEng, MIMechE, extracting, analysing and modifying information contained
FIEE is Professor of Control Systems Engineering in the Department of within signals. Signal processing, or digital signal processing
Mechanical and Marine Engineering at the University of Plymouth and a
member of the Marine and Industrial Dynamic Analysis Research Group.
(DSP), is an extremely important engineering technology
His main research interests lie in the application of advanced control spanning many disciplines including image processing, condition
engineering and artificial intelligence techniques to control problems. monitoring, instrumentation and control, telecommunications and
Ongoing research is concerned with the use of fuzzy logic, neurofuzzy biomedicine(18). Signals generally carry time-varying information
algorithms, artificial neural networks and adaptive search algorithms in such as acceleration, force, temperature and strain. Simple analysis
the design of novel control systems for industrial and marine plant. of such data has enabled engineers and scientists to determine
Contact address: Department of Mechanical and Marine Engineering, derived properties relating to the physical process/system under
University of Plymouth, Drake Circus, Plymouth, Devon PL4 8AA. E-mail: consideration (for example derivation of Young’s modulus from
[email protected] stress/satran data).

344 Insight Vol 45 No 5 May 2003

 In the analysis of vibration data, time-related changes to a
physical variable may not be sufficient to characterise fully the
dynamic behaviour of a system. Often the most valuable data in
such areas relates to how a system reacts to changes in frequency.
In the determination of this frequency response data two options
are available. Either the system’s response to direct sinusoidal
excitation can be analysed or, more commonly, the time response
can be transformed via mathematical methods to the frequency
domain. Fourier showed how simple periodic functions could
be represented by an infinite series of sinusoidal functions. His
work eventually led to the development of the Fourier transform
named after him. The Fourier transform is an extension of the
simpler Fourier series, which allows non-periodic signals (such as
those of interest in DSP) to be broken down into a finite number
of sinusoidal functions of differing amplitude and frequency. The
Fourier transform reveals both the frequency components present
in the time signal and the magnitude of the oscillations (in other
words the frequency response). For a review of the key concepts
refer to the literature(18-23). Despite the continued dominance of the
Fourier transform, the technique has a number of well-established
constraints. The Fourier transform is defined over infinite time.
This means that the effect of transforming a time signal, having
finite duration, to the frequency domain is to convolve1 (*) the
Fourier transform of the signal with the Fourier transform of the
window function, Figure 1. The frequency spectra of the window
function can often prove detrimental to the time signal’s frequency
representation, reducing the power and introducing additional
components in an effect known as leakage. Further frequency
components can be introduced due to edge effects where the
window function causes discontinuities in the time signal.
An additional limitation on the use of the Fourier transform Figure 1. Fourier transform windowing and leakage
is the lack of time information in the resulting transform. Fourier the CWT, the DWT is then explained in section 3. Examples are
techniques were designed for use with stationary signals. A signal used throughout this and an attempt is made to relate the transform
is stationary if its statistical properties, for example average and of the CWT to that of the DWT. To conclude these two sections
variance, do not change with time. However, most practical signals a note on the selection of appropriate transforms (CWT or DWT)
contain time-varying frequency data and so cannot be classified in and their corresponding wavelet functions is given in section 4.
this manner. An attempt to solve this time resolution problem has A review of the relevant mathematics associated with the CWT
been made through the development of the STFT. STFT uses small and DWT, including further recommended reading, is given in
time-shifted windows to approximate time/frequency information, Appendices I and II.
giving bands of frequencies over time increments. The accuracy
of the STFT in terms of time and frequency resolution is highly
2. Overview of the continuous wavelet
dependent on the size of the window selected for the transform.
Large windows offer good frequency resolution but poor time transform
resolution while short windows offer good time resolution but poor Of the wavelet transforms available, the CWT is possibly the
frequency resolution. simplest to visualise, providing a convenient introduction to the
The resolution problem associated with the STFT has been subject as a whole. The CWT is similar to the Fourier transform
found to be a result of Heisenberg’s uncertainty principle(18), which where an arbitrary function of time can be represented by an
states that time and frequency cannot be resolved simultaneously, infinite summation of sinusoidal functions and their multiplicative
in other words it is impossible to achieve accuracy in both coefficients. In wavelet analysis sinusoidal functions are replaced
quantities at the same time. One possible solution to this problem with wavelet functions. The essence of the calculation is then to
is to adopt a flexible windowing strategy and selectively transform determine the coefficients necessary to accurately portray the time
the signal according to the time/frequency requirements. This is the function. (See Appendix I).
method used in wavelet analysis, a technique formally presented in The wavelet is a mathematical function usually of time. The
the 1980s(24), and fast gaining popularity within DSP and in many terms wavelet and wavelet function are used interchangeably
industrial applications. herein. A wavelet is a small wave: it must be oscillatory and have a
Undertaking an analysis using wavelets is a relatively limited duration. The use of functions localised in time removes the
straightforward process. Complications are introduced through the windowing requirements commonly found in Fourier techniques.
differing approaches adopted and the conditions required to satisfy However, the term windowing or tiling is still common in wavelet
each implementation. The two most commonly used forms of the analysis. The leakage effects suffered by the Fourier transform (see
wavelet transforms are the continuous wavelet transform (CWT) section 1 above) caused through windowing are not present in the
and the discrete wavelet transform (DWT). Section 2 reviews the wavelet transform and therefore signal information is retained.
CWT providing a simple illustration of how the transformation Wavelets must meet a number of mathematical conditions which
process is achieved. Having gained a fundamental knowledge of limit the types of functions that may be used in the analysis. The
1
Convolution is a special kind of ‘product’ of functions. Convolution discrete transform must satisfy a greater number of conditions
has the property that the Fourier (or Laplace) transform of f(x) *g(x) is than the CWT. The reasons for this are given in the following
the Fourier transform of f(x) multiplied by the Fourier transform of g(x). two sections. Three examples of wavelet functions are shown in
Reference(22) provides further details of convolution theory. Figure 2.

