Professional Documents
Culture Documents
Wavelets K P Soman PDF
Wavelets K P Soman PDF
,
/-
, ,
Insight in to
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
R... 375.00
0 20 10 by PHI Learning Private Umiled. New 0eIII. AI rights .--ved. No pari 0I1hIs book may be
-----...,-..........- ----.....-..
reproduCad in any klnn. by mimeog...llln Of any 0Ihef ~. -..iIholA pemIisaion in writing from " .
publishef.
ISBN-978-81-203-40S3-4
The .~potI rights 01 IhII book Ire 11951ed IoIIIy w;., !he publisher.
P~bli5h.d by A50h K. Gh-osh , PHI leamlng Privati llmitad. 1.4-97, Connaught C ircus.
New [)aH.ll000t and Printed by Mudrak. 3O-A. Patparganj . Delhi-l10091 .
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Contents
Prrjoce
PrrjllCe /0 the Fir" Edition
A.ckno ...led~mellU
3.2. 1 of ~gnal$ 39
3.2.2 The Founc! Transforml 40
3.2.3 The tli 41
3.3 Propenics of Wavelets In 45
3.4 Conlinuow; versus Discrete Wavelet Transform 45
SWllmQry 47
E:arr:isLL 48
Suggt~!td F... ~r Rtadi~&s 49
.
8.1 Bionho&onali!J: in Vector S~ 127
8.2 Biorthogonal Wavelet Systems 129
S.3 Sigru.l~ UsinS_BiortlKlgonal Wavelet S)'Slem H2
" Conte nts
S"""""ry 202
Conknts xi
Summary 228
1,3~.8~.~1>lB~"~~~;";8~O~f~""";~'~I.~';'~O~:25"
13.8.2 J; _ I 258
13.9 259
162
xii Contents
14.2. 1
14.2.2
14.2.3 Using Lifting 291
[4.2.4 Higher Order Wave let Transfonn 295
[4.3 Lifting Scheme in the Z-domain 296
14.3.1 Desi,n Example I 301
]4.3.2 Example 2: Lifting Haar Wavelet 305
14.4 Mathematical Pttliminaries for Polyphase Factorization 307
14.4. 1 Laurent Polynomial 307
[4.4.2 The Eoclidcan AI,orithm 308
14.4.3 Factoring Wavelet Transform into Liftin~A Z-domain Al!Pf"OICh 314
]4.5 Dealinl_wilh Signal Bolmdir)'_ 1i9
1"-'..1 Circ I,laLCooYOlutiOll_ 319
14.5.2 Padding Policies 310
14-'.3_ lteraLioo_Beh.aviour------111
SU/MIOr)' 31 J
Exercuu 311
.8.
439-443
445-447
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Preface
Wavelet theory has matured and has entered into ils second pllase o f development and evolution
in which practi tioners an: finding newer applications in ever-widening scie ntifi c domains such
115 bio-informatics, computational drug discovery and nano- material simulation. Parallelly. the
theory of wavelel$ got more and more demystified and has become an e~~ryday tool for signal
and image processing. Postgraduate courses in mathematics and physics now ioclude a subject
on wavdet theory either as a separate eJoctive or as part of other related subjects. In many
techn ical universities, w3sclel has been introduced cven at the undergraduate level. In this third
edition of the book we have taken into account thi s increasing popularity and the needs of the
relatively ' young' readers from such wide range of backgrounds.
One of the main additions in this third edition is that we have shown how the ubiquitous
electronic spreadsbcct can be utilized for wavelet ba.o;cd signal and image processing. The theory
behind the algorithm for computing e)lact values of wavelet and scaling funct ion is simplified
and implemented in Microsoft E)lcel as a workshcct function. Onc can now draw the fUlICtion
by writing and dragging an e)lcd formula in a ce ll. Many of the intrigu i ng properties of wavelet
and scaling functions such as orthogonality of integer tran slates, partition of unity and
refinement relation of scaling functions can be easily visualized in spreadsheets. The
accompanying C D contains several worksheets that demonstrate the power of spreadshcct
packages as a computational and visualization 1001.
Recent years have secn heightened interest in ' parametric wavelet filter design' whic h
allows the tuning of wavelet fiheB for various applications. Theory of its design procedures arc
added in respt(;tive chapteB with several e:o;amples.
Another new feature is that parametric and !"IOn-parametric biorthogonal wavelet des ign
are e:o;plained in more detail.
M-band wavelets are finding increasing applications in Communication Engineering as a
tool for multi rate signal procciSing and as signal modulators. So the chapter on M-band wavelet
is eJlpanded to include the more recent and simplified design procedures.
A scparate and elaborate chapter on Multiwavelet theory is added. Muhiwavelet re presents
the highest level of generali:.tation in wavelet theory and provides short filtcrs with most Of all
of the desi rable properties that a filtcr should possess. 1beory of baI:ma:d and interpolating
mu lti wavelets are discussed in detail.
"
xvi Preface
We earnestly hope that this edition will meet the: needs of readen of different academic
backgrounds for their undcrgl1lduate, poslgraduate and research le\'el sludies.
Finally, we acknowledge OUT heanfelt Ihanks here to our ex-students Ms, K. Hemalalha
and T. Ar.nhi who prepared all Ihe worksheets given in the accomp311ying CD.
K.P. SOMAN
K,I, RAMACHANDRAN
N.G. RESMI
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
In the past few years, the study of wa\'c1ets and the exploration of the principles governing thei r
behaviour have brought about sweeping changes in the disciplines of pure and applied
mathemaTics and sciences. One of the most significant development is the realization thai., in
addition 10 the canonical tool of representing a furn:tion by ils Fourier series, then:: is a different
representation more adapted 10 certain problems in data compression. noise removal, pauem
classification and fast scientific computation.
Many books are available on wavelets but most of them are wriuen at such a level Ihat
only research mathematicians can avail them. The purpose of this book is to make wavelets
accessible 10 anyone (for example. graduate and undergraduate students) wilh a modest
background in basic linear algebra and to serve as an introduction for the non-specialist. lbe
level of the applications and the format of this book are such as to make thi s suitable as a
textbook for an introductory course on wavelets.
Chapter I begins with a brief note on the origin of wave lets. mentioning the main early
contributors who laid the foundations of the theory and on the recent developments and
applications. Cbapter 2 introduces the basic concepts in FotJrier series and orients thc reader to
look at everything found in the Fourier kingdom from a geometrical point of view. In
Chapter 3, the focus is on the continuous wavelet transform and its relation with soort time
Fourier transform. Readers who have oot had much exposure to Foorier transforms carlicr may
skip thi s chapler. which is included only for the purpose of completeness.
Chapter 4 places the wavelet theory in a concrete selling usinS Ille Haar scaling and
wavelet function. lbe rcst of the book builds on Illis material. To urKlerstand the conceptS in Ihis
chapter fully, the reader need to have only an understanding of the basic concepts in linear
algebra: addition and multiplication of vectors by scallUl;, linear independence and dependence,
orthogonal bases, basis set. vector spaces and fUllClion spaces and projection of vectorffunction
on to the bases. TIle chapter introduces the concept of nested spaces. which is Ille comer stone
of mulliresolUlion analysis. Dauhechies' wavelets are also introduced. The chapter concludes
wilh a note on the fact that moSt of the wavelets art fractal in nature and that iterative methods
are required to di splay wavelets.
Designing wavelets is traditionally camed out in the Fourier domain. Readers who are IlOl
experts in Fourier analysis usually find the theoretical arguments and terminology used quite
ntl
xviii Pref:>C(: to the l-irst Edition
baffling and tOlally out of Ihe world. This book. the~fore. adopts a lime domain approach 10
designing. lltc: orthogonality and smoothness/regularity constraints are di~ctly mapped on to
constraints on the scaling and wavelet filter coefficients, which can then be solved usi ng solvers
available in Microsoft Excel-a spreads t.eet package or a scientific computillion package like
MAn.AB or MOlhematica.
Engineers onen view signal processing in terms of filtering by appropriate filters. Thus.
Chapter 6 is devoted to eSlablish the relationship between 'signal expansion in terms of wavelet
bas.c:s and the filter bank approach to signal analysis/synthesis.
Chapter 7 discuss.c:s tile theory behind parametric wa,elets in an intuitive way I1Ither than
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
by using rigorous mathematical approach. The chapter also discusses various methods of
plotting scaling and wavelet functions. The focus of Chapter 8 is 011 biortoogonal wavelets
which is relatively a new concept. To drive oome this concept to the readers, biorthogooality
is explained using linear algebl1l. The chapter then goes on to di scuss the design of elementary
S-spline biorthogonaJ wavelets.
Chapter 9 addresses orthogonal wavelet design using the Fourier domai n approach.
Chapter 10 is devoted to the lifting sche me which provides a simple means to design wavelets
with the desil1lble propetties. lltc: chapter also shows how lhe lifting scheme allows faster
implementation of wavelet decomposition/reconstruction.
Chapters 11 to 13 describe applications of wavelets in Image Compression, Signal
Denoising and Computer Graphics. The notations used in Chapter 13 are that used by
~searchers in this patticular area and could be slightly different from those in the rest of the
chapters.
To make the book more useful 10 the readers. we propose 10 post lhe teaching malerial
(mainly PowerPoinl slides for each chapter, IlIld MATLABlExcel demonstralion programs) at
the companion webs-ile of the book: ww w.umritG.tdultrnlpublictdionslwu. vrkts.
We earnestly hope Ihat this book will initiate several persons 10 this exciting and
vigorously growing area. Though we have spared no pains 10 make this book free from
mi slakes, some errors may slill have survived our scrutiny. We gladly welcome all com:ctions.
recommeOOatioos, suggestions aOO constructive criticism from our readers.
K.P. SOMAN
K.I. RAM ACHA NDRA N
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Acknowledgements
First and foremost. we would like 10 express our gratitude 10 Brahmachari Abhayamrita
Chaitanya. who persuaded us to lake this project. and never ceased to lend his encouragement
and support. We thank Dr. P. Venkal Ra ngan. Vice Chancellor of the university and
Dr. K.B.M. Nambudiripad. Dean (Research}-our guiding Slars--who c,mlinunlly showed us
what perfection means and demanded perfection in everything Ihal we did. We would like to
thank, especially, Dr. P. Murali Kri shna, a scienti st at NPOL, Cochin, for his endearing support
during the summer school on 'Wavelets Fractals and Chaos' thal we conducted in 1998. It was
then Ihal we learned wavelets seriously. We take this opportunity 10 thank our research students
C.R. Nitya. Shyam Divakar, Ajilh Peter, Sal1lhana Krishnan and V. Ajay for their help in
simplifying the concepts. G. Sreenivasan and S. Soornj. who helped liS in drnwing the various
figllres in the textbook, dcserve a special thanks. Finally. we express ou r sincere gl3titude to the
editors of PHI Learning.
K.P.SOMAN
K. I. RAMACIiANDRAN
N.G. RESMI
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
INTRODUCTION
Wavelct analysis is a new development in the area of applied mathematics. 'Illey were firsl
introduced in seislTIQlogy to provide a time dimension to seismic analysis that Fourier analysis
lacked. Fourier analysis is ideal for studying stationary data (data whose statistical propcn ics
are invariant over time) bul is nol well suited for studying data with transient events that cann<ll
be statistically predicted from the data's past. Wavelets were: designed with such non-stationary
data in mind, and with their generality and st rong results have quickly become useful 10 a
number of di scipl incs.
Research can be thought of as a continuous growing fractal (see Figure 1.1) which often folds
bad: onlo itself. This folding back definite ly occurred several times in tne wavelet field. Even
though as an organized research topic wavelets is less than two decades old, it arises from a
" oll stellat;on of related eonc:t:pts developed o""r 11. period of ne;orly two " "nturies, repell.ledly
redisrovered by scientists wbo wanted to solve techni"al problems in their various di sciplines.
Signal processors were seeking a way to transmit clear messages o~r te lephone wires. Oil
prospectors wanted a better way to interpret seismic traces. Yet "wa~lets" did 001 become a
oousehold word a mong scientists until the theOf)' was liberated from the di~rse appliCalions in
which it arose and was sy nthesiud into a purely mathenwical thCQf)'. This synthesis, in turn,
opened scientists' eyes to new .ppli Calions. Today, for example, w.""lets are !K)( only the
workhorse in computer imaging and animation; they also are used by the FBI to encode its data
base of 30 million fi ngerprints (see Fi gure 1.2). in the future. scientists may put w.~kt analysis
for diagnosing breast cancer, looking for heart aboonnalities (look at Figu re 1.3) or predicting
the weather.
2 Insighl inlo Wavelets-From Theory 10 Praclice
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
FIG URE 1.2 An FB I-digilized left thumb fingerprint (1lIc: image on the left i. the original ; the one on the
right is reronsuuccr:d from a 16:1 """'pI"'ssion .)
,.,
LO
-
<
0.'
'" 0.' -- Hcallhy
H~an failun:
0.'
0.0 0.'
, 0.' 03 0.'
FIGURE I.J MuhifractaJ 'pec(fum of hcar1 heat oscillations. h is a ""ph of singularity mcasun: versus flllCl&l
dimension. Thi'i spWrum clplum I di ff~n[ kind of iofoonalion Ihill CIIlOOl. be cljlluml by
a ' 1nl"''''')' spectrum. Wn-.:1c1S an: used for muhifraclal spectrum estimation.
The Age of Wavelets 3
Wavelet analysis allows researchers to isolate and manipulate specific types of p<lttems hidden
in masses of data, in much the same way our eyes can pick out the trees in a forest, or our ears
can pick out the flute in a symphony. Otle approach to understanding how wavelets do this is
to stan with the difference between two kinds of sounds-a luning fork and lhe human voice
(see Figures 1.4 and 1.5). Strike a luning fork and you get a pure tone that lasts for a \'ery long
time. [n mathematical theory, such a tone is said to be "localized" in frequency, that is. il
consists of a si ngle note with no higher-frequency overtol1oe.'l. A spoke n word. by cOntl1lSt, laslll
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
for only a second. and thus is "localized'" in lime. It is not localized in frequency because the
word is not a single tone: but a combi nation of many different frequencies.
f\ f\ f\
V V
I'TGURE lA Graphs of the sound WllveS prodoced by tuning fQr\< (top) and thr spol.C1\ wool "grea.ly"
(bottom) iliu!MItc the diffm:l"ICc bet'.ieen tone I<x:alized in freqUfflC)' and one I<x:ali<i in
ti ..... "The tuning fork produces a .imp'" "si"" wa>c".
1/\
Graphs of the sound waves produced by the tuning fork and human voice highlight the
difference, as illUlitrated here. 1be vibrations of the tuning fork trace out what mathematicians
call a si ne wa~. a smoothly undulating curve that, ill theory, could repea l forever. In contrast,
the graph of the word "greasy" contains a series of sharp spikes: there are nO oscillations.
4 Insight into Wavelets-From Theory to Practice
In the nineteenlh century, mathematicians perfected what might be called the tuning fork
version of reality, a theory known as Fourier u1Ullysis. lean Baptiste l oseph Fourier, a French
mathematician, claimed in 1807 that any repeat ing waveform (or periodic function), like the
luning fork sound wave, can be cllpressed as an infinite sum of sine waves and cosi ne waves
of various frequencies. (A cosine wave is a si ne wave shifted forward a quarter cycle.)
A familiar demonstration of Fourie(s thcory occurs in music. When a musician plays a
note, he or she creates an irregularly shaped sound wave. n.e same shape repeats itself for as
long as the musician holds the note. Therefore, according to Fourier, the note can be separa~d
into a sum of sine and cosine waves. The lowest frequency wave is called the fundamental
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
frequency of the note. and the higher fn::quency ones are called o,ertoncs. For example, the
note A. played on a violin or a nute, has a fundamental frequency of 440 cycles per second and
ovenones with frequencies of S8O. 1320. and so on. Even if a violin and a flute arc playing the
same note, they will sound different because their ovenones have different strength s or
"amplitudes". As music synthesizers demonstrated in the 1960s, a very convincing imitation of
a violin or a nute can be obtained by recombining pure sine waves with the appropri ate
amplitudes. That, of course. is ellactly what Fourier predicled back in IS07.
Mathematicians later elltendcd Fourier's idea to non-periodic [unctions (or waves) that
change over time, rather than repeating in the same shape forever. Mosl real-world waves arc
of this type: say, the sou nd of a m(){or that speeds up. slows down. and hiccups now and then.
In images, too. the di stinction between repeating and non-repeating patterns is imponant.
A repeati ns pattern may be seen as a texture or background while a non-repeating one is
picked out by the eye as an object. Pe riodic or repeati ng waves composed of a discrete series
of overtonc:s can be used to repres.c:nt repeating (background) pauerns in an image. Non-periodic
features can be resolved into a much more eomplu spectrum of frequencies, called the I'-ourier
transform, just as sunlight can be separated into a spectrum of colours. The Fourier trans form
portrays the structure of a periodic wave in a much more revealing and conccntraled form than
a traditional graph of a wave would. For elllmple, a rallle in a motor will show up as a peak
at an unusual frequency in the Fourier tnulsform.
Fouri er transforms have been a hit. During the nineteenth century they solved many
problems in physics and enginee ring. This promirtence led scientists and engineers 10 think of
them as the preferred wa y to anal yze phenomena of all kinds. This ubiquity forced a close
examination of the method. As a n::SUII, througOOut the twentieth century, mathematicians,
physicists. and engi neers came to realize a drawback of the Fourier transform: they have trou ble
reproducing transient signals or signals with abrupt changes, such as the spoken word or the rap
of a snare drum. Music synthesizers. as good as they are, still do not match the sound of eoncert
violinists because the playing of a violinist contains transient features--such as the contact of
the bow on the string-that arc poorly imitated by representations based on sine waves.
The principle underlying this problem can be illustrated by what is know n as the
Heise nbe rg Indeterminacy Pri nciple. In 1927, the physicist Werner Heisenberg stated that the
position and the velocity of an object cannot be measured CIlactly at the SIlIl"Ie time even in
theory. In sig nal processing terms, this means it is impossible to know simultaneously the exact
frequency and the exact time of occurrence of this frequellCy in a signal. In order to know its
frequency, the signal must be spread in time or vicc versa. In musical terms. the trade-off means
that any signal with a short duration must have a complicated frequency spectrum made of a
The Age of Wavelets 5
rich variety of sine waves whereas any signal made from a simple combination of a few sine
waves must have a complicated appear:J.nce in the time domain. Thus, we can't e;t;pect to
reproduce the sound of a drum with an orchestra of tuning forks.
Over the course of the twentieth century, scientists in different fields struggled to get around
these limitations, in order to allow re presentations of the data to adapt to the nature of the
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
infomullion. In essence, they wanted 10 capture both Ihe low-resolution forest- the repealing
background sigr"lal- aoo the high-resolution trees-the individual, localized variat ions in the
background. Although the scielllislS wcre trying to solve the problems panicular to their
respective fields. they began to arrh'e at the same conclusion- namely. Ihat Fourier InlIIsforms
themselves were 10 blame. They also arrived at essentially the same solution. Perhaps by
spliuing a signal into components that were not pure sine waves. it would be possible to
condense the information in bolh the time and frequerocy domains. This is the idea that would
ultimotely be known as wa\"l~lcts.
The first entrant in the wavelet derby was 0 Hungarian mathematician nanw:d Alfred Haar,
who introduced in 1909 the furoctions that are now called Haar ",.. ave1ets. These functions
consist si mply of a short positive pulse followed hy a short negative pulse. Although the short
pulses of Haar wavelets are e;t;cellent for teaching wavelet theory, they are less uscful for moSt
applkations because they yield jagged lincs instead of smooth curves. For example. an image
reconstrtlcted with Haar wavelcts looks like a cheap calculator display and a Haar wavelet
reconslrtiction of the sound of a i1me is too homh.
From time to time over the ne~t several decades, other precursors of wavelet theory arose.
In the 1930s, the English mathematicians 10hn Linlewood and R.E.A.C. Paley developed a
method of grouping frequencies by octaves thereby creating a signal that is well localized in
frequency (il$ spectrum lies within one octave) and also relatively well localized in time. In
1946, Dennis Gabor, a British-Hungarian physicist. introduced the Gabor transform. analogous
to the Fourier transform. which separatcs a w""e into "time-frequency packets" or "coherent
states" (see Figure 1.6) that have the gre.1test possible simultaneous localization in both time and
frequency.
And in the 1970s and 19805. the signal processing and image processing communities
introduced their ()wn versions of wavelet analysis. going by such names as "subband coding:'
"quadrature mirror filters" and the "pyramidal algorithm".
While not precisely identical, all of these techniques had similar features. They
decomposed or transformed signals into pieces that could be localized to any time interval and
could also be dilated or contracted to analyze the signal at different scales of rcsolulio n. These
prC<:urso(S of wavelets had one other thing in common: no o ne knew about them beyond
individual specialized communities. But in 1984, wavelet theory finally came into its own.
6 Insighl into Wa~lelS-From Theory 10 Pnctice
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
- -~I't------- --,,------
FIGURE 1.6 Oocomposing lignal inlo t;"",f~l>e1lCy alom~. BOllom of the pictun: ~ twO time fn:qumcy
atom . The signal and the timef""lllCnc:y map i. """"'n "'->ve that.
Morlel 's method wasn't in the books but it seemed to work. On his personal computer. he
could separate a wave into its wavelet components and then reassemble them into the original
wave. But he wasn't satisfied with this empirical proof and began asking other scientists if the
method was mathematically sound .
Morlct found the answer he wanted from Alex Grossmann. a physicist at the Centre de
Pltysiq ue ThOOrique in MlIIOeilles. Grossmann worked with Morlet for a year to confirm that
waves cou ld be re.constructed from their wavelet decomposi tions. In fact. wavelet transforms
turned out to work beller than Fourier transforms because they are much less sensi tive to small
errorn in the computation. An error or an unwise truncation of the Fourier coefficients can turn
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
a smOOlh signal into a jumpy onc or vice versa: wavelets avoid such disastrous consequences.
Morl et and Grossmann's paper. the first to use the word "wavelet", was published in 1984.
Yvc:s Meyer. currently at the Ecole Normale Suptrieure de Cachan. wide ly acknow ledged as
one of the founders o f wavelet theory. heard about lheir work in the fall o f the same year. He
was the first to realize the connection between Morlet 's wavelets and earl ier mathematical
wavelets. such as those in the work of Littlewood and Palcy. (Indeed. Mt!yer has counted 16
separate rediscoveries of the wavelet concept before Morlet and Grossmann' S paper.)
Meyer went on 10 discover a new kind of wavelet. with a mathematical property called
orthogonality that made the wavelet lTan sform as easy to work with and manipu late as a Fourier
transform. CQrthogonality" means that the information captured by one wavelet is completely
independent of the information captured by aoother.) Perhaps most importantly. he became the
IlCXUS of the emerging wavelet commu nity.
In 1986. St6phane Mallat (see Figure 1.7). a former student of Meyer's who was working
on a doctorate in computer vision. linked the theory of wavelets 10 the existing literature on
nGURE 1.7 S\ephane Malla! (CMAP. Ecoic PoIyIcchniq .... 911211 Palais.eau Cedeo.. Fr'anc:e).
sUbband coding and quadrature mirror filters which are the image processing communi ty's
versions of wavelets. The idea of multi resolution ana1ysis--that is. looking at signals at different
scales of resolution-was already fami liar to experts in image processing. MaUat, collaborating
with Meyer. showed that wavelets are implicit in the process of multiresolution analysis.
Thanks to Mallat's work. wavelets became much easier. One cou ld now do a wavelet
analysis without knowing the formula for a mother wavelet. The process was reduced to simple
8 In sigh t into Wawlets-From 'Theory to Practice
operations of avel'llging groups of pixels IOgether and taking their differences, over and over.
'The language of wavelets also became more comfortable to e l~trical e ngineers, who embraced
famil iar teons suc h as "'filters"', "'high frequencies" and "Iow frequencies".
The final great salvo in the wavelet revolution was fired in 1987, when Ingrid Daubechies
(see Figure 1.8), whi le visiting the Courant I nstitUl~ at New York University and later duri ng
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
her appointment al AT&T Bell Laboratories, discovered a woole IICW class of wavelets [sec
Figure l.9(b)] which were not only otthogonal (like Meyer's) but whic h could be implemented
using si mple digital filtering ideas, in fact , using shon digital filters . The new wavelets were
almost as simpl e to program and use as Haar wavelets but they were smooth, without the jumps
of Haar wavelets. Signal processors flOW had a dream tool: a way to breair;: up digital data into
contributions of various scales. Combining Daubechies and Mal1at's ideas. there was a simple.
orthogonal transform that cou ld be rapidly computed on modem digi tal computers.
,, /
~,) ,
,
I ,n
o.
-, " -0.,
1\
,
o , 3
(a) Haar wavelet (b) Daulxdlic. 4- ... p Muna( WlI>"ek(
0'
0.'
0.2
0
-0.2 /
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
- 0.4
.
- 0.6
-5-4 -3 -2 - 10 , , ,
(c) GaIIsi ... ".""lel
The Daubechics wa\'eleLS have surprising features-such as intimate connections with the
theory of fractals. If their graph is viewed under magnification, characteristic jagged wiggles
can be seen, no matter bow strong Ihe magnification is. This exquisite !;omplexity of detail
means, there is no simple formula for these wavelets. They arc ungainly and asymmeuic;
nineletnth-eemury mathematicians would have recoiled from them in hOllOr. But like the
Model-T Ford. they are beautiful because they work. The Daubechies wavelets turn the theory
into a practical tool that can be easily programmed and used by any scientist with a minimum
of mathematical training.
,.
"
,1----"1\ I1
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
,.
,.,
,
I
,
"
-<l.
-. , ,
(b) Mexican Hal ..... ""lel
FIG URE 1.10 Wavelets uoed in conlinuoul wavelet tranJ(O,TU .
Second gene,flIion wlll.'elefS; Here: one enti rely abandons the idea of translation and
dilation, This gives elClra flelCibility which can be used 10 construct wavelets adapted to
im:gular samples.
,
0
I
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
-,
o 2 3
,
nGURE 1.11 Coinel wavelet.
Wavelet ar.alysis is unquestionably one of the beSt achieveme:nts of matllematics in the twentieth
century. Its initial appl ications were mainly in spa rse signal representation and denoising of
signal and images. Probably, it was due to the fact that the theory was hard to understand at
that time. In recent years wavelets have spread in many fundame:ntal sciences other than
mathematics. such as me:dicine, biology, geophysics. physics, mechan ics. economics, etc. This
is e,ident from the ti tle of the books that are appearing.
At present there are more Ihan hundred books on wavelets most of which are on
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
mathematical aspects of wavelet theory. Slowly but steadily many books are appearing in
specific application domains. Following are some: of them.
Wavelets in medicine and biology (by Akram Aldroubi and Michael Unser,
CRC Press, 1996)
This book explores application of wavelets in me:dical imaging and tomography biomedical
signal processing. wavelet based modeling of problems in biology.
Wavelets for sensing technologies (by K. Chall, Cheng Peng. Arlech House
Publishers, 2003)
This reference book focuses on the processing of signals from Synthetic Aperture Radar (SAR).
Specific remOle sensing applications presented in the book include noise and clutter reduction
in SAR images. SAR image compression. texture and boundary enhancement in SAR images,
directional noi se removal and general image processing.
'TlIe Age of Wavelets 13
wave lets were proposed as those wavelets satisfying tlK: linear wave equations in acoustics.
Harmonic wavelets were suggested by Newland. The Newland hannonic wavelets can be
referred to physical family of wavelets for many reasons, btu mai nly be<:ause they are especially
proposed for the analysis of physical problcms on oscillations.
(b) Flexibility ...irh ,ime-frequency ,i/ing: Another advantage of wavelets lies in their
ability to arrange the time-frequency tiling in a manller that minimizes the channel
disturbances. By nex ibly aligning the time.frequency ti ling. the effect of noise and
interference on the signal can be mi nimiZo!d. Wavelet based systems are capable of
ovt'rcoming kTlOwn channe l disturbances at the transmitter. rather than waiting to
deal with them at the receiver. Thus. they can enhance the quality of service (QoS)
of wireless systems.
(c) Semi,i"iry ro chmmei ~ffecl5; The performance of a modulation scheme depends
On the set of waveforms that the carri ers use. 11le wavelet sc heme. therefore. holds
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
the promise of reducing the sensitivi ty of the system 10 harmful channel effects like
inter-sy mbol interference (ISO and inter-carrier interference (ICI).
(d) Vltra widdxmd llpplicmiolls: Impu lse radio. or ultra wideband (UWB) radio. is a
promisi ng new technology for wireless communicruions. Rather than modulating the
information on a carrier. the data is transmiued usi ng a coded series of very narrow
pu lses. carrying informatio n in the time and the frequency domain. Wavelet bases are
good candidates for these pulses.
Figure 1.1 2 shows communication engineering areas where wavelets are fiBding new
applications.
ChanllCI clumlct.erizluion
I. ChanllCl modeting
2. Electrorroagnctic compuuuions and "Jlltnna desis.n
3. SpeW estimation
[nlt"l'f=e
COlnitive rmio 1. Signal denoi.inl
Inl~11i~ nt co;>mmunkation 2. O'lta amval estimation
syst~ms
3. Intcrf"= mitigation
4. ISI. ICI mitil'llion
In summary, we may safely bet that wavelets are here to stay and they have a bright future.
Of course wavelet do nOI solve every difficulty, and must be continually developed and
enriched. We can expecl proliferation of specialized wavelets each dedicated to a panicular type
of problem and an increasingly divcrse spectrum of applications.
When asked 10 justify the value of mathematics, mathematicians often point out that ideas
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
developed 10 solve a pure mathematical problem can lead to unexpected application s years later.
BUI the story of wavelets paints a more complicated and somewhat more interesting picture. In
(his case, specific applicd research led to a new theoretical synthe sis, which in lurn opened
sciemists' eyes to new application s. Perhaps the broader lesson of wavelets is that we should
not view basic and applied sciences as scparate endeavours. Good science requires us to see
both the theoretical forest and the practical trees.
SUMMARY
Though wavelet is an organized research IOpic. it is only two decades old, and has been in use
for a long time in various disciplines under different names. Morlet and Grossman were the first
10 use the word 'wavelet'. Stephan Mallat brought out the relat ion between wavelet
methodology used by Morlel and filter bank theOI)' used in image processing applications. TIle
greatest contribution came from Ingrid Daubec hi es who put tile whole theory on a strong
matherruuical foundation. Wavelets are now emerging as one of the fastest growing field with
appl ication ranging from seismology to astrophysics.
INTRODUCTION
Deemed one of the crowning achievements of the 20th Century, th e Fourier series has
applications Ilia! are far n:aching in varioLls fields of science and mathematics. The Discrete
Fou';". "ansfurm i . one p"I1icular tool widely u&ed in today'~ age of compulers and sol id state
electronics. From graphic equalizers in stereos 10 the moSI advanced scientific sampling
software, the userullless of this mathematical feal is ISlolJllding. Most of the readers of this book
might already know this fact but many of them may nor be knowing tha! Fourier series. as a
mathematical tooJ in analy:ting signals, h:J.S strong connection wilh geometry. In this chapter our
aim is to understand the theory of Fourier series from a geometrical viewpoillL
We assume that you are familial" with vector algebra and co-ordinate geometry. If you are
really comfonablc in using Ihc!iC topics, you can very well understand w hat Fourier series is,
computation and interprelalion of Fourier series coefficients, Fourier transform. discrete Fourier
transform. fast Fourier transform. etc. l1iere exists a strong analogy between what you do in
vector algebra and what you do in signal processing. This analogy will help you to visualize
and give interpretation to the processes and output of the processes that you do on signals.
Development of tllis geometrical mental picture about signal processing is the main aim of this
chapter. Einstein once said "Imagination is more imponant than knowledge". Signal processing,
which many students consider as dry and nonintuitive. demands imagi nation and some form
of abstract thinki ng from the pan of students. In fact. it requires only that. Once )'OIl develop
a conceptual link between geometry (vector algebra) and signal processing, you need nO(
remember any fonnula and every formula that you come across will become transparent to your
mind. Here we will refresh concepts from "Ulor spoce. The detai led exposition of vector space
is given in Appendix A. In this imroductory chapter, we won't try 10 be 100% mathematically
precise in our staternems regarding vector space. Instead our aim is to relate geometry and
Fourier series in an intuitive way.
.
Fourier Serie$ and Geometry 17
2.1.1 Bases
We call i.]. k as bases of space 9{1. By this we mean that any vector in 9{l can be represented
using I. J. k vectors. We also say i. J. k span the space 9{1. Let V = al + b] + et be a vector
in 9{1. Continuously changing scalar a. b. c: we will go on get new vectors in 9{1. We imagine
that. set of all such ,'eClors constitute the vector space 9{1. We express this truth by
2.1.3 Projection
-
Given a vector V how shall we find its normal component vectors or in other words how shall
- - -
we find the scalar coefficients a. b and c'! Wc project V on to bases i , j, le to get a, b and
e. This is the direct resul t of orthogonality of our basi,; vectors. Since V = (JI + bJ + ef,
Y' I = JI + bJ + ef) I =(J
ii J = JI+bJ+ek)J=b >od
ii f = (al+b} +cf)k=c
18 Insight inlo Wa~cLcts-From llIeory to Pmctice
So projecting vcctor 0 on to a base of space 9\J (spanned by a sct onoonormal vectors) gives
the corresponding coefficient (component) of the base. There is a parullcl for this in Fourier
scries analysis.
With these ideas firmly cngral'cd in your mind. you arc ready to understand and visualize
Fourier series.
In vcctor space. we represent a vector in terms of onoogonal unit vectors called bases. Fourier
serics is an extension of this idea for functions (cleclrooics engineers call them as signals). A
function is expressed in terms of a set of orthogonal funclions. To com:13le with ideas in
geometry. we now need to introduce the concept of onhogonality and norm with respect to
functions. Note Ihat. funclion is quantity which varies with respect 10 one or more running
parameter. usually lime and space. Orthogonality of functions depends on multiplication and
integration of functions. Multiplication and integration of function must be toought of as
equivalent to projection of a vector on to another. Here comes the requirement for abstract
thinking. mentioned at the stan of tile chapter.
J-
two functioos have high simi larity. Otherwise. tlley are dissimilar. Magnitude or absolute value
of f t (l )f2(Odl will be high if they \"ary sy nchrOllOusly and zero if they vary perfectly
asynChronously.
For example consider two functions ft(t) = sin t and 12(1) :: cos I for f in the interval
(0,211"). Figure 2.1 shows the two functions and their point wise product. Observe that the net
area under the cur\'C sin I cos f is zero. That is. the llfCa above the I axis is same as tile llfCa
below the I axis. For these two functions, it ean be easily shown that. for I in the range of
multiples of 211". area under the product curve is always zero. TIlerefore. we say that si n I and
cos I are onhogonal in the range (0. 211). Specification of range is very important when yOU say
orthogonality of functions. Because the deciding factor is area under the curve which in turn
Fourier Series and Geometry 19
, ,, , ,,
o.8 , ,, / \( , f
,, ,,
o. ,, ,,, ,
o.
,6
.
V ,, V, j ,, \
\.,
I
depends on the range of inlegralion. Now here observe Ihe varialion of sin t and oos /. Whcn
si n I is increasing, cos I is de<:reasing and vice \'crsa. For range of I in multiples of 2K, the two
functions considered are said to be ortllogonal.
The goometrkal analog of f It (t ) 11 (/) d/ '" 0 is the dot product, ; . J '" O. For the given
,
interval. when!,(t) is projected on to!ft) (proje<:tion here mearu multiplying and integrating).
if the result is tero we say Ihe IWO functions are orthogonal in the given interval. Clearly look
al Ihe analogy. Taking B do! product in ve<:tor space (multiplying corresponding terms and
adding) is equivalenl to multipl ying and inlegrating in Ihe funclion space (point wi~
multiplication and integration). The dOl product is mro;;mum when t'NO vectors are oollinear and
zero when they arc at 90". This analogy will help )'ou to casily visualize and inlerprel various
results in signal processing.
and
-f
Two real fUI>C,j<>"s/,O) a",I/1(1) a", said ,0 be onhonormal if and only if
I> j
oi~ T
I.,
,
0.'
0 V ~
-0.'
-, "- / T
-I>
" ,
1/
"'" "-
03
0 T
-03
-, /'
- 1.5
What is the area under the curve sin 0)1 in the interval (0. 7)'! Obviously i!.ero. Mathematica ll y.
T T
Ju(t) sinllJl f
,.,
dl = l sin(2mfT)dl = 0
We say U(I) = I is orthogonal to sin WI in (0. T ). Do not forget the Il'lar.io n between wand T.
Simi larly we can show that u(t) '" 1 is o nhogonal to cos rut in (0.7).
In general.
T T
J
,.,
u(t) s in nox dl '" 0 al50 f
,.,
Il()) cos nox dt = 0 for all n '" I. 2. ...
Fourier Series and Ge<>mctry 21
Consider cos Wf and sin Wf in (0. n. Figure 2.3 shows the two functions and
COITespondillg fu oction obtained after poin t wise mu ltiplication. Area abo\'e and below of the
2
, ~ ., /-
0 "- T
-, "- /'
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
-2
I'
,
0.'
0 / "".n QW
T
/
""
-0.'
-,
-u
0.6
0.<
0.2
(\ om "" 00$ fUr
0 T
...(1.2
-0.'
-0.6
V V
)'IG URE 2.3 !'Iou of sin WI. 00$ WI and si n (Qf cos QW.
resulting curve is same implyillg that the net area is zero. Th is ill turn imply that cos rut and
n
sin rut in (0. are orthogollal.
Generalizing.
,
Similarly.
,..
f sinllM cos mWfdt = 0 . for all m. n = 1,2.3 .....
~ /1).