Insight Vol 45 No 5 May 2003 345

 in the positive time direction and a new coefficient calculated, this
process being continued until the end of the signal is reached. At
this point the scale is increased, the wavelet being stretched and
dilated, and the analysing signal returned to the starting point,
ie t = 0, allowing calculation of the wavelet coefficients at the next
scale. The transformation is complete once the signal has been
analysed for all scales, Figure 3.
The wavelet transform performs a comparison of wavelet
to signal. A high degree of similarity exists between the two
functions when the coefficients at that translation and scale are
large. If the two are dissimilar the coefficients are small. In this
way the transform process gives an indication of the frequency
content of the signal. As an approximate relationship exists
between scale and frequency, a large coefficient at a particular
scale implies the presence of a particular frequency. The ability
Figure 2. Examples of wavelets – (a) Morlet; (b) Mexican hat; of the wavelet transform to give accurate time and frequency
(c) Gaussian
information is constrained by the uncertainty principle in the same
The definition of a wavelet requires an understanding of the way as Fourier transforms. Stated simply, this means that it is not
process of scaling and translation. To scale a wavelet means to possible to achieve both good time and good frequency resolution
stretch or dilate: this ensures that the energy contained within simultaneously. This is clearly demonstrated with the Fourier
the scaled wavelets is the same as the original ‘mother’ wavelet. transform where there exists no time information at all due to the
As the wavelet is stretched in the horizontal x-axis direction it is infinite window implicit in the technique. The STFT achieves time
squashed in the vertical y-axis direction. Translation moves the resolution through the reduction of this window: as the window
wavelet, usually in the positive direction, along the x-axis. In most is reduced, time resolution improves at the expense of frequency
cases the x-axis will represent time, while the y-axis is amplitude. resolution.
In practice, the horizontal axis is not always ‘time’ but will always Analysis of differing scales in the wavelet transform provides
contain time-varying information. The transform process is flexible windowing. The resulting transform will have varying
achieved through continued scaling and translation of the mother time/frequency resolution as demonstrated in Figure 4. It is clear
wavelet along the length of a signal. from the wavelet transform process that the lower scales (higher
Before a transform can be performed, a wavelet function must frequencies) occupy small windows where the wavelet is highly
be selected. For the current discussion a generic function will be compacted. As the window size is reduced, time resolution
assumed; for information on wavelet selection see section 4. The improves at the expense of frequency resolution (as in STFT).
term scale is often used in relation to the frequency of the wavelet Figure 4 illustrates this by the tall thin tiles located at the top of
function. Scale is inversely proportional to frequency, so low scale the diagram. The higher scale (lower frequency) components are
relates to the most tightly packed (high frequency) wavelets. A stretched and occupy a larger window, thus having poorer time
consequence of this is that typical wavelet decompositions are resolution and better frequency resolution (again this is similar to
a function of time and scale as opposed to time and frequency. STFT with large windows). The flexible tiling scheme adopted by
Wavelet transforms generally begin using low scale (high the wavelet transform provides a trade-off between frequency and
frequency) wavelet functions progressing to high scales (low time resolution. This usually proves advantageous in the analysis
frequency) where the wavelet is at its most dilated. The analysing of most signals. Real signals tend to exhibit high frequency
wavelet is set at the beginning of the signal, t = 0. The product of components for short durations and low frequency components for
the wavelet and signal are then integrated over all time. The results long durations.
are then normalised giving the wavelet coefficient for that scale and
translation. The wavelet is then moved by some small increment

Figure 4. Wavelet analysis flexible windowing scheme

Having briefly examined the CWT transform process the

obvious question that remains is what does a signal look like in the
time-frequency/time-scale domain? Consider a simple sinusoidal
signal of finite duration having a frequency of 50 Hz: the frequency
domain representation is a single vertical line (neglecting leakage),
see Figure 5. At any point within the duration of the signal the
frequency will be 50 Hz, hence in this instance the time-frequency
graph of this function can be constructed by projecting the single
component of Figure 5 throughout the signal’s time duration,
Figure 3. Wavelet analysis overview Figure 6. Time-frequency and time-scale plots can be displayed in