22 I"sight inl(l Wavd"i5--From 'Theory tu Pra.c:licc
Alw
,
,
J
.. cosnaN cosmaN dl = 0
Lel us look at the concept from another angle. Consider Ih" sumj(t) for each / in the range
(0. T) given in Eq. (2.3). Choose a set of coefficie nts Do. al ..... and b l. bz, .... and make the
function (or signal) f(l> in the range (0. T). Now go on choosing different set of coeffici"nts 10
gel a different signal. The set of all such possible signals generated is our signal space Ll in
the range (0. 1). Practically all smoothly varyi ng fUllCIions that you can imagine in Ihe interval
(0.1) belongs 10 Ll. Or in other words. practically all smoothly varying functions thal you can
imagine in Ihe illlerYlll (0. T) can be represented by Ihe equation:
panicular base by projecting the vector on to that base. The propeny that lil = I jl = 1.&:1 " I
where Ii 1= .Jt 7 7.]. k are onhogonal implies thal for a
and veclor If e 9\). if
If i '" a. If ] = b, V k =c Ihen V '" ai + b] + ek . [I is a unique representation using Ihe
bases i.]. k. To reilerale the principle, 10 find a panitular coefficie nl , project the vector on 10
the torrespondi ng basis. 1be same mental pittuTe is applicable to Founer series representation
of signals.
In the last section. we have foufld out a set of functions that are onhogonal among
themselves in a given range (0. n. These functions are onhogonal but we haven't thccked
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
whether they form an orthooormal set. In function spa<;e, onhonormalily of two functions 11(1)
and h(l ) in (0. T) requires that
, , ,
f, /i(/) 12(1) dl = o. f, ft( t ) 11 (t ) dl '" 1 and f, f (t) f
2 2 (t} dt '" I
f,' . .
smnM smlltIX d1 = 'f, ['~'J
SIn T . ['~'
Tt'"
smJ d 2"T
The result is a function of T and is IlOI equal to I. For any integer n > 0, the integral is TI2.
So. we say our si ne bases are not normalized.
Similarly it can be easily SMwn that,
Therefore, none of our bases are normalized. So how shall we normaLize our bases? Multiply
all tile sine and cosine bases with J2fT and u(t) '" I with M.
N~.
'f, = =.
...,2/TsinnM...,2/TsmnMdt= T s m ' 'f, . ['''''J
T . [''''J
Tsm 'T
dl "'T ' "2'" I
, ,
f, = = 2f, [2"'' J [2"'
...,2fT cosllM...,2fT cosnM dl = T cos -T- cos -r' J 22" dl = T . T = 1
24 Insight into WavdelS-From 1brory to Practice
, ,
JJIIT
,
../Ifrdt:o ~ fl dt:o ~ ' T=I
,
The set of nonnaliz.ed bases of Fourier series are:
{~, # # #
cos M, sin M, cos 2ar, H sin 2C1X .. , g cos IIflJ1. #
sin IlflJl . . }
All these functions are defined in the interval (0. 1) and ro = 21f1T.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Let us now take a functionf(t) ELl defined in the interval (0.1) and express il using the
nonnalized sine .and cosine bases defined over the same period (0, T). NOIe that the length of
bases are same as the length of the signal. Also note that the furKl.amcntal frequency of the
sinusoidal (also 'oosinusoidal") series is given by flJ:o 21f1T. (This is something which students
tend to ignore or forget and hence the repetition of the fact in many places.)
, f2 " f2 .
f (1) Or! 7r + at V"T cos ar + "I V"T Stn
,I
:0 fa + ..
'H
+ aW T
- cos IlCIX + b -
T 'H' Sill liar + .. . (2.4)
Here 00 . al' ai, .... bj. bi .... are called normalized Fourier series coefficients.
Now. given th3!. such a n:presentat ion is possible. how shall we find out the nonnaliU!d
coefficients. The answer is, projeaj(l) on to the corresponding bases. Suppose we want to find
a;. Then project fit) on to J21T cos IlltW. that is,
a~ = j f(t)H cos mu dt
,
Note the analogy. In vector algebra, to find a component coefficient, we project the vector on
to the corresponding unil vector (base). Let us verify the troth through mathematical
microscope. Multiplying both sides of Eq. (2.4) by J21T cos Ilwt and integrating over (0. T).
we obtain
, ,
J, f(t )hlT COSIIM' dt :0 aO ~.
,T
J, JUT cos mm dt
,
+ a; JJ2fT cos wtJ2IT cos df 1l{J}l
,
,
+ hi f J21T sin OX J21T cos lIax dt + ...
,
Fourier Series and Geometry 2S
,
J
+ a~ ../2/T COS nM ../2/T COS nM dr
,
+ b~ J ../21T sin nM ../21T sin nM dr + ...
=
, /I'
,
Thus. J10) since J
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
normalized base.
What about <lQ1 ProjcctJtt) on to the base w(r) = .JlfT defined over (0. 7). giving
,
~ = Jf(l)M dr
Well. we ullderstood how to get normalized coefficients in Fourier representation of signals. To
get a particular coefficient simply project 10) on to the com:spondi ng normalized base.
Projection in function space meaRS poiRt wise multiplication and integration of tile functions
concerned over the defined period.
After obtaining the IIOnnalized coefficients. we substitute in Eq. (2.4) 10 get Founer series
representation of the signal. Let us simplify Eq. (2.4) to avoid scalars ..J2IT and .JIlT.
IfJT = I/o
Let <lQ so that
, ,
<lo = .JlfT <lQ =.JlfT J./liT I{t) dl =~ Jl(t) dl
(Note that we substituted the Cltpression fo r <lQ to get /10-)
,
a.=: f f(t) cos nar dl. n = 1,2, 3, .... (2.7)
,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
N~,
(2.9)
Note that the formulation involvcs complcx cxponcntials and the concept of negative
angular frequcncy -ru. 1ltere is nothing imaginary in any signa l and there is no negative
frequency. Complex exponcntials fac ilitate easy manipulation of multiplication and integration
involvcd in thc calculation of Fouricr serics coefficicnts than multiplication and integration
using plain sine and cosine fUlI(:tions. Let U$ filS! try to visualize t}oH, when: fI) = 2trrr and T
is the period of o ur signal.
Fouricr Series and Geometry 27
By Euler's fannula,
~OA = cos (J)l + j sin (J)l
We can th iok of this fUlICtion as 11. 'bundle' of sine and cosine fUlICtion kept togethcr side by
side such that they do not 'mix' with each other. The fUlICtion of j cao be thought of as a
separator indicating that the two parts. I'iz. cos 001 and sin oot are orthogonal.
The set of bases. (I. ~. t- iu ... " ejoOA. e-jolll ... ,,) are orthogonal over the interval [0, n,
where 00 = 21rIT and T is the duration of the signal which we are representing using the bases.
However. there is a twist with respect to interpretation of orthogonality and projection of a
fUlICtion on to a complex function. To make you prepared fOf" that twist and to lICCept that twist,
let us consider If.: aT + bJ + ek. an element of ')\1. n.e nonn of this vector (or magnitude) is
obtained by projecting V on to itself and then taking the square root of resulting quantity. That
;,
to find the magn itude. we have 10 multiply l with its complex conjugate l and then take the
square root. In case of real function, say f(t), we find square of its nonn by
Jf( t ) f(t) df
In case our function is compleJO, say Z(f), its square nonn is given by
Jl(t) l(f) df
This has implication in the formula for projection of a fUl\Clion on to a complex function.
We must multiply the fil'$t fu nction with complex conjugate of the second funct\on a~ then
integrate. With this twist in mind, we will prove that the set of functions u(t). t"", e-JfM . . . . .
ej<lu. t - j<lu, ... ) are orthogonal. We denote the projection of ft") on to h(f) as (fl(f), fit.
CtI$e J: ep. .., and e J...,' where 11 and m are either positive or negative integers with m ~ 11.
, , ,
(~. ~) = J~M t!/ow dr = f ~.-. ) .. Jeil dt = .-",){2JffT)' dt ". 0
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
0 0 0
Cases l. 2 and 3 prove that the functions are orthogona!. What about orthononnality of
comple" e)[ponential functions?
Since.
, ,
r) = Jr"'M t ftttU dt = f
(l!'*",
...
] dl=T
>od
,
(U(I ). U(I = f Idl = T. the functions are not orthooonnal
The functions though orthogonal do not salisfy the requirement of orthononnality. So we
multiply the functions with a scalar to make them onhOflOrmal WllOng themselves.
For unit function and comple" uponentials, scaling constant is .fliT. Therefore,
orthonormal bases are:
We denote the signal space generated by these bases as L 2. Now. any signal which is an element
of Ll can be written as:
(2.10)
To get a particular coefficient, we project f(l} on to the corresponding base.
,
I.e., Col = ff(I).jlfT u(t) dr (2.11)
,
C; = Jf(I ).JIiT r-;- dr (2.12)
I'ouricr Series and GeomcU'y 29
,
C~. = f 1(1) JilT t i- dl (2. 13)
"
This resull follows from the OJIhonormality of bases used in (2.10).
Now, we simplify the expn:ssion given in Eq. (2. 10) by introducing new constants.
Let Co = coJIiT
Since
,
Co = f l(lr/lfi 1.1(1) dt
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Therefore,
, " , ,
Co = J IlT f f(t).fliT 1.1(1) dr = ~ f f(t ) uet) dr = ~ f f(t ) dt (2.\4)
,,' "
,
C. = C:. JUT = f
~ l(rY- dr (2.16)
"
Thus, we oblain complex fourier series in slandard form:
(2.17)
SUMMARY
It is no undcrstatementthat the Founer senes especially discrete Founer transform is considered
one of the crowning achievements of the 20th Century. The theory of Founer series has strong
connection with geometry or at lcast we can understand the Founer transform thoory from a
geometrical viewpoint. We h3d the honour and a privilege to bring it to the reader in th is user
friendly, low fat and no cholesterol form. In the forthcoming chopters on wove lets, this
geometrical viewpoint will help you to digest the theory in a beller way. It will surely prevent
intellectual diarrhoea.
30 .. Insight into Wave lets From Theory to Practice:
EXERCISES
2.1 Consider the interval [- I, IJ and the basis functions (1. , . , 2, ,1, ... ) for L 2!_1, I] .
,
The inner prod uct on this vector space is defined as <to g) = J/(1) 8(1) dl . Wc can
-,
show that Ihese functions do nOl fonn an onoonormal basis. Given a finite or
countably infini te set of linearly independent vectors Xj, we can construct an
orthonormal set )'j with the same span as X j as follows:
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
.. Therefore.
2.2 Project the function j(x) : x on to the space spanned by ji(x), 1x), 12x ),
!l'(2x - I)e L2[O, I] where
1. O SxS I/2
fI(X) :
1o. otherwi se
I OS x S l12
If(x): - 1 112 SxS I
(
o
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
otherwise
2.3 Show that the set of functions /", (x ) '" H sin mx . m = l. 2, ... , is an or1hononnal
system in L2[O, If].
2.' Give n that
112 112 1/.J2 0
In In -11../2 0
H. :
112 - In 0 11../2
,n - 112 0 - 1/./2
9
7
is an orthononnal basis for :It. and l : [l IE '" e :lt 4 where E is the standard
,
3
[Hinl: Acoording 10 Plancherei's formula I If = L. 1c. 12, where the C,.s are
""
complex Fourier series coefficients of the function I(/). U/li2 in the given problem
is 2~ J" 2
If(t)1 dt I
-"
2.6 Find the Fourier series of
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
fH
I(x)=l n
for-nSx<O
forOSx<n
INTRODUCTION
The aperiodic, noisy. intenninent, transient sig nals are the type of signals for which wavelet
transforms are panicularly useful. Wavelets lIave special ability to examine signals simuh-
aneousJy in both lime and frequency. This has resulted in the development of a variety of
wavelet based methods for signal manipulation and interrogation. Current applications of
wavelet include climate analysis, financ ial lime series analysis. hcan monitoring, condition
monitoring of rotating machinery, seismic signat denoising, deooising of astronomical images,
crack surface c haractcril.ation, characterization of turbulent intermittcncy. audio and video
oompression, compression of medical and thump impression records, fasl solution of partial
differential eq uations, compu ter graphics and so on. Some of these applicat ions req uire
(:(lntinuQUs wavelet transform, which we will explon: in this chapter and find out how it differs
from classical mc:thods Ihat deals with aperiodic lJOisy signals.
Wavelet means 'small wave' . So wavelet analysis is about analyzing signal with short duration
finite energy functions.1bey transform the signal under investigation into another representation
which presents the signal in a more useful form. This tnlnsformation of tbe signal is called
wa velet transfOl'm . Unlike Fourier tnmsfonn, we have a variety of wavelets that ~ used for
signal analysis. Choice of a particular wavelet depends on the type of application in hand.
Figures 3.1 to 3.4 show examples of some real and complex wavelets.
34 I
""
"gh t .,"IQ W m le ~~
F """" " p,,,,'"
1
1
0.2
0.2
0 "
0
..... -<1.2
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
-<1.2 -1
- 20 o 20 - 20 - 10 o 10
FIGURE J.l Real and imaginary pans Qf Shan wavdet
1.2,--_-_--_-_-_-_--_-,
0.8
0'
01--_,
-<1.'
-3 -2 -1 o 3
FlGURE).2 Quintic Splinc wavelet
1.0
1.0
0.'
0.'
o0 t-~
o.ot------
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
-<I.'
-1.0 ':----:----;--:---0 - 1.0 ':-----";----'-;----;:------;
-4-20 4
-4-2024
, ,
FIGURE 3.4 Real and imaginary potU of Morlet wavelet.
We manipulate wavelet in two ways. The first one is translation. We change the central
posi tion of the wavelet along the time axis. The second one is scaling. Figures 3.5 and 3.6 show
translated and scaled versions of wavelets.
,", '" ,
,, ", ,
~, ;.
,'.''"'~, "'.,
FIGURE J.5 Translation (change in position) of wavelets.
f~ucnq
(Scale)(Lcyd)
,
vv
TIme shin
Figure: 3.7 shows a schematic of the wavelet transform which basically quantifies the local
match ing of the wavelet with the signal. If the wavelet matches the shape of the signal well at
a specific scale and location, as it happens to do in the top plO( of the Figure 3.7, then a large
transfonn value is obtained. If, however, the wavelet and signal do not correlate well, a low
value of transform is obtained. The transform value is then ploned in the two-dimensional
transfonn plane shown at the bottom of the Figure: 3.7. (indicated by a dot). The transfonn is
computed at various locations of the signal and for various scales of the wavelet, thus filling
up the transform plane. If the process is done in a smooth and continuous fash ion (i.e., if scale
and position is varied very smoothly) then the transform is called conti n uous wavelet
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
transrOnR. If the scale and position an: changed in discrete steps, the transfonn is called
diM: rde wavelet transrorm.
transform plO!
Wavelet
_ Cu=n' ",ale (Two-dlmeru'onol)
"", I ~---
-----------------------~
: Higher lhe coefficient more
Position - - -_.
FIGURE 3.7 n.e signal. wovelel and lnUI,form.
I
~. (,). JI"I ~ ('-b)
-;;-
Continuous Wa" elel and Shon Time Foorier TfWlSform 37
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Ti mel)
HGURE J.lI SiglW iIId iu ""iI,(]tl tJillSfl1l1ll lipc.;lJUm.
where b is location parameter and 11 is scal ing parameter. For the function to be a wave let. it
should be time li mited. For a given scaling parameter a. we tran slate the wave let by varyi ng the
parameter b.
We define wavelet tran sform as:
According to Eq. (3.1). for every (Cl. b) ....e have a wave let trans form coeffic ient.
rePf"CSenting how much the scaled wavelet is s imilar to the function at location I = (bla ).
In the following section (Section 3.2) we will explore the classi cal time-frequency
representation of signals. and associmed uncenainties in frequency and time. 1llen we shall
compare the same with that of wavelet transfOl"ms.
If fELt. the integral makes sense fOl" every value of w and f is a continuous. bounded
function which goes to zero at infi nity (thi s last fact is called Riemann -Lebesgue Lemma).
38 Insight into Wavelets- From Theory to Practice
The Fourier transform is also defined for [E:' L2. In this case, the integral may oot be
defined in the usual sense. One way to define it is:
" = I
c:= L Il' II f '
In this case j is also in Ll.
few)
,,2K R -J ...
-. [(I)e-"OJ dl
I(r) I
= r::-=
,2. - I(wje"'" dw.
where the integral may need to be imerpreted as a limit of finite integrals, as before. In general,
every locally integrable function [has a Fourier transform but j may not be a fu nction any
1IlQI"C, rather a generolized [unetlort or dlslrlm.lion. We will assume from oow onwards that all
functions have suitable imegrability properties so that all Fourier transforms (and other
operations that show up) are well defined. Some propenies of the Fourier U"lUls form are:
The Fouricr transform preserves L 2 norms and inner prod ucts (thi s is called the
Parseval-Planchen:1 Theon:m). Thus. if f, g E Ll then
(I,g) = <i,g)
11Je Fourier tran sform turns convolutions imo products and vice vema. 1lte convolution
of f, g is defined as:
J
(f. g)(I) = I(y) g(t - y) dt = J I(t - y) g(y) dy
We fi nd
(f. g)" (w) = J2K j(M)g(M)
The Fourier transform turns translatioo into modulation and vice versa. The lrons/mion
of I by a E 9t is defined as:
T.[(t) = I(t - a )
. .
(EJ) " (m) '" f(m - a) '" T.f(I)
The Fourier trlUlsform turns di lation inlO inverse dilation. The dilOliQn of f by SE 9f
is given by
-112
I 1 f(tls)
D.f(t ) '" s
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
The factor in front is choscn so thal ID.f l If b.The Fountr transfonn relationship
l
'"
IS:
1/2 " "
I 1 f(sm) '" o",.f(m)
( D.f)'" (m) '" s
The Fourier transform turns differentiation into multiplication by iro and vice ve~
.
f(m ) '" ~'f -JO)t-"'" dl (3.2)
,2. -
is used to decomposc f into its frequcocy components. The inversion formula is o;pressed as:
(3.3)
transform near time I, and frequency ro i$ interpn:tcd as: [he signal! contai ns a large componenl
willl frequency W near lime I. lllere is a lot more theory [0 botll of these transforms IlIan we
will cover in [his chapter but our main interest lies elsewhere.
----'---+_L---::+---'------+ ,
Sigma
.' IGURE 3.9 Char.lCtm>lic function of W .. XI-l.It
(ii) W", (J +cos 21)12 for I ( ~.R"12. xl2 j, IT == J/.J3 (see Figure 3. 10)
Sigma
(iii) w" _'_ t-r'12 (Gabor wi lldow). 0'" I (see Figure 3.11)
J2Jc
CQnlinUOlU Wa""kt and ShQrt Time F<>urier Transfmm 41
Sigma
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
';"e-.. n
FIG URE 3.11 PlO! Qf window fUI>Ction W ~
,2, .
The WfT wi th window W of f is defined as:
Ifwf(a.b)
-
= ,j~lf L !(r)to-;bl W (t a) dl (3.4)
\// y -
Note how the time resol ution u (rel3led to the window width) is C()nstanl. independent of
the frequency.
(3.5)
42 In si):hl intQ Wavekts-Frorn TheQry IQ Practice
(3.6)
Remark: If! E Lllhen !2/I!f is a probability distribution. i.e. a non-negali\e function with
integral L. IJf and CIf are simply the mean and stlllldard deviation of this distribution in the
statistical sense.
IiJ measures where! is localized in time and CIf measures how spread OUI (or uncertain)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
this time measurement is. Pt localizes the frequency and Cl; measures the uncertainty in
frequency.
H,~
(3.7)
where
!" (w) = ~
I
-J .
!(t)e-" dr
,2. -
"d
(J.8)
(3.9)
If a function is localized in time, it must be spread out in the frequency domain and vice
versa (see Figure 3. 13). The optimal value of 112 is achieved if and only if f is a Gaussian
distribution. To visualize thi s, consi(je r a hypothetical function F(t, w) over the time-frequency
plane. F(I, w) represents the component of! with frequency wat time /.
The uneertai nly princi ple says that it makes no sense to try to assign point-wise values to
F(I, w). All we can do is to assign meaning to averages of F over rectangles of area atleasl 2.
(The signal is localized in time to [,u - Cl, IJ + a:l and likewise for freq uency, so the rectangle
has area (2C1p x (20).
The uncertainty principle applies 10 both WfT and CWT but in di ffere nt ways. Let 1J... 1lIld
Cl... represent 'localization ' (mean) IlIld 'spread' of the wirKiow function W(I) in time domain
which are formally defined as:
1-1... = 1 2
-fIjW(t)11dl (3. 10)
IWO)I _
Conlinuous W Ueltl and SIKH1 Time Foune, T r.msfonn 43
:;>
D Good frequency resolution
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
(3.1 1)
Also Itt A.. and 0-.. represent F014,iu lrons/orms of 'locaJiution' (mean) and 'spread' of
the window funclion WCt). Then the inner product (f, W) contains information on / in
[.uoo - u.".u .. + u .. J. Since <I, W) '" tj, IV) , it also contains (nfoonation on /(0) in
[.it.. - o-.. ,.it.. + it.. ]. Thus, (J, IV) represents the frequeno:ies between Pw - o-w and .it.. + 0-..
thal are present between time p .. - u.. and .u.. + u"'.
For WfT, the test funct ion is the shifted and modulated window .
which is localized in time near .u.. + a with uno:ertainty a.".. Its Fourier transfonn is
which is localized near .it.. + b .... ith unCtnainly U...
Here E and T are respectively modulation and translation operators.
(3.12)
For a WFf with fixed window WCt), the time resolution is fixed at u..' the frequency
resolution is fixed at 0-,... We can shift the window around in both time and frequtncy but the
uncenainty box always has the same shape. n.e shape can only be changed by changing the
.... indow W(I). Refer Figure 3.14.
44 Insight int o Wave lets From Theory to Practice
D
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Time axi.
Time and frequcocies are resolved equally well (or equally bad). A discrete WFT (with equally
spaced lime and freque ncy samples) gives a unifonn tiling as in Figure 3.15.
,
I' IGURE 3.15 lime.fre<j""'''''Y Lilinl of windowc:d Fouricr t.,,".form.
(3. 13)
AI high frequency:
Good lime re""luliOl1
Bad freqllCOC)' r~lulion
[
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
TIme aJIis
.'IGUIlE .3.1' f'lmiblc limc-frcquclI(Y tHings in the case of continuow wa ...det transform.
A wavelet 1f(1) is si mply a function of time t that obeyS a basic rule. known as the wavelet
admissibi lity condition:
number of translations at each scale. This is the discrele wavelet transform , Although discrete
wavelet transform may be derived without referring 10 CWT, we may view it as a
"di scretiz.alion' of the CWT through sampling specific wavelet coefficients. A criliCllI sampling
of the CWT
f
lV(a. b) : ,/(1) Jlal
1 ('-b)
VI - , - d,
is obtained via a ,. 2-i, where j and k are imegers representing the set of discrete translations
and discrete dilat ions. Upon th is substitution
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
f
, /(1) JI1 1VI ('-b
(J
)
- , - d,
becomes
o t
H GU RE 3.17
.-+
Critical $ampling of the ti~.frequency (~alc) plane by discretizing the cwr parameters ..i.
a '" rJ and b W. "The CWT i. defined al 011 points in the plan<: and CQm;$pondo to
redundant rcprescnLation of the information JlfUCnt in the function. A... inerea<e5. the number
of codficicn15 s.;ImplCld ""''''" lime gel doubled. By using only the point. provided, the
minimum ,.,mbcr of w.""lct coe fficicnt< an: uocd 50 th~t the funetion may be: perfectly
reconstructed.
Continuous Wavelet and Soon TIme Foorier Transfonn 47
CWT. we find wavelet coefficients for every (a. b) combination whereas in discrete wa\'elet
transform. we find wavelet coefficients only at very few points denoted by the dOls and the
wavelets that follow thesc values are gi\'en by
(3.18)
These wavelets for all integers j and k produce an orthogonal basis. We call 1I'(l.o(t) = Vf(t )
as lIIolhtr WQ\del. Other wavelets an: produced by translation and dilation of the mother
wavelet. More de tails about the discrete wavelet transform will be dealt in forthcoming
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
chapters.
The discrete wavelet transform results in a timefrequency tiling as shown in Figure 3.18.
_ F""I""ncy ui.
Time uil
In many applications (such as speech analysis). high frequencies are present very briefly
at the onset of a sou nd while lo .... er frequencies are present later for longer periods. To resolve
all Ihese frequencies well. the WIT has to be applied several times. with windows of varying
widths. The CWT and DWT resolve all frequencies si multaneously. localized in time 10 a level
proportional to tIleir wavelength. Thm is what tile wavelet people always claim.
SUMMARY
Fourier analysis is a global scheme. While we are able to analyze the frequencies Ihal make up
a signal (and do so quickly). the Itx:al properties of the signal cannOl be easily detected from
the Fourier coefficients. For example, if our signal was a recording of a drum being struck once
and silence otherwise. we could look allhe Fourier coefficienls 10 ana lyze Ihe frequencies. 001
withOllt significant effortS. wc wOllld not be able 10 tell when the beat of the drum happened.
It is this phenomenon that is known as lIcisenberg's Un~rlainly Princi ple. It says that 11
signal cannot be simultaneously localized in time and frequency. Wavelets. are an allempt to
48 Insight inlO Wavelets-From 1beory 10 Pracli
EXERCISES
3.1 Show Ihal for f(x) given by f(x) = /!.-l. u~oo}=_1 and aJ = u.
4,
3.2 Find few) for
graphical display
.
f xc-<I(4-..... dr
of the
= -2i f xc- ax sin wx dr J
magni tude
,
of the STfT, V'w I(a, b) :.
-
.J~1f L f(l)e - ib< W (I a) dt, is called the spectrogram of the signal. Prove th al the
3.6 Compute lhe time-bandwidth prodUCI for the spectrogl1l.m of x(l) = c. . , computed
with the window function Wet). "The answer should be in tenns of Wet) and ~
C"",;nuou. w"...... and Shot! Time FoIIri... T"""f""" 49
Hlawatsch. Franz, Time-Jl1'quf/lC)' Analysis will SylllhtsiJ of Linear Signal SpaceJ. Kluwer
Academic Publishers. Boston. 1998.
Qill1l. S. and D. Ch.n. loo'", n",,!requt1!C) Arwiys;$- Mtlhods ,,,,<1 Appiicaliom. Prenticc Hal!.
>9"
Strang. G. and T.Q. NgU)'CH. Wa,,.,lels and Filtu Banks. Revi sed Edition. Welle.ley-Cambridg.
Prcss. Welles!i:y. MA. 1998.
Vaidyanalhan. P.P.. Mulrjrate Systems and Fil,U /hUlks. Prenliee Hall. Eztglewood Cliff NI.
1993.
Venerli, M. and I . Koncevie. Wa"e1m a"d 5ubband Coding. Prentice Hall. F..ng1ewood Cliff
NI. 1995.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
INTRODUCTION
In the las! c hapte r. we came across continuou s wavelet tran sform and di Sl;:rete wavelet
transform was viewed as sampled version of the continuous parameter wavelet lransform.
However theory of discrete wavelet uansform can be stud ied independe nt of its continuous
counte rpart and in thi s chapler we will dc'-clop the theory without any dependence on
continuous wavelet trnnsform.
Wavelets are a special kind of fUlICli ons which exhibits oscillatory behaviour for a short
period of time and then die out. Unlike Fou rier series. in wavelets. we use a single function and
its dilations am! translations 10 generate a set of o rthonormal basis fun.ctions 10 represent a
signal. Number o f such functions are infinile and we can choose onc that suits to our
application. Unfonunalc:ly. most of the wavelets used in di screte wavelet tran sform are fraclal
in nature. The)' are expressed in terms of a recum:nce relation !iO that to sce them we must do
several iterations. For a newcomer in this fie ld. it is the greatest hurdle. Fortunatel), we have
111'0 special fu nctions called Haar wavelet function and scali ng fun cti on which have explicit
express ions. To underst:md wavelet mathematics with geometrical insight. Haar wavelet s),stem
is ou r onl), hope. Scaling functions and wa velet funct io ns are twins. Corresponding to a
wavelet functi on. there is a scali ng function. To understand this relation ship. we must
understand the concept of a set of Iusled spaces and their complementary spaces. which are
di scussed in fonhcoming section s.
In discrete wavelet transform. we ha ve to deal with basically two sets of functi ons-scaling
functiOlls and wavelet functions. Understanding the relation between these two functions b the
main ai m of this chapter.
DiSCR:1C Wavelet Transform S I
Consider Hoar scali ng function r) defined in Eq. (4.1) and shown in Figure 4.l.
I 0:51:51
.r) '" ( (4.1)
o else wllere
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
o
FIG URE 4..1{a) H""" sealina: function.
~,) t
I ------------- .'
,
- ')
------------ .' .." ,
- ----_.
3
n GURE 4.l{b) Translalion. of Haar 5Calina: fuoclion ;(r).
-,
energy. ThaI is.
-f 11(1)1
The domain of tile function is [0. 1[. Note tllat tile function is lime limited and Ilave finite
2
dr e~ists and is finite. Firsl we will make a set of functions that are
-
trnns13tes of !/I(r). The n we w ill s how thal. these functions 11' o n hogona l and ean form .. basi,
set JUSI like sine and cosines functions formed a basis set for Foo rier series analysis.
again. According 10 Eq. (4. 1), the quantity inside the bracket 'O'of, may vary only in the
dosed interval [0, I J [0 get a non-zero value. Once the functiOn is defined, this remain same
irrespective of what you write inside the bracket. For example, once we define ?(t) as in
Eq. (4.1). then in ,Cr - I), the quantity inside the brlK:ket, that is, 1 - I may vary in the runge
[0. I] to have non-zero values for 1/1. Therefore 1 jn ? CI - 1) may vary between 1 and 2, function
value is I and for all other values of I, Ihe function value is zero,
For economy of thought. if we want to find out where does ?CI - A:) have non-zero value
for the function given by Eq. (4. 1). we set (t - A:) '" 0 to get I '" A: and set (t - A:) '" I 10 get
1 '" A: + I. So the function I - A:) has oon-zero value for I in the range fA:, A: + 11.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
This argumcnt can be utilized to find, say. wllcre 9(101) ex ists jf I/t(r) is as defined in
Eq. (4.1). We set 101'" O. \0 obtain 1 '" 0, and set 101 '" I. [0 get I '" 1110. llw::refore. the function
, (101) has non-zero \'alue in the interval [0. 1/ 10]. Thus, \XIOI) is a scaled version of 9(1).
What about 101 - A:)1 We sct lOt - A: '" O. to get I '" kilO. and set 101 - A: '" I. to get
1 '" CA: + 1)1 10. Therefore. the functio n 10, - .\:) has oon-zero value in the interval [kilO.
CA: + 1)110]. Thus. 101 - A:) is a scaled and translated ve rsion of .1).
Consider
-f P(I);(I - 1) dt. For HaM, 9 (1) 9(1 - I) is identically zero as shown in Figure 4.2.
-
Seeing the function. we are able to tell the location where the function existS. It can be easily
seen that point-wise multiplication of these two functio ns resuhs in uro for every value of I.
Therefore, Ihe integral is zero. Here it is quite obvious, since. there is no ovcrlap bet".,een the
two functions in the sense that wherever 4l(t) has non-zero value rjl(t - I) is zero.
,
0
~r _ I)
'1
0 2
.,
1
0
~I ) \l{1 - l)
2 "
FIG URE 4..2 Ortbogonalily of lnmslald of ~I).
Discrete Waveld Transform S3
When rn and n are equal. the functions II.J'e identical and perfectly overlap so that the area 15
equal 10
-f -,
J f
-
4(1-rn)4'<t-mldl '"
-. ;(1) ;<t)dl= lldl:: 1
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Equation (4.2) implies that the functions 9(1 - nI) and 9(, - n) are onhooormal.
, , , , , , ,, ,, ,, ,, ,, ,,
6 , _____
~ , _ __ _ ,_ ____ , ____ ..,_____ ,_____ , ____ ..,_____ ,_____
~ ~
,, ,, ,, ,, ,,, ,, ,, ,, ,,
,, ,, ,, , , ,, ,, ,,
, ,,
L ____ ~
,,
,_____
,,
,
J _____ _____ J _____
,,
L ____
,,
, ~
,
,,
J,
_____
,,
,
,,
, ,,, ,,, ft, ,,, ,,
, ,, ,,
1-----
,, ----- ---- , ----+----, ,
,, , ,, , ,, , ,,
,
J
,,, ,, ---- ----- -----r---- , , , , ,, , ----~,,
,
,, ,, ,,, ,,, ,,, , ,
, , ,, ,,
1 ,
t-----,.----
, , , , , ,,
-----,.----,-----',, -----r----'
, ,
I ,f-~Il
, --- -~---~- - ---~- ----~-- - ~----} -- --~- -- ~ ------
,
,
,,
,
,,
,
,, ,,
, . ,,
,
,,
'~f..tl' ;(I-21 ' ;(1-3) , f{f.4), f{f..S) ' .(1-6)' ~1_7)'
,
ol _____ ____ J_____ ! _____ ____ _____ ____
~ ~ ~ ~ ~
,, ,
, ____ ,,_____ _
~
,,: '1'2'3'4:S:6:7:S
, , , .,, ,, ,, ,,
,, ,,
,
,, ,, ..
,,
,
,,
,
,,
,
,,
,
- I ~-----~----~----- t -----~----~-----t-----~----~------
FIG URE 4.3 A representalive Jignal in V... spanlltd by Haar scalinS functiollJ.
54 Insight into Wavelets-From Theory to Prac:tia
Let Vo be the space spanned by the $Ct of bases [ ... , 4(1 -+ 1), 4(1), 4(1 - 1), .. ,1. Wc: denote
this as:
,
Vo = Span{;<1 k)} (4.3)
Consider a. function 1(1) = L- o~(J(t -1:) where a.s are real numben (scalars) which we
call as coefficients of tjl(1 - 1:)s. For, one set of a.s, we have one panicular sigoal. But assume
that we are continuously changing 0li to generate continuously new functions or signals. 11le
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
set of all such signals cOlIStitute the function space V(Io Note that, all type5 of signals cannot be
element of Vo- For a sigoal 1(1) 10 he an element of VI). we must be able to express 1(1) using
the bases of Vu. What is the speciality of a signal in Vo which is spanned by the Haar scaling
functions? 'They are piecewise constant in the unit interval. Figure 4.3 shows such a signal.
Here 1(1) is given by
,,
~ f{r)
.-f-::::::.J-:
.../ Ur)
, ,
ko-----T----~,.---~)-----f
.----~------+ '
FIGURE 4.4 Signah which arc no! in Vo (spanned by Haar lCaIing functions).
No. Both .ltl) and b(l) cannot he represented by the bases of Vo- The signal in Vo space
must he piecewise constant in each unit interval.
0;..,,,,,< Wavele, T11lIUf<>rm SS
,
v,.Span{P:2, - k)}
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
f,(I):
.2. al~21-k)
-
By varying Do' in Eq. (4.5). we can geneJ1ltc new functions and set of all such possible
(4.5)
function. constitute u.. 'p""'c V,. A signal in such a space is illu5tJ1lted in Figure 4.6.
I
'.,:It_ l) ,I f{2J-2) II f{2>-l} I
"'" ,
,, ,,
"
FIGURE 4.'
., )
, .
,
.., ,
,
VI Span{~211
, k)}
(4.6)
,
VJ Span {;(211 - k)}
Let us now mo'"C on to the conttpl of nested sp;lCcs. This corocepl is one of the comet
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Closely walch Figure 4 .3 and Figure 4 .5. We have a l ignal in Vg space which buically meam
Ihat lhe siglQI C:lll be represented UlOing thc bases of Vo. Our sign~1 is givcn by EQ. (4.4), that
is.
/(1) 3~t) + 4i1(I - 1) + 2jt(1 - 2) + SjI(l - 3) +4j1(l - 4) + 3~1 - S)
+ 3~r - 6)+ 4f1(r - 7)
TIle wne signal can also he represented using the bases of VI. that Is,
/(1) " 3f1(21)+ 3j1(2t - I) + 4~2t - 2) +4j1(2t - 3) + 2j1(2t - 4) + 2";2/ - ') + ... + (4.7)
Wc ha''!' substitulCd 3~:r) by 321) '" 3f(11 - 1), 4";1 - I) by 4";21 - 2) + 441(21 - 3) and
so Vo- Therefore we say thlll Vo is contaiDed in VI or
011. This is true fOt Iny signal in space
Vg c: V,.V, is a finer space than Vo and contains all the signals in Vo- V, is a bigger space than
V, .
This is poss ible beeluse baSC$ of Vg itself can be represented usin, baSC$ of VI . For
cumpk.
jl(1) 4'(21) ... 4'(21 - I ) (4.8)
~I - I) .. 2t - 2) + fl{21 - 3)
TIle ,dation is called ,.".Ung relallon or renMml:nt ",Jallon or dila tion tquallon .
Simil.... ly. V, c: VI, i.e .. any signal in VI can be represented IIsing the bl..!cs of V2. Again
Ihis is possible because any base of VI can be rcpresented using the bases of V2, i.c.,
4'(21) = 4'(41) ... 4'(41 - I)
In genernl. we can write
(4.9)
Pictorially. we visualize this I..! exhibited in Figure 4.7.
BiISed on the relalion Vo c: VI, it is nalll".] 10 alik, wbat is mi ss ing in Vo thlll makes Vo
a subset of VI' Wc p<I anoIhc:r queslion 10 understand lhe meaning of ahoyc question.
Disc~ Wavele! Tnnsf..,., 57
:---- '.
'r- - ',
-1Hr.I-- - '.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Consider the relation 91 ' C 91 l where 91 ' is !he space spanned by !he basi s vectors 7 and ].