346 Insight Vol 45 No 5 May 2003

 wavelet analysis are generally quite similar to those produced by
using time-frequency methods; there are, however, a number of
notable differences as demonstrated in the following example. In
this instance the signal to be analysed consists of a low frequency
sine wave (50 Hz) of duration 1 second, followed by a high
frequency sine wave (150 Hz). For the STFT and the continuous
wavelet transforms the time-frequency/time-scale plots are shown
in Figure 8(a) and 8(b). The two figures are inverted as scale is
inversely proportional to frequency. The effect of the resolution
trade-off discussed previously is also quite clear from the wavelet
transform of Figure 8(b). The low frequency components derived
from dilated wavelets offer greater frequency resolution (poorer
time resolution) than the high frequency components. The first
harmonic covers a frequency range of approximately 20 Hz
while the second covers about 70 Hz; this scale to frequency was
approximated using the following relationship:
Fc ◊ D
s=
Figure 5. Fourier transform of a 50 Hz sinusoidal function
f
where f is the frequency, s the scale, D is the sample period and
Fc the centre frequency defined as an approximate measure of
the oscillatory nature of the basis function at its centre. Looking
at the STFT of Figure 8(a) there appears to be an anomaly in the
diagram, a vertical line separating the two dominant harmonics.
This anomaly is a direct result of the time-frequency resolution
problem associated with fixed windowing (selection of smaller
time windows can go some way to reduce this problem but at
the cost of frequency information). Clearly, the solution to the
resolution problem offered by the wavelet transform produces
additional complexities in the interpretation of the coefficients. The
overall effect, however, is beneficial.

Figure 6. Discrete frequency components with common

frequency
either three-dimensions (time, frequency and amplitude) or two-
dimensions (time and frequency). The above example is plotted in
two-dimensions, Figure 7. In this instance the axes are time and
frequency due to the use of a STFT. For a wavelet decomposition
the axes will be time and scale. In Figure 7, high amplitude
components are shown in white and zero amplitude in black,
grey shading providing intermediate levels. The visualization
of a sinusoidal signal with multiple harmonics (sinusoids with
frequencies of integer factors greater than the fundamental
frequency) can be represented in the same manner as the above
example. Each horizontal band relates directly to the frequency
and duration of the individual harmonics. Time-scale plots using

Figure 8. Comparison of STFT and CWT – (a) STFT;

(b) Continuous wavelet transform using Morlet basis

3. Overview of the discrete wavelet transform

The above review considered the continuous wavelet transform.
This provided a demonstration of some of the key concepts
involved in wavelet analysis. The CWT is a computationally
demanding algorithm so, as with Fourier techniques, the discrete
wavelet transform (DWT) has been developed. The DWT differs
from the CWT in that the method of computation utilises subband
coding (an alternative technique for computation of the DWT
using pyramidal coding can be found in(19)). As there is a great deal
of redundancy in the data contained within the CWT, the DWT
utilises sampling of both scale and time data, thereby producing
a substantially faster algorithm. Scale and time are sampled in
Figure 7. Time-frequency representation of a 50 Hz sinusoid powers of two (21, 22… etc); in most texts this is commonly termed

Insight Vol 45 No 5 May 2003 347

Figure 9. CWT and DWT using Daubechies db7 basis
dyadic sampling. DWT data is generally displayed as a series of Y-axis notation is ‘A’ for an approximation and ‘D’ for a detail;
plots where each plot represents a specific scale. Using the notion D1 is low scale with increasing numbers for higher scales. The
of dyadic sampling these will be scales 2, 4, 8, etc. The data above approximation usually shows the underlying trend of the data:
(Figure 8) reanalysed using a DWT algorithm is shown in Figures 9 the low frequency components. In this instance it is more or less
and 10, where Figure 10 shows the left-hand-side of Figure 9. The a straight line. The remaining detail levels show how the signal
analysing wavelet in this instance was developed by Daubechies(25), changes between time samples. The DWT contains the same
known as the ‘db7’, see Figure 11. (See Appendix II). essential information as the CWT (see the lower four graphs of
Figure 9 demonstrates how the DWT diagrams relate to the CWT Figure 10, particularly ‘D2’ and ‘D4’ which relate to scales of 4
diagram (Figure 8). Examining the two diagrams simultaneously it and 16 respectively). As a point of interest, the ‘D3’ level found in
can be seen that the high amplitude parts of the signals relate to Figure 10 is not present in the CWT of Figure 8 (scale) due to the
the light areas in the CWT. The essential information displayed in adoption of a different wavelet function.
the CWT can be fully represented in the six plots of the DWT. The Because the DWT uses dyadic sampling it is easy to relate the
levels ‘D1’ to D5’ of Figure 10 to the scales shown in Figure 8. The
CWT of Figure 8 is analysed over 32 scales, hence the DWT must
be analysed over 5 levels: 21, 22…25 giving the samples scales 2,
4…32. These five levels represent the detail diagrams displayed in

Figure 10. Example of DWT using Daubechies db7 basis Figure 11. Daubechies db7 wavelet function