91 l Cl the space spanned by the basis vec!OIl i. j and f . We now ... k. what is missing in 91;1
that makes 91' subset of 'llJ, Yes. you guessed ;t right. The basis vector f. To answer the
first question. let us e~plore another fUI\Clion called H ur wavelet function .
1 0 $ 1 $ 112
"(I): - I 112$1$1 (4.10)
1o elsewhere
.,'
'"
-,
It is quite easy to see that the set of functions {!V(t _ k ), kEN} form an orthonormal set of
basis functions. Like Haar scaling function, there is 00 overlap between the translates of !Kt)
(see Figure 4 .9).
, ~" y.(1 - 1)
,.----','
,
y.(1 - 2)
,,, ,,
,, ,,,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
,
, ,
,,,
0
,n ,'2 J
,,, ,,,
, ,----~
,
Therefore,
J- \If{r) 11'(1 - I) dl =0
In general,
-
J- jll(1- m ) jll(t - n) dl = 0.. _ 8 (4. 11 )
- -J
That is. tlw: integral is I if m = nand 0 if m .. It. F igure 4.10 shows that \If{r) \If(1) dl = 1 .
-
Wl Wl
o f-----,-O-~
,n
.,
FIGU RE U G Firoding IIOI"m of ULlI WlI'o'Clet fUI>Ction.
Let Wo be the space spanned by the orthonormal set of bases {\If(r - k), keN} .
DiIC~ W... lct Transform 59
FonnaUy we define.
Wo Span {!f(1 - I:) } (4.12)
How does I , ia;na! which i-s an element of Wo look like? Fi gure 4.11 SboWI signal in Wo-
,, ,,, ,
,
,, c!- ,
, , ,
bi
-,
,,
,, , , , _2 . ~
~
-, , , , ,
ncuit 4.11 A lipal in ipacc W..
Wbat is the speciality of signal in Wo (spanned by Haa.. wavelet functions)? In every unit
imerval. if the function valkle in first half is lit units then in the next ha lf it must necessarily
be +m units. OthclWise. we cannot expren the signal using the bases of Wo- that is. 'Kt) and
iu translates. Th us, the space Wo is highly restricted space. The tignal in Figure 4.11 is
expressed using the Nses of Wo as:
COIlsider I Kalcd version Qf II'<t). We will 10. only for dyadie lCaling. Thai il. selling by the
integer power of Iwo. Consider al fint 1I'(2t). Given 11'(1) as defined in Eq. (4. 10). 11'(21) can
easily be shown to be
OS I S1I4
InS / S in
el lCwhe re
60 Insight into W.",lets--From 'Jht:ory to Practice
We deOOte the space .plllned hy the ..,1 of onhogonal base. {1V<21-A), keN} .. W,.
Formally we define.
,
W, . Span{[I'(21 - t)}
'1'<21 - J)
IjI(U 2) =~~:tt
1jI(2I _ I )
"'"
, ,
, "
Is Wo c: W,7 Or in other word$, can wc represent. signal in Wo by using the base. of W,.
The an.wer i. 'No. See Figure 4.13.
..,
" '" "
Since linear combination of 1"(2/) and it$ translates canOOl make I 1"(/), 00 l ignal in Wo
un be upressed using bases or W1. ThererOft. Wo et. Wt. However we shall lhow that
Wo.1. Wt . This IIlUIlS tbal biIsn of Wo an: onhogonal to bases of Wt. Figure 4. 14 shows that
f ,-(r) ,-(21) dr .0 .lso f ,-(r) jV(2I - 1) dr .0.
,
., .,
,, ,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
, ,
,
."
, , ,n
"':11 - I)
,n ,
, f+,-- - -
Note tbal the area under the fwx:lion, which is the product o f 1"(1) and 1"(21), is zero.
Similarly, the area undr:r the funclion which i, the prodUCI o f '(I) and '(21 - 1) is also uro.
Tltis implies that space Wo is Otthogonalto the space W,. Again it can be proved tllat W, .1 W,.
TltUI. we h l~ aOO!he:r most important relation in wa~lct theory.
.. W_ .1 Wo.1 W,.lW, .1. (4.\J)
So fir we h.ave learned tWO Yff)' imporunt COI"ICepts in wave let theory with respect to
fUllC1ion spaces. They...,:
Space spanr.ed by scaling function basel an: nestc4, i.e.,
Spau spanned by wavelet function bases an: ortbogonaJ among lhemselve$. Thus,
. W_ .1 Wo.1 W, .1 W2 .1 .. .
62 Insigh' into Wavdets-From Theory to Practice
We are now ready to answer the question we have posed earlier. What is that missing in
Vo 10 make it a subset of V,. To answer this question we consider a signal in VI as shown in
Figure 4.15.
, ,
,
,
,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
2 2
..,
2
h
:.. . - ,>
: I I ,
!' 2 ! !3
-2
..:!.
Since it is a signal in V,. we can express it using the bases of VI' Thus,
+ 2;<21- 6) + 29(21- 7)
It may be observed that the signal segment in the first uni t intervlll can be represented
using the bases of Vo and W(). I.e..
4 +2 4- 2 .
4 ;(21) + 2;<21 - I) = - , -;(1) + -y-lI'tl) = 3;(/) + \ll'(t )
Again
4+6 4-6
4;(21 - 4) + 6;{2I - 5) = -,-;(1 - 3) + -,-II'{t - 3) =5;(1 - 2) -II'{/ - 2)
Ditcn:le W clc:t Tran.form 63
,
,
,
.,
: ~ Jf(,)
, ~
: ~ 01( Zt -
l
fO" I)
,, ~
, <r)
" , "
, ,
, , ,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
FIGURE 4.1 ' Decompo>itiOll of a.igrW in V, in'" >igrW in V. and W..
Lastly,
2+2 2- 2
211(21 - 6) + 211(21 - 7) _ -,-11(1 - 4) +-,-11'<1 - 4) _ 211(, - 3) +011'<' - 3)
Thus,
ft.1) = 4K2I) + 211(21-1) + 611(21- 2) - 2;(21 -3) + 4K2, - 4) + 5KU -5)
+ 2K2I - 6)+2K2I - 7) (4.14a)
= 311(,} + 2K' - I) + 6Kt- 2) + 2K'- 3) + 11'<') + 411'<' - I) - 11'<1 - 2) +011'<1- J) (4.14b)
Wc could up~ss a signal in V, space, in .erIDS of bases of Vo space and Wo space. If we
combine the bases of Vo and Wo space. we can up""'s any signal in v, . pace. Wc upress !his
InIth malhcmalica.lly as:
V, -VoGl Wo
We call Wo .... ,he complcmcnwy space of VI> Now. we found out the "hing' tha, w.... rrJuing
in VI> ,ha,
made Vo a proper subset of VI' Wc call Vo and Wo spaces as complemenwy spaces
bee,use Vu and Wo space.! arc onhogonal and Iheir bases together can represent any signal in
.he ne.' 'higher' or liner space V,.
We ca.lllhi. as the decomposilion o f a finer sign.al into lwo coarser signal . Figure 4.17
show. the standard representation through diagram.
'.
3. 2. ,. 2
"
4. 2. 6. _2. 4. 6. 2. 2
1. 4. - 1.0
w,
FIGURE 4.1 7 Vi ... oli.l.lllion of Eq . (4.140) and (4.14b).
64 INigh' ink> W.vel<,s-From 'Theory '0_""
How ~hall .... e pro.e tha, Vo.! Wo 1 W~ mu>! show lh.aI the bases of Vo are orthogonal '0 base!;
of Wo- We need to show tM orthogonali,y only in tile Case where both the bases overlap (exist
in the same in'erval). If ,he ~. do not ""erlap. nalUtally they are ort/wgonal because the
prod"", of the 'wo funclion. will identically be zero at evtry poin!. We know that 4(1) and 1;(1)
are defined in the same in,erval [0. 1] and hence o"",lap. So we win try to prove that 4(1) is
orthogonal '0 ,...;1).
~ ,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Pic,orially this i~ depicted in Figure 4.]8. Produc, of 4(1) and \1'(1) gi"". \1'(1) wboe net area
" zero.
,"
.', ,
"
,
M 11<')
.. J JIIl ) ,,(Il tlI - 0
,
"
fiGURE 4. t8 On~'Y or f(1) ond 11'(1).
Let us go back to our previous result. We ha"" proved thal the bases Vo and Wo together
span V,.
That is.
(4.16)
Equation (4.16) tells that any signal in "I can ~ expressed using lhe bases of lVi_t .
WJ-.l .. . Wo and Vg. You win recognize laler thal Eq. (4.16) is the fouadation of wavelet based
de<;omposition of the signal . SchemalicaJly.this process is shown in Figure 4.19.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
V.I-<
'H
vH
vJ _, Wj _,
wj _,
" w;..,
WH
Our basis """tors in 3t' are orthonormal. In addition to onhogonality. they satisfy the relation
i i ",].] '" k k '" I . Hence OfIhonormaHty, In function s]laC<'. CQfTeSponding to dol prodllC'l
we have multiplicalion and inlegralion.
Consider base! ;(1- k)s of Vg. We loo ... (hey are ortbogonal. In addition. we have.
J ~r)~I)dl"'l
J~I -k.)p:r-k)dl.1
making the basc.s of Vc Qr\hononn.oI. We say. bases of Haar at ",ale level '0' are onhO/l()rmal.
The k1ual value of ",ale is giwm by 2i ",here j is the scale level.
What aOOul Of\hononnalily of~. of V" where scale level i. 11 n.. base! of space V,
are iran,lal'" of 9(21) . Lc:t us check the valuc of imegral
Figure 4.20 depicts lhe process.
J21) ;<2r)dr . [ts value is In.
66 Insight into WaveLets From 'Theory 10 Pr:>clice
fJ{2l) f(21)
o o
FIGURE 4.20 FindillJl the llOfflIllIizinll con" ...1 ror f(2Ij .
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
So to normaliu 9(21) and its translates. we multiply each by .fi. Refer Figure 4.21 for
visualization.
o o
F IGURE 4.2 t N"m>olivod _ tor v, .
(4.17)
Consider bases for V1. Here the scale level is 2. Heroce we call bases for V1 as level 2
bases.
Since Jf(4r) f(41 )dt = 1/4. we shou ld multiply each base by ./4 = 2. Sce Figure 4.22
for visualization.
f(4/) f(41)
o If4 o 114
We redefine V1 as:
Vz .. sp:n{2j1(41~1:)} "SPKan{29(21t~k)}
Let us generalize. Consider un-llOI1I\(l!ized base, for Vj. "The un-ltOIltlllliz.ed bases for
/lOW
Vi are translates of ~.2j,).
Since
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Whon wc express a signal in u:nns of normalized base,. we have ID repeal lhe oonnaliting
conSl.8l11 in u<h lenn. To avoid lhi. repelilion. an abbreviared nOUllion is introduced. We ",rile
2j129(2J' ~ k) as 'i.
(I). The first subscript denotes the ~ale level or ,imply Ievellll\d the
second sub~ri]ll denotes tnnslation in !hal ~ale level. Thus. the bases of Vo are:
{... fI!,-1 (I). !Ib.o(t) 11.1.1 (I) ... 1\,.. (I) ... }
{... Yj._1 (t). 11'J.o(l) IO'j.1 (I), .... 10'/.> (I) ....}. where ~.>(r) are given by
BoIh these rdation follow5 from the fat!. that Vo aJId Wo arc subset of V, . As a result of that.
any base in Vo and Wo can be e:cpr~ued u5ing tha! ba= of V,. With respect to normaliud bases,
this relation can be re-expres.sed as:
Ir.: 1,- I 1
f(t) .. J2 ... . 2tt(21) +Ji ... . 2;(21 -I) " Ji. 1'1.0(1) +"Ji .1'1.1 (I) (4.19)
Ir.: le I 1
\"(1) s "7..' . .... 2;11(21) - ~ ..... 2f(21 - I) ~ _ . 1'1.0(1) - ~. 1'1 , (I) (4.20)
.... 2 .... 2 ..,fi .... 2 .
TMse two relatioos. which seems quite ob.ious with respect to HaM wavelet syste:rn. are
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
IlK: ,omer slOne or w,'-cIC! theory. When we go on other wa'-cIeIIiYlilCrnli, \hili relation may not
be that ob.ious 5ince mol;! of other wa....,l~t .yst~ms arc f"",ul in nature. W~ cannoc draw
directly ,he fu!>Clion. for .isuali:u.tion.
Generally. wc write: the above relalion as:
(4.21)
\"(1) " :L g(k) 1'1.. ( I) " :L g(k) . .fi~(21- !) (4.22)
Later w~ will interpret {It(k). teN}. {g(!).... E NI as low pas. scaling filler and high
pass wavelet filte:r coefficients. In ronventiooal signal processing wha, really mUlte:rI is ,he
propeny of fihers (f~ueocy ""ponse. phase response. ClC). Wc will 'Iudy .uch propenie5 of
these filter coefficients in dewl laler.
Before proceeding fUrlher. ~ would like to introduce .ome: other 'ypes of wavelet
5y5t~ms. By wa""ICl .ystem ~ mean boch !he scaling function aJId wavelet function with their
refinement relation defined through a known set of coefficients. To introduce other Iype of
wavelet. at first Wc need to familiarize with another ,oncc:pt called s upport of the wa""let
syste:m.
Convemionally. we defi .... the .upJlOfl for only the base. of VD and Wo. In the case of Haar. wc
have defined ~I) and 1;(1) in the unit intel'\al. Therefore. the support of ~I) and 1;(1) is ]0.1].
There is an interesting relation be1wecn length of support and number of coefficienlS in the
refintment relation. For orthogoool wavelet syntm. length of support is always 0Jle less than
the number of coefficients in the refinement relation. This "";11 become clear. as we proceed.
I
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
OSISI
IS I S 2
elsewhere
It obeys a ""aling relation giveJ\ by
11'(1) K
, ,
2'11'(21) + ]21 - I) + 2'21 - 2)
Let uS ay to analyu the abo,.., relation . Since I) is defi .... d in ]0. 2]. ~21) is defined over
]0. I]. 21 - I) is defined over ]0.5. 1.51 and ~21 - 2) over [I. 2]. Thus. three scaled and
tJanslated version of ~I) cO\'er the interval 10.2] and reprod""c ~I). Figure 4.23 shows the
triangular scaling function and ,hows bow it. translates of ""aled version make up the original
O~.
,
,,,
,,
,"
,'
,, , " ,
I " ,
.21_1) +
,
'2(1(2> - 2)
deri'-nliQfl of Ihis relalionship will be plai.-ed la"". Al preJent it is sufficient 10 know mal. le
get 8(.1:). we nip h(k) and change (he sign of alternate coefficients. Flipping mt:ans reversing the
sequence of coefficient . In (hi. example lln. 1.ln) an: the sequence in h(k). l1Ic:n g(t)
coefficients arc either {-In.!. -In) or lIn. -I. In) .
4 .9 DAUBECHIES WAVELETS
1I i. Daubcchie . woo gave solid foundal inn for wave1c, Iheof)'. S~ <Je""lnped many wa""lel
sy.tem w;m oompact support (fini!e number of coefficients in the refiJle1!lenl relation). We will
deal the design of Daubechie. type wavelet sy.rem in a "'parare chap!er. In this section. we will
explore OJle .... avelet sySlem called l>aub-4 type ..-a>tleI 5Y5tcm .llIe peculiarity of this wavelet
system is that. thtre is no explicit function. so wc: caaOOl draw it dirtttly. Whal we an: given
is h(l) the coeffident< in refinement relation which """""Cl .,o(r) and tran.late. of .,0(21). l1Ic:se
coefficients for normali~d Daulr4 are ... follows:
hO).
'"I~ (3 +,fj)
M
h(2) '" 1~(3-..[3)
M
How shall "'10 see ilS $hapc the n? We Man with a ' uniform' ;(1), i,e. we .!.Sume f(f) to be
a KjUlU'C pul.se of unit heighl and spanning the support of fl/). Then we ust!he alxI\1o rdinemen!
n:lation and build ;(/) ilcra!i vel ~. Sti()ll 4.10 describes !he Sleps in\'ol\ed.
The problem hen: is that n:finemenl n:la!ioo lells us how bilser scalin, fUlIClioo can be
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
uprl'ssed usi ng !he smal ler 001.'. I! 00c1i no! !ell us anywu~ how to draw !he bigger one or for
that mat\~r how to dlllw the ~mallcr onc. 8c.-ause of Ihis. il has 50me pulIlIcl with fractals.
Many frac!als have this nature. The ..hole is made up of scaled and !ranslated \wsion of i!sel f.
For eJCamp1e. fern leaf as sho\r.n in Figure 4.25. This fern leaf has no explici! equation !o draw
itie lf. Here i! is generated !h rough 1UI i!ernli\c scheme . The companion Webllile of thi s boot
( ht\p:f/www..amribLfdU/ nlpuhl ic1llion"""nelds) rontains a spn:ad<heet implenlCntat ion of
this scheme. Open !he spreadshee1 'Fem.d s' and pn:ss F9 !O draw the piC!un: in Fi s ure 4.2.5.
To see Oubhies 4-tap wave ltu, _ will Initial ly use a brute fon:e approach. NO!e !hat
'1) of Daub-4 is defined in !he range I" ]0. 3] ~.use number o f cocffic~\ll in the n:finement
n:hltioo is 4. fib) lies in the ran&e I . [0, 1.51. f(b - I) lies in the range I . ]0.5. 2J. Similarly.
9(21 - 2) in [I, 2.5] and f(21 - 3) in 11.5. 3]. NO!e again thal four 21)0 5pan the range ]0. 3].
We now sample f(1). Sample size must be muhiple of 2("" - I) ""hen: N is the number o f
coo:fficicnt . Thi . is 10 place 4(r)s ",,,,,,,.11 in lhe &ivcn range. U! uS lake sample s ilt as (8().
i.e . we lake 18() samples rrum ~I), ~rform lbe fo llowin& l teps:
72 [Might into W".. lelS- Frorn Theory t<> Proctice
(i) Initialize arrays PH120. PH121. PHI2l. PHI2J of si,..: ISO with zeroes. Initialize array
PHI of size 180 with ones. Initialize another army PHI 2 of size 90 with zeroes.
(ii) Take alternate values from PHI array and fill PHI2 array. (We ..... making 41(2,) from
9(')')
(iii) Take PH I2 array. multiply with h(O)/i and .lOre in PHI 20 array from position I
10 90.
(iv) Take PHI2 array. multiply with h([),fi and .store in PHI2[ array from position 31
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
to ]20.
(v) Take PHI2 arrny. multiply with h(2),fi and Store in PHI22 array from position 61
10 150.
(vi) Take PHI2 army. multiply with h(3),fi and 5l0l"t in PHI23 array from position 91
to ISO.
(vii) Add PH1 20. PHI!]. PHI!2. PH123 anay. poim by point. and store in PHI IlmIY.
(viii) R'1"'at ' teps from 2 to 7 ..,yern] times (usually within to iteration, PHI army wHl
converge) and draw 41(1). Figure 4.26 shows the positioning of "moos array values.
1o(2N2IA.2r - 2)
1~(JtJ21A.21
o 0.' '.0
I ,
- 3)
C::===============JI.,;') "
C:========::::JI 11(2.)
(ix) Use converged 41(,) val"". 10 d""" If'{r) using the relation:
Figure 4.27 show. the output of the above procedure implenlented in EXCEL spreadsheel.
lbe left figure is wavelet and the <>!her is scaling function. lbe book's website contai n.
the Elcel ~t. Interested readers can download and use it.
A"uralC methods of finding ''lIlues of f<') and If'{r) is given in Chapter 7.
~"
e I ~~ 7J
,
~ ~
,.
/
.,
, 1// \
.,.
, L , , "VTrf ,
. . ~ ~ ~::;; 1';1::;
--- - -
M
. ,.... 1
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
116WWJ7
1$/SIII01
Itil11lii11l
1169Yl"b1
2l1Kli91
1166132271
1' / 'l2Ill4J
1!i81!9l11i
IIlHIt5J)j
....
IQi!lS8l!II 11611616S tll! 116I5JI!
SUMMARY
WII.,.~I~I~ an~lysi, of signals is an inle~Slinl. aod ~Iali " ely ~m. new 1001. Similar!O
Fourier series analysis, whe~ si nusoids are chosen as !he basis fuoction. wavelet anal ysis is also
baKd on a o:Ieromposilion of. sianal us;na an orthooonnal (typically. ulthough no! ne~ly)
family of basis fUI1l1;ons. Unlike a siRe' wa'l'. a WlIv",kl has its ~n"'iY rom...",tnllnJ in Iin,I',
Sinuwids an: useful in analyzinll periodic and time-invariant phenomena while wl\'cleu are
well suited for the analysis of transient. lime-\'llI)'ing signa ls. Discrele wa''eiel transform lheofy
~uires ',",'0 K IS of ~I aled functions called sclllng tuncllo n and w ~ltt Function. SIlS;S lielS
o f sealing functions span a nested spl:. Most of the se~Jing and wa.,.elet functions are f!'Klal
in nature and i!emli.'c methods are required to see its sllape.
L\ I .KCIS I .S
4.3 U1 Vi be tll< .pace of an finite energy signnl ~ Ithnt are continuous and piece-wise
linear, with po.. ibl~ di.>eontinuities OI.'culTing only at the dyadie points U2',
k E Z. This function, are canul the Unear spUnes, ut x) be the scaling function
defined as,
.. , -1 $, < 0
~t) = ~-I 0< t S I
1 I' I >I
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
INTRODUCTION
In che last chapler ~ u~ Daubechie~' 4-tap wa ..,kt system coefficumu. We ha,.., two S"U u(
eoefficicRls Ih(.)) and [&'(~JlIhaL delil\/:!he refinemem relation. 1'he>e coeffic~ms decide the
shape of the scaling function and ",a,..,let function. This shape in turn decides the application
wl\c:re we can use \he p;Lr1icular wal'Clct. The coefficienu (h{!) J and [8(1,) J BCt as signal fillers.
which we will discuss in due course of time. Vast amount of literature is available on filter
design for specific application but il i, inacce$sible to an average reader since practKally all
methods uK frcq~ncy domain as well .s rompb analysis o.:oncepu 10 amve I t the filter. In
lhis s.ecuon wC will find methods 10 anive at (be filler coe fficients derived by D<Ulbedues, As
mentioned flll'lier Oaubtthics u.sc\l ideas from Complex analysis and Fourier transfooom 10 derive
the >efficient . We. instead. 10 for a time oom;oin ~pp<OllCh for deriving die coefficienu.
"-,
'od \V(r) = L r(kN2f1(21-k) (~.2)
.~
"
76 Insight into Wa",\ea-From Theo<y to Prac!.ice
Scaling and Wav~lets m"~ satisfy some OI:C<'ssary conditions such as onhogonalilY and certain
OIh~r desirable p"'IX'"ies such as sITIOOlhnes.s. These condilions. in lurn, put reslrictions on
scaling and ",",'det fU""lion coefficients , We ",ill ...., 001: by one.
"-.
l~ L h(!)J2.~
.~
"-.
L It(k)".,f2 (SA)
.~
f ;(1);(I-kldl"O ifk.,.O
/1 -' /1_'
f iXl)iXl-k)dt = 2 L, h(l) L
loO .......
h(m) f jl(21-1) . j!(21 -2Jc-m )dt
/1 -, /1_,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
N_' N_'
" L h(l) L hem) 0,.1.."
''' '
"-,
. ""
"L
,. h(l) h(l- 2Jc)
"-,
So, J \Kt) jl(t - ")dl" 0",. =;> L
,. h(l) h(l- 2.1e) ~~... (B)
Similarly
"-,
J ;CI) !&(t +k )dl = 0"... =;> L W) h(I+2.1e) .. 0",. (5.6)
,.
If t '" O. wc have the "1= normalization =ndilion
"-,
L hl(I) = 1 (5.7)
,.
For t .,. o. wc have the onhonormality condition
,.
/1_'
'.
L h(l) h(l-2k) = L h(l) h(l + Ik) = 0 for all .le .,. 0 (5 ,8)
w. call the condition given in Eq. (5.S) as double shift onhogonality conditiOll .
Example: Let us see bow this equation ,ranslates 10 a wa,..,let synem with four coeffICients.
I.e.. t{" 4
Let .le " I (.le i, blLlically the inleser represcnting ,ranslation in ,he relalioo
f(r)
'0 ,
- f(,-I)
2 ohin
J f(r)f(r - I) d/ - 0". h(O) h(l) + /t(l) II(J) - 0
We have a refinemc:m relation that connects '1'</) and 9(21 - k)s. This relation define.
coefficienl5 (g(k)). That is
,-,
]0'"(1) .. I. g(k) J2?(2r - k1
,~
We IIOW establish the relation between the seric. (h(tll and [g(k)). We have """n earlier that
41(1) and 1/) are onhogonal. This implies J!X./)/iI(lldl = O. 11 can be shown thaI, for Ihis to
be we. 1/'(1) mUSt take the following form: (The derivation of tru, i. deferred to Chapler 9).
,-,
\f(l) " L (-l)'h(kN2;(21 +k - N t l). L (-I)II -' - lh(N - f - Irfi!X.21- f) (5.9)
This relalion implies Ihat g(k) .. (-1)", - ' - 1 h(N - " - I) (5.10)
g(k) '" (-I)" h(N _" _ 1) i, also. "alid solution.
For N. 4. thi' mun,
g(0) .. (_ 1)4-0-1 h(4 _ 0 _ I) .. _II(J)
g(1) '" k(2)
g(2),. --h(l)
g(J) " k(O)
i.e . 10 lIel coefficicnl ""rie, (g(f). simply reverse the coefficient ocri (l1(k)J and change lhe
sign of the coefficients in a\lem;lle positions.
In term, of the normalized ""15 of coefficients. lhe relalion is:
g(k) " (-I)' h{N - k - I) (5.1 1)
is in rewrse ordor to that of ~I) and with alternate nip of sign. we ClUl mal<e J <K1)'l'f.I)dl .0
" Ill')
F I GURt: 5.l The buikW1g blocla of ... ') >nd "" .... Il>o "",... "') <>11 b< ...... onhupul to ",'1 by
,~i.1 , .. """" """If", ... " .. ,t.. of ... nt"" '"
~"<"" onIef >nd "i,h .1"' ...... < nip ,,( ,i!'t.
of"'" <o<lr" ...".
Anutho< I""-'ibih ,y i. fOIIuwn ;n F'gUTe 5.3. Th;. ",Iu",," rorre>pOl>ds ' 0 g(k) " (- I)'
h(l - t). Her<: Ihe suppon of ~I) is [0. 3) .nd Iha, of O(I) i, I-I. 2). Th. k for "hich g(k) ha.
""" -"tcro ,.Iues or. 1-2. -I. O. I J. In .,any 'c"_ .s. 'hi' is Ihe ",Iu,km 2i'-cn
,",
." ." '"
" ""1
,
"
." ~21
'"
" ""
,
5.lA Condition 4: Approximation Conditions (S moothness
Conditions)
ID mmy >pp l"";OD>. "'c "c..,d '0 'pp,..";m,, ';gD.1 u,iDIl scaling (UDC';OD. S" "'. ask 'he
q""Mi"": To ",h.,lkg"'. p. Cl .. 1I.IInO",i.ls I. ,. 1'. be: repmduced euc,ly U'"'~ I b.,."
I"
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
<If - JIPt(I- k ) <11 lproJ<'C' " on 1<, lhe bas" f/l(1 - i)1 (S.U)
[Nol<" "'., <If .,. id, t. PI>, ,upc""'rip' indic.uing "'., <If I> llh rocffJCienl ffiI"J'e'P""din,1O
'he projc<lI"" of nlOnom,.1 /' on '0
llh h"'''1
To ""hi"," "'is. "".Iing funcllo n should PO"""" nnain propcrlie> (,uillbl" ,hapc). lIul ib
shapc depend ' un ,he c""flicicnl.> h(t).
Now ".'" .",'"me ,h.,
11 ;s possible 10 repr""'"! "",Iy 1110""",;,1, of ontor "PlO p u,inB
1 ,,,en "" . Iing funClion. Thl> will Impose cenajn condition> On h(l). Wh., arc Ihe .. rondllio",!
10 '"" """i"" . we will show lh 31 il ;, qUllc easy'o ,r,n., lat. 'he cond"ion on "".IiOIl ("nellon,
On '0 <""diti""5 "" wavdd funclion.
Le' uS m.... try 1<, projc<o' " on It> 1jI(I). Sc. what h'l'I"'n<:
lero. NOIe Ili::,. "'al. In literalu". ,b. inlegral J "[(1) <11 is col/od plb nlUm.nl or r"n''lon
fit ).
Le, P ~ o. Thi, """d,II0n implies ,h.,
For a 4-lap filler (corresponding 10 a wavelel syslem of support 4). Ihis reduces 10
11(0) - /J{l ) + Ir{2) - h(3) = 0
For p = I, we can prove Illal
f r!V(t) dt '" 0 ~L
, (- I)'; k/J (k):o 0 (5.15)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
In tllis manner. we can put conditions on scaling function coefficients, and then solve for
tile value of h(k), 1lIe number of conditions Illal you can put depends on tile suppon of the
wavelel system you are designing. If tile suppon is 6, we can put six conditions to get unique
values for h(k). We say we Ilave 6 degrees of freedom. Out of these 6. three (50%) degrees of
freedom will go for constraints on onhogonality. The remaining degree of freedom can be used
at our will. Wllat Daubllies did was tllat, the remaining degree of freedom was fully util ized
on constraints on smoothness.
This can be easily proved as we have done for the onhonormality of tl1lllslatcs of scaling
functions. The condition is redundant while designing wavelets becau5.e g(k) is related 10 h(k).
ConJlmin" ~n cOf!1fid~n/'
Wc h.>"c wri"en tile con",.in .... Howc"r il "an be c""ly ,""own Ih., tile "'lu.r<. non".I;,.,ion
rond'lion C.1l t.. obla,",~1 from the norm.li,.non. orthonom,.hty .nd p = 0 ~"Onduion. So.
dfcC1;I'dy Ihorc ar~ only fou, con"";n"'. Thi. C.n be c., ily so"'cd u>in~ MATLAB.
M.. hcm."c. 0( evcn Excel Sp,ud""'.1 "'".... The'" " al", an opcn ",",ce package co llod
s"g~ ("." " . sag,m,,lh.,,'1t ).
Tl>ore . r<: 'wo ",IUI;on,. lbe .. me uP'" pcrnlu,.,;nn. bul ;, ;, <"nwnlion.' 1o ,ake ,1>0
secood nf Ihe",'
'"'
1>(1) .. ~(3 "' ,fi)
M
1>(2)", 1[.;"i3 _ ,fi)
4 .. 2
1>(3) .. ~( I-,fi)
M
0"". "e oh'.;n m.mdonl.' (lI(k)l. "'< Can tind Ig(k). Ju" ""..,.,;e Ihe <mkT of <"oemd~n"
and change 'he .I~n ., tII~ al<ornato posili",,,. Therefore.
g(0) .. ~( I- J3)
'"
-, 0
g(1) ~ [.;"(3 - ,,3)
4 .. 2
,
,
0.'
o,
1\.
, '\
" ,
, , , , , , , ,
"
Fill.r "",,(f!Cient' wilh N " 8 .re:
1.2.10377813.lO9. 0.7 14M6570553. 0.6.\08801619.\0. "().O2?98)169"17.
-{I. I 8703481 1719. 0.OJOS.l1381836. 0.0328-830 11667. -{I.O I0597-1()1 7851
5.4 DES IGN OF COIFLET WAVELETS
Coill", "",,-,'oIe.. are obtained by imp<"mg " ' nl'ihinl: momoot cunditions on both ""aling and
"'",'<:1.,, function . More conditions on ,'anisllillg momenls imply mon: n il c""ffle;""". The
minllnom nU lllocr of "ps is flJllr. If Ihc numt.:. of tapS N. 6p. 'hcn. 2p numt.: , of ,'anl>hing
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
"",mem cond lnons ..~ in'f'O""d on w",'elel fUllcnon. 2p - I 011 ""ann, fUllclion "nd lhe
ren'.ining un nOnl,aJity and un"ugon.IIlY condiu""s. Thus. lItc mndlliun, in'po!idl arc,
J ?lJ)J'''' I
J ?ir)?lr-k)J,",q"
N01~ t"', J ;)(1) d, ,,0 ,"d J ?l1) \If(' - t) dl" 0 for all k. rel",~ h(n) an<! R(n).
For p I. filtcr <""meien!> iIr'< gi,'cn by
h(n) [0.038580777748. -0.1 269691253%. -0.077 161555496. O,607491().l1386.
0.74';687558934. 0,2265842651971
Fmp -'" 2
1,(11 ) _ 10.0163873.J.6.t63. -O.(l.I 14M936782. -0.067372554722. O.386II00668H
0.8127236J5450. 0041 7005184424. -tl,07ft.188599078. -tl.0594:J..141 86-16.
0.023680171947. O.00S6114J--f819. -0.00182320&\71. -O.OOO720~9.w51;
, ,
,
, ,
, ,
" "
," I "\
" 'J ,
, , , , , , , , ,
" "
86 Insight illto Wavdets_From Theory to Pnclice
5.5 SYMLETS
The solu tion for the wavelets which was given by Daubechies is not always unique. She gave
solutions with minimal phase (max imum smoot hlless). Other cho ices can lead 10 more
sy mmetric sohniOIlS. llley are nevcr completely symmetric though. Sym1ets have the followi ng
propertics:
I . Onhogonal
2. Compact support
3. Filter length N = 2p
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
o. ,
0.'
-0.'
0
o ,\
)'IGURE
.
S.,
V
6
1 ~---c:--L-:---:--'
o 2
Symlet ..,.,Iinll and w.""lct function wilh 8 taps.
6
As exp lained in' the Daubec hies wavelets have surprising features. such as intimale COIlnections
with the theory of fractals. If their graph is viewed under magnificaLion, c haracteristic jagged
' Hubbanl. B.B.. TIr~ World According 10 II-ln~I~ ..: The s'ory of" _~ntO'iCllI '/u,;q .... ;n ',,~ makj" B. oeoond
edition. A. K Petor. Ltd. 1998.
wlUlo Can be "<'Cn. nO """Of hOW .trong the "'.gnilkatlon is. Th!. c~qui'"< romplexi!y of
detail m~'"s that th.", i< 00 simpk> formula for these sc'Ii"R function .od "",,,lets. Tlwy . re
ungainl y and lSy",,,,..ric; nlnc'c<:n,h centu!), "'athe",o,ician> would hlv" roik:<i fro", Ih<:m in
I>orror. Bu, liK< Ihc Modcl T Ford. they..-. be.u,iful. bee.u they worl.
In 'h" se.'ion "e will discuss Ih., pm!","y of 'he sc.hnl! funclion 'hat would h ..
m><le nlndc<n ,n ccn'ury ",a,hen,.O<;a., I"oil from lhem in I>orror. It i. known IS part it ion
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
or .", tly . pm!","Y of ",. Ilnl! fun<1Ion. That" ",. Ilnl! function ~ . 'I "",,,fie, lhe ",tati""
L ;(.<+ n) '"' t. where n i< an integer llIId 0 ,;:.<,;: l. r"Or. 4_lap D.uoc,hie. "".hng funeliun
wl>ose .uP""" is (0. 3). (..., Figur< 5.7) mO'n' lh ...
~O.I) + ~1.1) + 2 .1) " (5.17)
Also. 11(0.2) + 11(1.2) + 2.2\ (5. (8)
Or. in g<""r.!ot.
.T\.f(I._T) . ~2 . _')"1. OSrSI (5. 191
,.'
1.2
,
..,
\
~/
, ,
TIll 1980s. Ma,hen"li<illIls did nOllilink of ",eh f"netions an<! wo"ld not hI" " belie'cd
lil .. "ueh fU"""un' .",,_ Tho grc .. "".~' "f D'uocenie, lie. ,n p,<wing Ihat ,,"ch. funcli"" n,-",
.nd .(so In ,'ho\>o'ng thallhe inlolle, ([1I0<latO.' "f such fUIIClions are o"hugo".1 "nd are lhus.
useful fur signal represenl",i,,". Unronunatcl y. l!>ere is no explicit ",,,,,,sen'ation or ".eh
fUn<1,,,,,, (ex<1-"JlI Hurl and hk. froelal,. 'big!!", fune1lOft i, e.p",~d in lonn' of "".lod
(reduccd) ..<I tran,lll"d versions of Ilself.
L< '.
To pro,'c lilc lbo>< ",I'lIun "'4u;",s Poi""'n ... ",,,,.lIOn fum,ul . W< ",1 1 discu .., il hCITinaflcr.
5.6. 1 Po isson Summation Formula
The P'''~'''n Summa';"" r"<)rmul, (PSF) "1""'''''
'he infin"" ",n' of I"'riodic .... mpks uf 'ny
funC1ion f(x). al Ill< mlcge. points. to be <q ual to a slmilor >lJlll. "htn t.~ <n in its Fouri~
Tran,form dom . in.
C"",mer <on"nuous fun<'ion ft.,"). (""""'n in Figure 5.8) in the time in,,,,,,"1 (0. I). In
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
t~v~
"
,
: =-~=---:-:-;;:,--~--::_==;?;::
J.C:E=:
L
n o. ~
=--;::
---::--=--::--:'-;;;-~d
(S.20)
.... 112
.-
112
f(II). L (I,~, ~ (>lOC<. ~,(.\.) ., ~.. , '" ~,,~,. '" "ill<. " 1 .. Oj " I) (5.221
"b,,,,, .
.om
(t.~.) '" J I(d e- j " ' , d . (S.B)
.+,n
~w.
(5.21
0" InlCtch:mging ,be Ml""""ion.
L I("'" ") '" L L (I.~.) ~'~". O:s y S I (5.27)
This is the general expression ror Poisson summation rormula. If y::: 0, we get a special
case of Pois!i(ln summation rormula:
u".