348 Insight Vol 45 No 5 May 2003

Figure 10 where level D1 corresponds to scale 2 (21), D2 to scale
4 (22), D3 to scale 8 (23), etc. The DWT differs from the CWT in
that it makes use of a scaling function in addition to the wavelet
function previously discussed; this is the reason for the additional
diagram in Figure 10. In DWT analysis, the wavelet function is
used to determine the detail coefficients while the scaling function
determines the approximation coefficients. The nature of the
approximation coefficients depends on the level of decomposition
that is undertaken. Each level of decomposition removes more high
frequency data from the approximation coefficients. In the case
of Figure 10 only a small amount of low frequency data has been
retained, leaving the general trend of the signal.
In the frequency domain the wavelet function has band-pass
properties as shown in Figure 12. Scaling (stretching) of the wavelet
function acts to compress the frequency band-pass window, due to
a reduction in the wavelet’s sampling time, and reduces the band-
pass frequencies. Time compression of the wavelet by a factor of 2
will stretch the frequency spectrum of the wavelet by a factor of 2
and also shift all frequency components up by a factor of 2; this is
shown diagrammatically in Figure 13 where the wavelet function is Figure 14. Daubechies db4 scaling function and its Fourier
transform showing the low-pass nature of the function
represented by the Greek letter y. From Figure 13 it is clear that the
wavelet transform can be calculated through the dyadic sampling 4. A note on transform and wavelet selection
and filtering of the wavelet function. The problem with this is that
The foundations of wavelet analysis lie in linear algebra and as
in order to cover the entire frequency spectra from zero to the upper
such some of the terminology is common to both. The wavelet
bounds, an infinite number of scaled wavelet functions would be
functions are commonly referred to as bases (basis in singular).
required. The solution is to produce a function having low-pass
A basis or basis set essentially describes a coordinate system
properties using the scaling function mentioned previously, see
which is linearly independent, allowing a signal or a function to be
Figure 14. The DWT is therefore calculated by passing the signal
written as a linear combination of the basis set. A simple example
through a filter-bank consisting of a high- and low-pass filter:
of a commonly used basis set are the unit vectors i, j and k used
the wavelet function and scaling function respectively. On each
in three-dimensional Euclidean space. Fourier analysis uses the
pass through the filter bank the scaling function captures the low
same principles of linear independence in the construction of an
frequency data from the previous approximation. The effect of
infinite summation of sinusoids, themselves being basis functions.
dyadic sampling of the scaling function is to progressively reduce
For clarity, the idea of linear independence with reference to the
the low frequency components contained in each subsequent
Fourier series means that none of the bases can be constructed
approximation. The filtering process as a whole is shown in
from a summation of any of the other bases. Choice of a wavelet
Figures 15 and 16, which demonstrate the actions of both the
can often appear quite arbitrary(27), but should be guided by the
wavelet and scaling function in the process of decomposition.
following:
Orthogonal or Nonorthogonal – The orthogonality of a wavelet
has a significant impact on the type of decomposition that can be
performed on a signal. Orthogonality implies that vectors are at
right angles to one another. The general definition of the scalar
product of two vectors A and B is A.B = ABcosf, where f is the
angle between the vectors. If the vectors A and B are orthogonal (in
other words at right angles to one another) this product is zero as
the angle f will be 90º. Strictly speaking, the use of the CWT does
not require an orthogonal basis, although most are at least partially
orthogonal. With reference to Figure 8, 9 and 10 it is easy to see
that the CWT contains a great deal of information as compared to
the DWT, which requires an orthogonal basis (note that despite the
additional information in the CWT representation all the essential
detail is fully captured in the DWT). A lot of redundant information
exists in the CWT and subsequently the transform and its inverse
are computationally demanding. Through the use of orthogonal
bases (DWT) it is possible to reduce the information required for
decomposition and reconstruction of the signal and hence provide
Figure 12. Daubechies db4 wavelet basis and its Fourier opportunities for data reduction. This is the approach adopted in
transform showing the band-pass nature of the wavelet
image compression and has been successfully used in (28) and (29)
function
to replace the more commonly used cosine transform. Calculation
of the CWT can use either orthogonal or nonorthogonal bases.
Despite its computational requirement, this algorithm can give a
good overview of a signal’s features where smooth, continuous
variations in wavelet amplitude are expected(27).
Complex or Real – As with Fourier techniques, wavelet bases can
Figure 13. Wavelet spectra resulting from dilation of the mother be found in real and complex forms. Complex functions provide the
wavelet (after Ref 25) opportunity of determining both magnitude and phase information