,.,
;C')= l , h(.r/2;C"")
.~
- ,
-
'" ..!..,,-J-...ul ~({J)12)
-
f ;(x),,-J"~ dx =_1_
, ~
[Lh(kJe-i....1) ~(J)12)
t
H(,,) = J, [~ W),;.)
11,,"
(5.3 1)
S,nce. we "i"Um. ,h .. ,(0) ~ O. (.in", ",ca "n,kr ,(. ) ;. I) i' is ."den' , ~ ., HIli) mu"
be equal 10 I. 8y p<rk><li'ily. We hal. 1I(2/fi.) '" I for any k E Z (se, o( na,unl num"",,).
Since. (5.)5)
"-
{),2 0}4.16 IH7S11 -O.I}22~
{J,f>6!(o -0.00,).1
"' 0,7'J~ 0.17.: 0.02'19
"'
""'.
"'
""
,-
O,Q3:lO
0,9701
I I65J
"=
omS8
-O.1)j51
...(l.IMS
...(l.1f>63
"~W
0.0041
_0.0055
~-
0.0001
"' 1.:1662
Table 5.\ sIIow. ,'aluos ()f DI"1>-4 ""ahng (""";00" inlCf\'al, of 0,1. Th~'" ,'oluos an:
C<lmputod "sing D.u"'->chio ... L' g'''.' Algonthm wh,ch "upl.ioed on del .. 1 in Ch'plcr 7.
SUMMARY
In ,hb ch.pter. we denwd D.ubeehie, <><1hOBOn,1 w,,."'t sy. ",m cocffkienlS (h(';;) I and IKI k) )
based On onl>o~onality aM smoolll""", condition, lhat must be sa".lied by ." " ling and ,,. ,.dcl
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
funcllon . The"" cond lllon . in IUrn. impo", re"ncti"". on 'he valuo o( fil'er c..officien"
Ihrough dil'"On (",fmemen1) .qu.,ion . A ""I of .. mu l"ne"u, IWn-linear equatiOnS on
cocffickn" "'" formulalcd and ool,'cd 10 oblO;" nu"'cricl I ,'.I""s (or lIIe coe[f,cien,", This
"""ghtforw.nl 'ppn",ch oh"i"" 'he n",od (or u",Ien;l1nding .d,'.need Fourie< tmn,rn,m .nd
spectral facloril"ion ",",hods t1131 ate co,,,,ention.ll y used for the purpose . W. nlso discussed
an in'nguinl: pro""ny of !iC.linll functi"n "hieh i. I>Ot ,al"ned b)' u"",al functions tha! we
e nCOUnler in malhem"i ... ,
EXERCISES
~. I For 8--lap Daul:>echlcs ""' elol sy5lcm. tleri,,, the equations Ihat mU'" be sali,ned hy
"'aling fu"",,,,", coefficient>.
5.2 Writo. gene<.1 MATI.AIl or maple mole In1l geneme lhe eqU.llons mcnlk,ned in
q",,"'ion 5. 1, You, inpul should onl y be Ih. numbe, of lap...
5..1 O<:ri,'o Ihe ne",s,'al)' cundions f.,,- P """,shing m"nlen" fo, ",olinll function. in
lerms of "".Iitl~ funclion codr,";en,",
5.4 I'ro,'e lhal "ny ." ,.Iing fnnt',io" in an onhogonal wavele, .y"em ~ti'fy the rol31ion
5.11 Show that Hat function is refinable but its integer translates are not orthogonill.
5. 12 Show that f ,.(1) ,.(1 - k)d, '" 6(u = L. g(m)g(m - 2.1:) '" 6o.~
5. 13 Show that f ;(1) \11'(1-.1:) dl = 0 = L. 1I(m)g(m - 2.1:) '" O.
Akan~u, A.N., and M.1.T. Smith (Editors), Sublxmd and W'lI'elel TrollS/arms: Design allll
applications, Kluwer Academic Publishers, Norwdl, Mil.';$ilchussets, 1996.
Burrus. C.S .. Ramesh A. Gopinath. and Haitao Guo, Introduction to WQ\'e!tIS alld "\1\'(/tl
Trons/amu: A primer. Prentice Hall. N1. 1997.
Daubechies, I., Ttfl UClwrtS on Wovdm, CBMS-NSF, SIAM, 1992.
Strung, G. and T.Q. Nguyen. Wal'elelS and Filter Banks. Wellesley-Cambridge Press. Wellesley.
MA. revised edi tion. 1998.
Vettedi. M. and 1. Kovacevic. Wavt/m and Subbond Coding. Prentice Hall. Englewood Cliffs.
NJ. 1995.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
INTRODUCTION
We never bave 10 deal most of the applications directly with scal ing functions or D.S.$()(:iDled
wavelets, We need only h(II) and g(n) in the refinement re lation (dilation equation) and the
discretized signal. In this contC)!;I, h(n) and g(") are viewed as filters and the sequence of data
as digita l signal. This is from the viewpoint of fi lterbanks developed in 1980s. Most of the
resulu in wavelct theory can be c)( plained either in terms of signal upansion or in tenns of filter
banks. SludenLS in signal processing must have both viewpoints 10 gel full understanding of tbis
rich and faSI growing field.
As said earlier, we never see scaling function or wavelet function playing any role directly in
the CQmputation of signal expansion coefficients. This is because of the relationship between
expansion coefficients at a lower scale and expansion coefficients at a higher scale. Our aim is
to establish this relationship. Further, we use Haar scaling and wavelet function to geometrically
interpret the result.
We start with refinement relat ion. the relarion responsible for signal decomposition.
We have
.... -1 .... - 1
"
Discrete Wavelet Transrorm and Relation to Filler Banks 95
. coe mICLents.
Note that h(O) and h( l ) are normahzed ' For Haar, h(0) = .fiI an d h( l) = .fi
I ' We
aim at relating jth level ~ with the next higher level scale level j + I . so we write
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
.... _L .... _1
This relation plays a very important role in tile derivation. Therefore. we try to visualize this
using Haar.
In Eq. (6.1).j and.l: are fixed by LHS and so is not a variable in RHS. On RH S. m is !he
only variable. Let us for simplicity take j = 0 and k = 3. Then what does the resulting RHS
mean? See Figure 6. 1 for interpretation with respect 10 Haar scaling function.
,,
,,
,,
,,
, ,
Normalized bases or VI
-+------r
, , t.1(1) .. ./2t(21 - k)
.fi I
1-""'~'t:'C"i"':'i-'C'C'ic':"i-':'C'i'c':"~:='='=:t=== ~'_ll
, -f -',
,, ,, ,, ,
,,'.
, no_I) ',
,,
,2
Substituting j '"' O. k '"' 3 and N = 2 (for Ham") in the Eq. (6, I), we get
,
;(2 1_ 3) = L, h (m - 2k)./2; (2t l-m )
.~
J
l h .... _ 1
sj(k ) = L hem - 2k ) f( t )i j' )/2;(2 J* II_ m)dr (6.6)
.. _ l l
The integral in Eq. (6.6) is basically the projection of fit) on to the normalized base i j+t )12
\fl(2fr l t - m) and according to Eq. (6.S ), it is sjo l(m).
Therefore.
2hN-t
sj(k ) = L.
.. .. 2.1
h(m-2k ) sj* l (m) (6.7)
Discrete Wavelet Transfonn and Relati.,n to Filter Banks 97
In signal processing, one of the most frequently used operation is the convolution. Convolution
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
= L.- s(m)hl-(m-k)1
,.-
Note that the running parameter in the summation is m. k in the summatiOn is fixed by the left
side: of the equation and henee is a constant. Also. the series {hem - k) I, is a k-times translated
(or delayed) version of the series {h(m)!. The series {h(-(m - k))) is a time reversed form of
{hem - k)). Therefore. Eq. (6.9) can be interpreted as translation and folding(time reversal) of
one series followed by multiplying and summing with thc: other series. 1bc process is then
repeated for each value of k. Since the process involves folding o f a translated seri es and then
multiplication and summation with other series. the operation is aptly called convolution.
Lel us try 10 visualize the operation by taking two small finite series. Let {11(k). k = O. I.
2.31 and (s(k). k = 0, 1.2.3.4). The two series can be visualized as rollows:
Il(m))
, 2 J .~
,0,T T
h(1)
T
h(2)
T
h(J)
/11(".)1
-4 -J -2 -I ,
, 2 J
T T T 10(1)T 10(0)
T m
h(3) ")
L o(m)h(-m ): 11(0)0(0)
..-
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
c(0):
[11(....... )
L- s(m)h(J-m):.r(O)h(J)+.r( J)h(O)
c(l) =
,
2 J ,
T T
(s(m))
, T T T T, 2 J -+ m
[he... - I))
-4 -J
T h(3)
-2
T T T
-I , T , , J
-+m
"')
FIGURE 6.4
"" " )
Findin~
(loO - m)
convolution SlIm fOf l " \.
So we get the convolved seq uence c(k) by translating one of the sequence by k units (or k
sample positioru;), folding it backwards, multiplying the COITesponding (overlapping) terms in
the two sequences and then adding il.
Discrete Wavelet Tnnsfonn;md Relation (0 Filttr Banb 99
Let us now try 10 visualize Eqs. (6.7) and (6.8). We can rewrite Eq. (6.7) in convolution
format wherein, the su mmation of product of two terms, the running parameter is '+' in one
term and '-' in the other term. In Eq. (6.7) the running parameter in the summation is m.
Therefore we write Eq. (6.7) as:
2';''''_t
s/k) = L h( -(2k - m)]sj +t(m) (6.10)
2f
U+N- L
= L h'(2k-m)sj+L(m) (6.1 1)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
.. . 21
where N = 2 (because we have only h(O) and /t(l Therefore. for k = O. we have
,
si(O): L. 1l(m)sj+,(m) = Il(O)S;+L (0) + h(l)s}+,(!) (6. 12)
For k = I. we have
,
si ( l ) '" L. hem - 2)sj+l(m) =h(0)s;H(2) + h(1)J}+,(3) (6.13)
For k = 2. we bave
,
sP) = L. hem - 4)sj+' (m) = h(O)s;+1(4) + h(l)sj+' (S) (6. 14)
To give interpretation in tenns of convolution. let LIS convol ve {h(-m) series with the series
(sjt,(mH. In our present case, {he-m)} series becomes h( l) and h(O), the time reversed (onn
of series (h(O), h{Ilj.
100 In~ighl into Wavd C:ls From 1lIeory 10 Praclice
,
The silumion is depicted in Figure 6.5.
(s(m)
, ,
0 2 3 _+, m
Ir(l) 1r(0)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
11r(....... )
When we convolve these two series, let the OUlpUI sequence be represented by eCk) (see
Figure 6.6).
1.5(... )
'j*I(O) 'j<I(I) '}<1(2) '}<1(3) '}<1(4) 'j<'(S)
, ,
,
~O)
,0
~I)
2 3 ' m
{h(m ))
It can be easi ly seen that s/.O), 1/.1). sP) given by Eqs. (6.12}-{6. 14) are c(1). c(J) and c(S).
Thus. Eq. (6.7) is equivalent to convolution followed by downsampHng of eve n indexed terms
in the output series.
So Eq. (6.7) is visual ized as shown in Figure 6.7.
--"-"',--'-'i1L_h(_
~_)}---I' j 2 I f - I- ' , - - - '- ,
Finally, the analysis. the decomposition of signal in Vj+t into sum of two signals, OJlC in space
V} and the other in Wj is visualized as shown in Figure 6.9.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
I,(-m) )2 "
si"
d,
t(- m) )2
FIGU RE 6.9 Relation between si'" sJ and d}
This is OJlC stage decomposition. In 'filter bank theory', it is called two band a nalysis.
The seqllence of coefficients (5Jk)} can be fU l1her decomposed into sequellCes {sj-I(k) and
{dj-I(k)}. At eac h stage. the length of the sequences approximately halves since. usually. filter
length is very small due to decimation by half after the convolution. F igure 6. 10 shows,
IW(l-stage. tW(l-band analysis tree.
h(..... ) ~2
" 'j-'
~-) ~2
~-) f---.IE}-;d,_,
~-)
~
FIGURE U O Two-band analysi, of signals.
11 will be shown later that the FI R filter represented by he- m ) is a low pass filter and Ihe
one implemented by g(-m) is a high pass fi lter. The sum of the number of coefficients in
sequence {slk)} and { dJ k)J is almost equal to the number o f coefficients in the sequence
IS}+t(.!:)). Hence, there is possibility that no infonnation is lost in the splitting of the frequency
bands and that the original sequence {s}+I(k)) can be recovered from the sequence Isp:)} and
{d}(J:)). As we shall sce. tha! is indeed the case. Downsampling will surely result in aliasing hut
the aliasing occu rri ng in one channel can be undone by using the downsamp led signal
(sequence) from the other channel.
102 Insight into Wavelets From Theory to Practice
,
6.3 FREQUENCY RESPONSE
The frequency response of a digi tal filter is the discrete-time Fourier transform of its filter
coefficients h('1). This is given by:
The magnilUde of this comp!e)[ valued function gives the ratio of !he output to the input of the
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
filter for a sa mpled sinusoid at a frequency of w in radians per second. The angle of H(w) is
the phase shift between input and ootput.
The first stage of tile IWO banks (channel) divides the spectrum of {sj. 1(k) I into a low pass
and high pass band. It resul ts in scal ing coefficients and wavelet coefficients {slk)) and Id!.!:).
The second stage then divides thal low pass band into another lower low pass band and high
pass band. The firsl stage divides the spectru m into tWO equal parts. The second slage divides
the lower half into quarters and so 011. Th is results in a logarithmie set of bandwidths as given
in Figures 6.11 to 6. 14.
I H(ru)1
w,
."
"-
o
Discrete Wavelet Tr.ansfonn and Relation to Filler Banks 103
IH(cu)1
v, w, w,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
o
FIG URE 6.1l Spliuing of f!aiuency band (aflCr second 5Iag.o).
I H(1Il)1
'. w. w, w,
o .14
FIGURE 6.14 Spliuing of frequency band (.nc. third su.~).
Here we address the problem of finding the original fine scale coefficients of the: signal from
the combination of scali ng function and wavelet function coefficients at a coarse resolution. In
other words we will Lry arri ve al a formula which finds ..,ries /sj+l(k) 1 from lhe: ""ries \sj.k) 1
10
and Id,{k ). To derive thi s we assume that signalft.l) e Vj+L'ft.' ) can be written in terms of basis
fUlICtions of Vj+h i.e.,
~ (j+L)l1 .
/(1) = "", sj.l(k).2 i/IIy+l t-k) (6. 16)
(6. 17)
104 in5iglu into Wavelets From Throfy to Pnw:tice
"d
Substituting the above two relation s in Eq. (6.17), we obtain
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
(6. 18)
1(1) = L. Jj(m) L. h(lI) 2U.tlll!l(2jl,_ 2m -11)
+ L. dj(m ) L. S(II)2(j 1JI2 p(2 j t t I - 2111 -11) (6. 19)
Multiplying both sides of Eq. (6.19) by 2U<-1Y2 9 (2;'1, - k) and integrating, we obtain
J 1 (1)2(jH)12p(2 j l , - k) d,
+ J[~dj(m)~g(II)2U.l)I2;(2jt 1- 2111 - n) 2U
t)i2 ;(21 t1- k )] dl (6.20)
On the left hand side of Eq. (6.20). wc arc projecting /f.1) on to the base 2(;'IYl (2}1-1, - k)
which is equal 10 $}l-I(k). On the right handside, si nce the bases are onhogonal. only one integral
of the product of ifI()s remains and all other vanish. Non-zero (unity) value for the product
()C(:urs when (2m + n) = k or when n = k - 2m. Therefore.
(6.21 )
6.4.1 Upsampling and Filtering
Let us try to visualize RHS of Eq. (6.21) in terms of convolution. If the relation was of the form
L. sj (m)h(k-m ) + L d;(m)g(k - m). then both the summation are perfect convolution
sums. Since the translation quantity is 2m. each of h(k) and g(k) is not getting multiplied with
Discrele Wavelcl Tnoru;fonn and Relation 10 Filler Banks 105
s and d terms. For a given k, either odd indexed term of h(k) and g(k) or even indexed term of
h(k) and g{k) are panicipating in the summation. So, to make it a perfecl convol ution sum,
without affecti ng the filUll result, simply upsample (add 'zero' bet ween each lcrm in 1 and d)
and do the ordinary convol ution. This can be visual ized as shown in Figures 6. 15 and 6.16.
0,
I' ~m)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
d,
1I, I ~m)
,}-,
f, ~ ~)
0,
f, i-' 10(111)
dH
t , i-' ~m)
d,
t , I-< om)
" "
" "
" "
106 Insight into Wavdets From 1l>rot'y 10 Practict:
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
The same result would have been obtained by upsampling !s}(m)! and !dilll)! series and then
convolving with {h(O), h( l )! and {g(O), g( 1)!, respectively.
Figures 6.17 and 6. 18 show the pictorial representation of the compulalion of Spl(O) and
spl(l). Note that the result is same.
.J.O) .,(0 'P) ',(3) ,J.4)
T ,0,
T
"" ------------t------------------
11,(0) dj l ) dP) liP ) 1i,(4)
I
~I'
T
~o,
' J*I(O) ',(O)Jo(O) + d,{O)z(O)
0.
dJ.l)
d,(2) d,(3) d,(4)
H
8'(1) 8'(0) '/*,(1)',(0)1(1) + <1,(0)8'(1)
Wave Let-system analysis filters Ih(-m). gC-m)) decompose the signaLimo frequency bands and
wavelet-system s.ynthesis filters Ih(m). g(m) reconSlruct the decomposed signal back into the
original signal (see Figure 6,19). Set of such system of filters are called ]M:rfcctly matching
filters . Since hem) and gem) are orthogonal (in the veclorial sense) and analysis filters are mirror
image of the synthesis filters. these filters are also called quadrature mirror filtus or QMF.
1~ ,
';
" t,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
h(-nI) h(... )
" t,
Sj . ,
';
&(-ml
+' <fm)
In all the previous di scussions we have convenie ntly assumed that oLlr signal fit) is an eleme nt
of space Vitt and {.I'}tt(k). k I: NI are the corresponding scaling fLlnction coefficients. There are
tWQ issues here.
How do we know that the signal actually belong to space spanned by scaling function
of our chosen wavelet system?
Secondly how do we compute (sjot(k). k e NI?
The answe r to the first question from practical signal processing point of view is that
signal may not be expressible as linear combination of our chosen scaling function exactly bot
we can very closely approximate it. We can think of this as the projection of oo r signal on 10
the space spanned by the scaling function and its tran slates.
Given the closely sampled continuous signal, though we can compute the .l'jl-t coefficients
usi ng Eq. (6.5) for a panicular j + I. L1sually we assume thal sampled sig nal value itSelf as litt
coefficients for some high enough value of levelj + I. As the scale leve l j becomes high. scaling
function gelS compressed and approaches a pulse signal of unil stre ngth and therefore sampled
signal itself be<:omes scaling function coefficie nts. This assum plion is not 100% right but for
IT\OSt of the practical application this does not matter much. Once we have sit] coeffidcms, we
proceed 10 find out Sj and d j and so on. Also note that. through the synthesis filters we can get
back sjot coefficients which is ou r original signal.