Insight Vol 45 No 5 May 2003 349

Figure 15. Decomposition of a signal through application of the
DWT

Figure 17. Haar wavelet basis

and time/frequency data but is difficult to use practically due to
its computational demands. Where real-time analysis is required,
the CWT is unsuitable despite being the more stable of the two
algorithms in terms of shift invariance.
The DWT has been suggested for use in data reduction
applications and it is clear that in real-time analysis this is the more
logical of the two algorithms to use. The shift invariance problem of
the algorithm is an area of current research and solutions have been
suggested through the use of complex bases(30). A further difference
is clearly evident in the type of basis that can be used with each
of the algorithms. The CWT makes use of a single wavelet basis,
whereas the DWT requires both wavelet and scaling functions; the
DWT also requires that both functions are orthonormal. A further
extension to the DWT algorithm is available with wavelet packet
analysis. This enables each of the detail coefficients to be filtered
in the same way as the approximation coefficients, hence providing
flexibility in signal representation with respect to the design
objectives.
There are a number of time/frequency algorithms available,
complementing the wavelet transforms. The STFT is possibly one
Figure 16. Decomposition process using high(right) and of the most well-known time/frequency techniques. STFT can be
low(left) pass filters extremely useful in the analysis of certain signals providing an
from a signal; this has been suggested to be useful in the analysis appropriate balance can be made between time and frequency
of oscillating signals(27). More recently, complex wavelet functions resolution. The wavelet transform and STFT are both linear
have found application in signal processing where the difficulty transforms obeying superposition; an extremely useful nonlinear
of applying the DWT to consistently process time-shifted signals algorithm exists in the Wigner-Ville transform(31) which gives
(fundamentally the same signal captured at a different point in instantaneous energy information between time samples. Cohen(32)
time) has prevented the use of pattern recognition techniques(29). presents a good review of time/frequency distributions including a
discussion of the relative merits of each.
Width – Remembering the windowing problems of the STFT,
the width of the wavelet basis has a similar impact on the time-
frequency resolution. For a wide basis function, time resolution
6. Conclusion
will be sacrificed at lower scales whereas a short basis function will An overview of the continuous and discrete wavelet transforms
sacrifice frequency resolution at high scales. has been given in the above text including where applicable
examples of their use in the analysis of signals. A number of
Shape – The shape of the wavelet is one of the most important
references have been given for additional study, in particular
considerations in the selection of a basis. Generally, the shape of
detailing the supporting mathematics. This paper attempts to
the function should show similar characteristics to the signal being
develop and explain some of the key terms attached to wavelet
analysed. In some cases a basis can be selected on the function
analysis providing a good introductory level text for those new to
it is to perform, for example the Haar function (see Figure 17) is
the subject. (See Appendix I and II).
particularly good at detecting discontinuities in a signal due its own
discontinuous nature. Acknowledgement
The principal author (CML) is grateful to the EPSRC for support
5. Discussion under the Doctorial Training Accounts scheme.
The wavelet transform (CWT and DWT) is essentially a time/
frequency analysis tool which goes some way to solving the World Wide Web Resources
resolution problems of Heisenberg’s uncertainty principle. There 1. Valens C. 1999, A Really Friendly Guide to Wavelets: http://
are clearly some notable differences in the two representations perso.wanadoo.fr/polyvalens/clemens/wavelets/wavelets.html
highlighted by this review. The CWT offers maximum detail 2. Institute of Electrical and Electronics Engineers 1995, An

350 Insight Vol 45 No 5 May 2003

 Introduction to Wavelets: http://www.amara.com/IEEEwave/ 15. S T Quek, Q Wang, L Zhang and K H Ong, ‘Practical issues
IEEEwavelet.html in the detection of damage in beams using wavelets’, Smart
3. Wavelet Digest: http://www.wavelet.org/wavelet/index.html Materials and Structures, Vol 10, pp 1009 – 1017, 2001.
4. Polikar R. 1999, The Wavelet Tutorial: http:// 16. B Sasi, B P C Rao, S Thirunavukkarasu, T Jayakumar and
engineering.rowan.edu/~polikar/WAVELETS/WTtutorial.html P Kalyanasundaram, ‘Wavelet transform based method for
5. Lewalle J. 1995, Tutorial on Continuous Wavelet Analysis of eddy current testing of cladding tubes’, Retrieved January 4,
Experimental Data: http://www.ecs.syr.edu/faculty/lewalle/ 2003 from the World Wide Web: http://www.nde2002.org/
tutor/tutor.html papers/017P.pdf
6. Altmann J. 1996, Surfing the Wavelets: http://www.wavelet.org/ 17. X Yang, ‘Reducing measuring points of automatic damage
wavelet/tutorial/wavelet.htm detection system based on wavelet packet analysis under noisy
7. Wavelet Forum: http://www.ondelette.com/servlets/com. conditions’, Retrieved September 9, 2002 from the World Wide
o n d e l e t t e . s e r v l e t . w e b f o r u m . We b F o r u m S e r v l e t ? _ Web: http://www.wbmt.tudelft.nl/pto/research/conferences/
FORUMFILE_=forumWavelet.properties Proceedings/Sci2001/Paperspdf/P001221.PDF
8. Mathsoft’s Wavelet Resource Page: http://www.mathsoft.com/ 18. E C Ifeachor and B W Jervis, ‘Digital Signal Processing: A
wavelets.html Practical Approach’, 2nd Edition, Prentice-Hall, 2002, Harlow,
England, ISBN:0-201-59619-9.
NB: All web links were known to be active on August 2002.
19. R B Randall, ‘Frequency Analysis’, 3rd Edition, Brüel and
Kjær, 1987, Denmark, ISBN: 87-87355-07-8.
References
20. D E Newland, ‘An Introduction to Random Vibrations, Spectral
1. P Hajela and F J Soeiro, ‘Recent Developments in Damage and Wavelet Analysis’, 3rd Edition, 1993, Longman Scientific
Detection Based on System Identification Methods’, Structural and Technical, Bath, England, ISBN: 0-582-21584-6.
Optimization, Vol 2, pp 1-10, 1990. 21. J A Cadzow and H F Van Landingham, ‘Signals Systems and
2. D Armon, Y Ben-Haim and S Brun, ‘Crack Detection in Beams Transforms’, Prentice-Hall, 1985, Englewood Cliffs, New
by Rank-Ordering of Eigenfrequency Shifts’, Mechanical Jersey, ISBN: 0-13-809542-6.
Systems and Signal Processing, Vol 8, No. 1, pp 81-91, 1994. 22. S V Vaseghi, ‘Advanced Digital Signal Processing and Noise
3. K D Hjelmstad and S Shin, ‘Crack Identification in a Cantilever Reduction’, 2nd Edition, Wiley, 2000, Chichester, New York,
Beam from Modal Response’, Journal of Sound and Vibration, ISBN: 0-471-62692-9.
Vol 198, No 5, pp 527-545, 1996. 23. P V O’Neil ‘Advanced Engineering Mathematics’, 4th Edition,
4. C-P Fritzen and K Bohle, ‘Model-Based Damage Identification Brooks/Cole Publishing, 1995, California, USA, ISBN: 0-534-
from Changes of Modal Data – A Comparison of Different 94320-0.
Methods’, Key Engineering Materials, Vols 204-205, pp 65- 24. P Goupillard, A Grossman and J Morlet, ‘Cycle-Octave
74, 2001. and Related Transforms in Seismic Signal Analysis’,
5. S Y Huang, G Z Qi and J C S Yang, ‘Orthonormal Wavelets for Geoexploration, Vol 23, pp 85-102,1984.
System Identification’ 14th Modal Analysis Conference, Vol 1, 25. I Daubechies, ‘Ten Lectures on Wavelets’, CBMS-NSF
pp 696-701, 1996. regional conference series in applied mathematics 61, 1992,
6. Q Wank and X M Deng, ‘Damage Identification with Spatial Philadelphia: SIAM.
Wavelets’ International Journal of Solids and Structures, 26. C Valens, ‘A Really Friendly Guide to Wavelets’, 1999,
Vol 36, pp 3443-3468, 1999. Retrieved June 24, 2002 from the World Wide Web: http://
7. S T Quek, Q Wang, L Zhang and K K Ang, ‘Sensitivity perso.wanadoo.fr/polyvalens/clemens/wavelets/wavelets.html
Analysis of Crack Detection in Beams by Wavelet Technique’, 27. C Torrence and P C Gilbert, ‘A Practical Guide to Wavelet
International Journal of Mechnaical Sciences, Vol 43, pp 2899- Analysis’, Bulletin of the American Metrological Society,
2910, 2001. Vol 79, No 1, pp 61-78, 1998.
8. Y-F Chang, ‘Wavelet Deconvolution Before Scanning in 28. C M Brislawn, ‘Fingerprints go Digital’, Notices of the AMS,
Ultrasonic Non-Destructive Testing’, Insight, Vol 44, No 11, Vol 42, No 11, pp 1278-1283, 1995.
pp 694-699, 2002. 29. D Lemire, ‘White Paper on Image Compression using the Fast
9. Q Wang, and X Deng, ‘Damage Detection with Spatial Wavelet Transform’, Concrete Industries Research Association,
Wavelets’, International Journal of Solids and Structures, 1999.
Vol 36, pp 3443-3468, 1999. 30. N Kingsbury, ‘Complex Wavelets for Shift Invariant Analysis
10. B Mazieres, ‘A new approach for fingerprint image and Filtering of Signals’, Applied and Computational
compression,’ Tech. Rep. LA-UR-98-4298, Los Alamos Nat’l. Harmonic Analysis, Vol 10, pp 234-253, 2001.
Lab, Jan. 1998. 31. J C Goswami and A K Chan, ‘Fundamentals of Wavelets
11. C J Lian, K F Chen, H H Chen, L G Chen, ‘Lifting based Theory, Algorithms and Applications’, John Wiley and Sons,
discrete wavelet transform architecture for JPEG2000’, 2001 1999, Inc., New York, USA, ISBN: 0-471-19748-3.
IEEE International Symposium on Circuits and Systems 32. L Cohen, ‘Time-Frequency Distributions – A Review’,
(ISCAS’2001), Sydney, Australia, May 2001. Proceedings of the IEEE, Vol 77, No 7, pp 941-981, 1989.
12. J N Watson, P S Addison and A Sibbald, ‘Temporal filtering of 33. A D Poularikas, ‘The Transforms and Applications Handbook’,
NDT data using wavelet transforms’, Retrieved February 18, 2nd Edition, IEEE Press, 1999, Florida, USA, ISBN:0-8493-
2003 from the World Wide Web: http://www.ce.utexas.edu/ 8595-4.
em2000/papers/JNWatson.pdf 34. C S Burrus, A G Ramesh and H Guo, ‘Introduction to Wavelets
13. M Niethammer and J Jacobs, ‘Time-frequency representation and Wavelet Transforms: A Primer’, Prentice-Hall, 1998.
of Lamb waves using the reassigned spectrogram’, 35. Y M Hawwar and A M Reza ‘Nonlinear Filtering in the
Retrieved March 23, 2002 from the World Wide Web: http: Wavelet Transform Domain’, IEEE International Conference
//www.ce.gatech.edu/~ljacobs/Marc_note_JASA.pdf on Image Processing, Vol 3, pp 266-269, 2000.
14. M A Hamstad, ‘An illustrated overview of the use and value 36. D B Percival and A T Walden ‘Wavelet methods for Time Series
of a wavelet transformation to acoustic emission technology’, Analysis’, Cambridge Series in Statistical and Probabilistic
Retrieved February 12, 2003 from the World Wide Web: http: Mathematics, Cambridge Press, 2000, New York, USA, ISBN:
//www.vallen.de/zdownload/pdf/hamstad.pdf 0-521-64068-7.

Insight Vol 45 No 5 May 2003 351

Appendix I – Continuous Wavelet Transform (admissibility); this also implies that the Fourier transform, y(w)
evaluated at zero frequency must be zero:
The bases used in the CWT and DWT, whilst sharing some
similarities, demonstrate a number of differences as will be shown y (w ) w =0 = 0 ..................................[10]
below. The CWT’s forward transform is defined by the inner
product of the signal with the wavelet basis. The inner product, Using knowledge of Fourier techniques it is clear that the frequency
also known as the scalar or dot product, maps the projection of the domain representation of the wavelet is a band pass function
basis function onto the signal. Equation [1] defines a measure of localised in frequency. Transformation of a band pass function
similarity between the signal and the basis function: from the frequency domain to the time domain will always produce
a signal with an oscillatory nature. It can therefore be stated from
W (s, t ) = f (t ), y (s, t ) = Ú f (t )y * ( s, t )dt .............[1] Equations [9] and [10] that the wavelet basis must be oscillatory in
where f(t) is some function of time (the signal), W(s, t) are the order that a signal can be reconstructed using the inverse function
wavelet coefficients, y(s, t) is the wavelet basis, s and t are defined in Equation [7]. A further condition ensuring the wave-like
the scale and translation (time) parameters respectively and * behaviour of the wavelet function is that it should have a zero
represents complex conjugation. As described in section 2 the mean, in other words:
wavelet transform is calculated through scaling and translation •
of the wavelet basis. To achieve scaling of the wavelet function Ú-• y (t )dt = 0 .................................[11]
it is divided by the scale parameter s. However, using this alone
In order that the CWT avoids the windowing problems inherent
is insufficient to produce appropriately proportioned energy in
in Fourier techniques it must be localised in time. This way
the wavelet coefficients. The scaling is therefore normalised
information is not lost through transformation due to leakage.
producing:
Combining this with the results of Equations [9], [10] and [11]
1 leads to the definition of a wavelet namely: a function localised in
y (s, ∑) = y ....................................[2]
s time with zero mean and having band pass characteristics.
Equation [2] restricts the type of basis that can be used for A review of the relevant mathematics relating to the CWT
this type of analysis to those which are square integrable. In other has been given above including the wavelet requirements needed
words: to satisfy decomposition and reconstruction of a signal. The
reader wishing to find out more about the CWT is referred to key
y ΠL2 ( n ) .......................................[3] papers (25, 27, 32), while additional detail of the CWT can be found in
and books(30, 33, 34).

Ú y ( x )dx < • ....................................[4]

2

Appendix II – Discrete Wavelet Transform

In fact due to the energy normalisation implied in Equation [2],
Equation [4] can be restated as: The CWT, while being extremely useful for performing exploratory
analysis, is a very slow algorithm. From a practical stance there
Ú y ( x )dx = 1 ....................................[5]
2
is a need to produce a fast decomposition tool akin to the fast
Fourier transform (FFT) which then provides real-time application
The translation parameter shown in Equation [1] can now
opportunities. The main problem that exists in the discretization of
be introduced to Equation [2]; to ensure that each successive
the CWT is the reconstruction of the signal from the wavelet bases.
translation occurs at the unit time as opposed to unit scale the
The DWT forces some additional requirements on the type of basis
scale parameter is also included giving the scaled and translated
that can be used in the transform process. The CWT is highly
wavelet:
redundant; this redundancy occurs because the CWT analyses
1 Ê t - t ˆ ............................[6] a signal at all scales and translations also requiring an infinite
y (s, t ) =
y
s Ë s ¯ number of analysing wavelets. Appropriate sampling of scale and
Hence [1] can be expressed as, translation and a reduction in the number of wavelets used for
decomposition produces a substantially faster algorithm, the DWT.
1 t - t ˆ .....[7]
W (s, t ) = f (t ), y (s, t ) = Ú f (t )y * Ê dt The first step in the generation of the DWT is the discretization of
s Ë s ¯ the wavelet bases. Recalling Equation [6], a wavelet is expressed
In order to reconstruct the function f(t) from its wavelet by:
coefficients the inverse CWT is defined: 1 Êt -tˆ
y (s, t ) = y
1 1 Êt -tˆ ds .............[8] s Ë s ¯
f (t ) = ÚÚ W (s, t ) s y Ë s ¯ dt s 2
Cy To create a discrete version of Equation [6], both scale and
translation are sampled giving:
where:
y (w ) 1 Ê t - kt 0 s0 j ˆ
Cy = Ú dw .................................[9] y ( j, k ) = yÁ ˜ .................... [12]
w s0 j Ë s0 j ¯

Note that in equation [9] above that y(w) is the frequency where j and k are integers, s0 is a fixed dilation step and t0 is a
domain representation of the wavelet basis. Equation [8] leads to translation factor; t in this instance gives the time steps over which
one of the most important conditions of the CWT wavelet basis; the wavelet is defined. It is clear from Equation [12] that t0 is a
that of admissibility: function of the dilation parameter s0. Typically s0 is selected as
2 (dyadic) and t0 as 1; in this way both scale and translation are
y (w ) dyadic.
Ú w dw < • Using dyadic sampling of scale and time means that only
special types of bases can be used to reconstruct the signal. To this
It is clear from Equations [8] and [9] that in order to recover end Daubechies(25) presents an equivalent admissibility condition
a signal from its wavelet coefficients Cy must be convergent for the discrete version of the wavelet transform vis:

352 Insight Vol 45 No 5 May 2003

 2 2 2 1
A f £ Â f , y j ,k £B f ....................[13] dj = Â y j ,k , y j -1,n d j -1 .......................[19]
s0 2 n

Equation [13] defines a frame with bounds A and B. These Using equations [18] and [19] the reconstruction of the signal
can be calculated from s0 and t0 and describe the accuracy of can be written:
the reconstruction. Equation [13] produces a trade-off between 1 1
( ) ( )
M
accuracy of the reconstruction and the constraints on the basis f (t ) = Â cMf 2 - M t - k + Â Â dif 2 - i t - k ..[20]
2
function: as A and B come closer the reconstruction becomes s0 i =1 s0 2
more accurate and further constrains the selection of the basis. For Equation [20] is the discrete wavelet series decomposition and
the discrete case (A = B) the wavelet coefficients behave like an enables practical implementation of the reconstruction formula.
orthonormal basis. In this case, for reconstruction to take place the Equations [18] to [20] are the essence of decomposition and
wavelet bases must also be orthonormal. reconstruction and result in a transform that is more efficient than
The practical implementation of the DWT makes use of the the FFT.
subband components (section 2). Decomposition is achieved A brief overview of the mathematics required for the DWT has
through filtering of the signal using high pass and low pass filters. been given. For readers wishing to study this area further, reference
This filtering technique is known as multiresolution analysis (MRA) can be made to the following(25, 27, 33, 35 and 36 ) which provide detailed
because the signal is broken down into discrete frequency bands of studies and proofs.
varying resolution (recall the time/frequency resolution issue from
section 2). The DWT introduces the scaling function providing the
low pass filter part of the filter bank. The scaling function f(t), is Nomenclature
not quite as strictly defined as the wavelet function as it does not Notation
need to satisfy admissibility nor need it be oscillatory. However, it ci Discrete wavelet coefficients for ith level
must satisfy the condition of orthonormality. Hence, for the scaling decomposition.
function and the wavelet function the following applies: di Discrete scaling coefficients of ith level decomposition.
D Sampling period.
Ï1 for j = l and k = m
f j ,k (t ), f l ,m (t ) = Ú f j ,k (t ), f * l ,m (t )dt = Ì ..[14] Fc Centre frequency.
Ó0 for j π l and k π m f(t) Continuous valued time function representing a signal.
f(∑) Scaling function.
Ï1 for j = l and k = m j Level (index) for scale in discrete form.
y j ,k (t ), y l ,m (t ) = Ú y j ,k (t ), y * l ,m (t )dt = Ì ..[15]
Ó0 for j π l and k π m k Index for time in discrete form.
The wavelet and scaling function are sampled and translated in L2 Vector space of all square integrable functions.
a dyadic manner as shown in Equation [12]. For implementation ¬n Space of real-valued n dimensional vectors.
these can now be expressed as: s Scale.
y(∑) Wavelet function – both continuous and discrete forms.
1
f j ,k (t ) =
j
( )
◊ f 2 - j t - k .......................[16] t
t
Actual time – continuous.
Constant for time-shifting of wavelet functions.
s0
W(∑) Wavelet coefficients.
1 w Frequency.
y j ,k (t ) = ( )
◊ y 2 - j t - k ......................[17] x Dummy function variable.
s0 j
Let us define the wavelet coefficients produced through low Operations
and high-pass filtering as cn and dn, where n represents the level
∑, ∑ Inner product
of decomposition. The signal is passed through the scaling and
∑ Used to represent an arbitrary functional variable.
wavelet filters giving the first set of decomposition coefficients c1
* Convolution
and d1, these are low and high frequency respectively. The second
n Modulus or absolute value of n.
iteration filters c1 into c2 and d2, the third c2 into c3 and d3 etc. until
the appropriate level of decomposition is achieved ie to level n. Norm or length of a vector.
Using equations [16] and [17] and taking the inner product the ΠContained in.
wavelet coefficients are defined as:
1
cj = Â f j ,k , f j -1,n c j -1 ......................[18]
s0 2 n

THE NEXT IN THE SERIES OF INTERNATIONAL CONFERENCES ON CONDITION MONITORING

2-4 July 2003 - St Catherine’s College, Oxford

For further details contact: Cindy Bailey, BINDT, 1 Spencer Parade, Northampton, UK NN1 5AA
Fax: (0)1604 231489 or respond via our web site: www.bindt.org/comadit

Insight Vol 45 No 5 May 2003 353