In this section we will study how filters designed with van ishing moments imposed on wavelet
function behave when applied to digital signals.
108 Insight into Wavelets From'J'heof)l to Practice
For 4-tap Daubechies wavelets we imposed two vanishing moment condition s. This
imposes two l;onstl1l.ints on the filter I;Qeffidents. 1l!ey art:
h(O) - h( l) + h(2) - h(3) = 0 (6.22)
Gh(O) - hO) + 2h(2) - 3Ir(3) = (6.23)
Consider diocn:le polynomial of o rder 0 as inputs 10 4-Ulp filter bank.. Let it be an infinile
sequence ... , ... ,1, 1,\, I,I ,I,... (Figure 6.20) and assume that filter coefficients satisfy conditions
e:l:plained in Eqs. (6.22) and (6.23)
!t(II)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
11 1111
g(lI)
Let us see what happens when the signal is convol\"ed with {h(k) and Ig(k)l.
We have.
{h(k)} = {h(O).h(1).h(2).h(3)}. {g(k)} = {-h(3).h(2).-h(I).h(O)}
and the output seq uence when convolved with {h(k) is a constant sequence with each value
equal to L h(k) =,fi but the output sequem:e corresponding ill filteri ng with {g(k) proouces
a sequenl;e of uroes. This I;an be visualized from Figure 6.21.
-11( 1) -lI(3)
. ' IG URE 6.21 Coovolution of consWll l.ignal with waveLet function filler IrlkJl.
Upon convolving, each term is h(O) - hO) + h(2) - h(3) which is uro
Consider pol ynomIal or order I, assume the signal of type 0, I, 2, 3, 4, 5, 6. 7
Consider Figure 6.22.
o J , , , ,
~O) -11( 1) -lI(3)
FIGURE 6,22 Convolution of fi~t ordc:r polynomial sisrW with .....""'Iet function filter [gCk) J-genellllion of
one tenn.
On shifting we get the next term in the convoluti on (see Figu re 6.23).
o 3 , , 6 7
'" ""
-h(3)
nGU RE 6.23 Convolution of fin;t onkr polynomial signal with wavelet function filter (g(/:) I~JIlencration of
""",her term.
We have h(O) - 2IJ(I) + 3h(2) - 4h(3). We have a~sumed that our fil ter coefficients salisfy
Eqs. (6.22) and (6.23). if that is true then h(O) - 2Jr( J) + 3h(2) - 4Ir(3):: O. This follows from
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
n GURE 6.24 Convo lution of firn order polynomial signal with wa""let function filter \I(k) l~&"ncration or
a third term _
SUMMARY
The recognition and the study of several phenomena strongly depends 011 the resolution at
which the analysis is carried out. The avail ability of an efficient way to rep!"esent the same
phenomena at di fferent resolutions is. cherefore. very important (from both a mathemalical and
110 Insight into Wavelets-From Theory to Practice
Burrus, C.S .. Ramesh A. Gopinalh. and Haitao Guo. fnlroduCfion ID Wavtl..1.! and Wave'f..f
Trallsform~: A primtr. Prenlice Hall. Nl. 1997.
Mallat. 5., A Wavel ..t Tour of Signal Proc..ssiJlg. Academic Press. San Diego, California. 1999.
Slrang. G. ;llld T. Nguyen. Wave/tU and Fift.., Bal/ks. WeUesley. Cambridge Press, 1996.
Vaidyanathan, P.P., Mu/tirat .. Sysleltll and Fil,.., banks, Pn:ntice Hall. Englewood Cliffs. 1993.
Vetterli, M. and J. Kovacevic, Wav.. I..1.! and Subband Coding, Prenl ice Hall, 1995.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
INTRODU CTION
Mostly. enginl."Crs an: onl), mle"'''c</ in tile propen;., of fi ller< a"-",,,i>led with ",aling [WK'.;un
.. Ill "' ..-c l", f"n"';oo nuhe. lh'" Ihe funenon<. 1I,,",'c"er. lbe prol"'nic. of ",",'det and "'"Iing
functions "IT ,-ery I"'""nanl in field, ,uch as wo,-de' based ad'l"i,'c pani.] difkrcnli., cqu31ion
soh'.", Further. lis 'Iudy npc"' way f(lf new funclion >P""""- The Daubecn ic< scal,"~ fu,><"oo
is hIgh ly in~gul ... \Jul Slill. 00"-.'0' funellons and .n.ooln ""I)'nomiol fUnClion, ca" be
C>1",,,,,,,d "_,;ng the D.uil<chks ,.,ahna fune,;'m and ilS ,",cgct 1!'lln>I.s. Tu <km""""'''c the
,..me in d""room. "c need '0 nnd ,he ""lioB f"nclion ,'.I"e 0' .11 pom" "Un high pr;.i"".
D.ubc<'hks and L'gari.sl.l de,,,lopc<l on algo<nlll" 10 ""mpule sc.h ng .nd w."ck. funClion
,-.Iue, 3' "ny J>O;I\, '" Un arbitrary pr"";'i",,. Ho"-ewr. Ihe proof of con"ergen ... requIre. be .. y
uS<: of malhe""'ic>.
In Ihi. ~hapl<r. we will "",,,,nl 3 .imple and ,"I""h'c nplan."on for the Daubh,c ....
La,,,i lgOfilhm. FUflhc,. lbe algorilhnl Call be ullIized '" demon .. ,,, .. 11>0 inlrigulng PMMion
"f Un ." properly of IIIe sc.lin~ f"n"'ion which was explained in Ill. lasl ch'pler, Th" P<orcrly
,,,le., Ih... Ih~ "".hng fun,~ion II(x) "",,,,,fl lhe ,d.. i"" L ~' + n) = I. JC E (0. I). whe ... n
i, .n Inl<ger.
I D.,brth;'. I . .,..j I. L>pri>o.. l'I.u-o<"'" wff.... """ <q ......... I fui,,,,,,,, _ ' ...... 1 tr3'lui')' uf ...,,"'~ .. ~
SIILII J Mu," Anal. 21(SI. 1J8S..1410. 1991
'O"bhie> I. "'"' I Lapr .... Tvro-",'" ~itf...n I"""""" 11 Loc", "'t,""ily, ",flui" f'I"'luC1' of n,",.ne..
on<! f,,",taJ.~ SIAM 1, MUlh A...I. 2) 14,. IO.ll-I01'l , 19')l,
'"
112 Insight into W",'eltc'<-r-r<:>m l1>>ry to I'r.e'-'co
T he", i. no mp l,dt "I""'''i'''' for the onhoi,m.1 ",.Iing fUIK'IIon f(x) .nd w ~ "d", fu"""i..n
'f(') (excepting )loar). IS ,hey Me ",1.let! by nleans of tile ",r,""nlCnl ",Ioh"" Ihol .xpresse,
,(x) ""d ",(x) in lermS of ..,akd and " ... ,Iad ve.sions of w,df. There are ",rum m""',,,l>
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
whkh .Oow u' 10 ,,"olual" Ihe ".Iues of 11>< "".hlll and w"'del funClio", "' dy.dk (in,'.""
p""'ers of 2) poin ... Th.", n'hods i"dude 11>< C.,.,adc . 111<"","'111 (dir,,<,, e,.IU 3l1"n 3. dyadk
... lIon31 p<>in~), ",,,,,,,,,,,,. "Ppro"m.,jon method ... d ., ubdi,"ion "",,",mc, How",'." tIle i.
only 0"" melh"" ~"""n lill dalc, which enob les the <akulation "f the scaling and wa .. ele,
funell"" .alue . . . ny poml, 10 .ny tlcsirc<i .ccura..'y. Le. uS get Inlo the . lgo<illllll.
For 11,,, D,uOcchlcs 2-band system, with N = 6. we h.... 'he dil.lion <quallo,,:
".,
1!(1) .. ,fi ~)(~) fl21 - k): 0 s ,< N (7. ))
~O)" J2lhtO)' f(O) .. h( I)j!(-I) .. h(2ki'( -2)" h(3)i'( - 3) .. h(4}jt\-4) .. Itf:'lki'( -:'I) 1
."
,Ol .Ji
h(2) h(l)
h(4 ) h(3) " "hlO)
h(2) h(l) h(O) 9(2)
il(3)
.',
i.< ..
" " " h(5) !t(4)
."."
<11(0) = To '1>(0) (7.2)
4>(0) _ 1'(2)
OJ'
."
Equ.'ion (7.2) impl;e> lh .. <1>(0) i> an cijl<n >"""or of m.tri. To rorrcspondi n ~ to an cigen
".Iue of I,
Thu .. finding function ,.Iues at In'eger poln .. reduces 10 a problem of findIng an dgen
,'cr",r of T. eorrcspoodlng '0 an eigcn >alue of I.
For hal f Inleg .. ,'.I ues. app ly!tlB lh~ d,lallon equ."on. we ob'ain'
0(112) = ./2[11(0) ~I)+h(l) "(0)[
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
o(S12) = ./2lh(l) ~4)+ 1,(2) 1(3) + h(3) 1(2) -+- h(4 ) ;'(1) -+-h(S) 1(1))[
I ..
Here there is p.ttern. l1Ic biMry equi,'akn' of '~ Is 0.01. Hence. 'he cquotion ""''''''P<'nding
'o'~ .I:>ov~, can be Inlerpreted as ",,<!ucmlally mulllpl}'lng m.tnc." T. ,nd Tt, rl<p<ndtnB 011 'h.
dig"" in it> binary cqu;,'akn' being 0 or I.
AlIain we C'n ..now that,
41 (~) = T,' 41 ( ~) ~ 1; T, 41(0) (note "glin bin ary equ,v.lenl of -11 is 0, 11)
To bc ,,,pilei,.
11 4 Inslgh' into W",'eltc'<-r-r<:>m l1>>rr to I'r.e'-'ce
Following 'he .b.:",o procedure. S'r:ang' h.. -",,,,,,'n ,ha' 'h~ d,la"o" "<Iuat;"" in ''''' discre'e form
Cln "" 'Hin .. n as:
<1>(1) To <l>(2t) + T, '<1>(21-1/. 0,.; 1 <; I
E (7.4)
.,'
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
\1(1 + 1)
<1>(1) K
\l(1 + N-2)
""'0,,,"'
Th" folio"" rmm Iwo foe'"--H", ITor = ",.trix wilh all colum".", <1>(0)
Lar:eSl Eilt"" ,. Iu" of To m'lri, " I .nd corres""ndmg E'I"" "cClor is <It(0)
Each column ,u", or To " I ."d. heoce. b"'e the p",,,,,nr of """h .., ic mllrix li~.
M... ~o, main tram,Uon ",."i>.
The con"cr:eoc< ",le of 'h" " qUite faSl,
Th" Ih~'OI'Cm impl;", tha, Insteld or compuling Eigo" ,'oc{o, of T.. we may n'"I'ipl)' To
'bdf wi,h T,,_ a r~w ,ime,. ,<> ilo, m.,ri. wi,h .11 column' ., <1>(0).
Example: Compute <1>(3/8)
<l>(31l1) = T.T,l',4>tO) = a,'e'''l:e of {he column ,'eClors of T.T, T, ~ ToToT. TQ
Wc :on- '''''in& .,'cr'ge of "'~ column, of ,;gh' hand si<k m.. ri~ bee>uS<" T. T. T. T.
pnxlu",,~ , llIaln~ wllh ideQllcal ('{Iiumns ",d when this I< leflmulllphed wi'h T.T,T, mU>l
produce. mm';. ,dlh iden,ieal column . Sine" Ion b""d side is ,'c clor We ,like 'be average
of ,be colum" ,,,e,ors whicb also will reduce eomp", .,)"" . 1 crrur.
Ul U.' now mm'e '0 D,ul:>cchieo-Lagari... IRori,hm. Giwn a numl:>cr.< sIXh 'hat 0 S.<
< I. upress .... s a binary numl:>cr uplO' d<>in:d precision. say 20. UI .' " O. dId, .. <1,.,. ",h<re
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
each <1, 's 0 '" I, Then b~ Daube<;h,c .... Lag.ri ..' algoritbm. we get
~ .\)
jilt + XI
<1>(X) ..
"'1"-2 + .\)
NOle lhal "'" ",,,,It is I ,"octor cOn>i>ting of N - t funehOd vllue . each al unil di'lance op.n.
Funher. the Sum of to. ,"oclor element> of any 'I>(,T). 0 S.<" I Is I acconlin~ 10 panilion of
unity ,h.""",,,, discus.s.:d and proved in Chap'.' 5,
D.u"""hit, and L"lIanas took "",und 70 page> ,,, pru,,, ,bal 'I>(x) w ,lI ntoootomoally
c""ve'g' k) i" actu.1 ,. Iu. M"", 11(.) ,'.1"", a..., ,r.... tional and hen, ...... will ~'" "nly
appro,imatc ,'oltlC . TO. only problem Is lhat to gc, 1110..., KC"racy. lhe .. luo of ." need, to be
e,panded with more binary vol" . nd ""nee we n><i m"..., multiplication.
Now a q.tslion arises", to how slloll \I..t rind ,h. scoli"g fU"ClI"" ..lues beyond I. say
.t 1.2 . SIO<'<' ... occd> '0 I>e I:>clwc,n 0 .od I... fifSl wo "rip off in l <~< r parI .nd Ia~O ""Iy the
<k:<;imal pa". So he", ... i, 0.2.
And .... e nnd <1>(0.2)
jIIO.2)
,.,1. 2)
Rut. <1>(0.2) E PIU)
PIN - 2 + 0 ,1)
From lhe ""Ull obtained we lake a .. cond ... Iu. from li>c "cC1Or ..' lb. 'n,"'.r ,,hith i_ \H1.21.
Fin.lly. we an: ",ody for for",al ""scriplion of lhe algorilhm "" ~i\.n in 11. 21 .
d yikl (x. n) dcnolcs "'" fI,.". n digllS from the dyat! of -'. let h" (h(O,. h(I) .. h(N-I)1 be
"i. can define IWO (N -
. , ''1'(101' of scali"R fun"';o" ",,,,mcients.
Ih~ Th~n. I) X IN - I) mllrice~
~;.j"N _ (7.7)
'""
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Equalion. (7.6) and (7.7) gi.,. I moolic closed (""m "'I'f""ion for genera'ing malrh tlcmtnlS
ofT~ and T, for any .,en N (nulIloc. of taps). Wc ,-on e ... il)' ,,,,Hy 11 '0 oc true.
~. O.J .. 0.010011 .. 64
"
7.1.4 Implementalion ill Microsoft Excel (or Any Spreadsheet
Package)
For 6.,ap~r.cc", 'H tan &<ncratc binary ,'al ucs as fo llow.:
Fune"nn bina,y(o "-' d"ubl<. n ... tn'eg.,) . . .-.ri.n'
Dt:em $ as in'c~<r
- 1111(<
If" > '5 Th<n
Prin' "Ou, o r RanKc"
End If
r= ~ - J
Fori= IT"n
r z r 1
If c >.. Then
t = c-
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
.1(i) .. I
El ..
1/(') = 0
End If
N", t i
I~gure 7,1 ""'ws an excel !'lot "f 'he Daubechics scaling funell"n . The ~mirc a1lO'ilhm is
implemented in Excel as a fu"clion. The .,:comp.. ,ylng CD contain' Mic"'''''[1 10,,,,,1 sllcelS
... ilh dct.,!; of the VBA prugr.",. U!iCd [0 d" ... ""a!ioll .nd " '.,'<le[ fun('''o n , of different
support.
:. .' ! .: .
831
1111 In,lght into W",'eltc'<-r-r<:>m l1>>rr to I'r.e'-'co
Subdi.i.ion ,;cheme i nlCihod uS! in rompuler lraj>hle, (or interpol.lion. Th;, i<k. ;, .>cd
~orc in plotting '-:11 and Il'(r). The idea i, .imple. In 'he .n.lp" ",age a sign.1 in V -", '"
<'pressed ""ng !he ba,... of V and Wt And ,n tl>c 5)"nlhesi. MOgo we 110 (",m lower .. .. Ie V,
and WJ '" higher .cak V-"I' Th , in 'he ou 'I"" of ... a!ysi. "age. if ,,'e hO\'c inglc codflcien'.
I. ",prc""nting' ilit). (In the!ow p. filler side) and no """fflC;"nts on 'be high P'''' ftlter 'ideo
lite .)'nllt<>i ...age will lake: (hal I .nd con,'en inlO <<>cmei"nl corrc.'f'Onding (0 bases .,0(211'.
If the ",,,,I, b fed b""k 10 ,h. synthes;s m'er (~)W pa..,,) all"n. the ",.oh i , coemcien"
oo ...... ""'ndini; to ~4')$. If
Ill" fe<tlb . .. k .. oontinu<tl the I"<'UU will be IIIe "'p",>cntation of the
ori~i"al I ~1) In l<'rms of eXlremely small ~al"d wrsion of f(tu "nkh for al l pr3Ctje.1 P"' fWIfoeS
ean be consld<:rtd as <id", p"b~. Thu,. !he outpUt bcoolll<s a .. mpled version of ~J).
The .y>lern for gencrnting 41(1) ,s bo<i<a lly Onc chonncl tbot we uS< III recoo.'lI"II<1 the 'ign.l
fmm w.,-,,{e, <".me,,,n,, sJ ...d d,. We "'-,urnc. " pul", of uni' ""'nglh " &"'<n '0 the ').""m
"' Inp"t (S Flg"re 7.2). Aflt:r "p""rnplin8. it is oon"ol>'od .,illl fllte. h(n) Ind Ill" output is r.d
1>(.)
hock as input. Repe"' tltc up"'mph ng eonmi u,ion proc<,-' '0 IIIe desired number of ,im . The
nn.1 outPUI is cqu,lIy """"d .. mple' of ~t) in 'he range 10. N - 1I f,,,.. ",."nll fune'Hln W""
N lOp,. M"1Ii p lie'lion factor .,fi can I>e .h",""'d in the nUer itself. follmoing i< MATLAS cooe
for generallng D,ul>ech",,' 4-t.p "'""'eJel ...,.linll funcllon. III this code .,fi i, absorbed In lilter
eocfficicn"'. The OUtpu, of IIIe MATLA8 oo,k is >lto"'n in Figu'" 7_;.
,., ,-----,-------------,
""
""'
""'
x _ l i n spacelO , J ,l e ngth(d)l:
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
plot 1><, dl
Ji
t' ..,
Ji
- t, ..,
output we oo"'ned .fter ",,",'o lu,;on wi,h g(n). ~OC, 10 u"","'pkr--romolution loop wi'" h(n)
I.. filter, Th. number of ioo{>inl """.nd!; on the number of equity ",,",,cd sam pled paim, of
1/'<1) that we ""I"II'e in the .. ngo 10, N - 11. Thc following MATLAB rod.: ge, .. ratc "'(1) of
D."bcch ,,,. 4_tap w",'ck, func",,".
h _ 10 . 683 l.183 C. 3l? -0.1831 ;
9 - 1-0.183 -0.317 1.lB) - 0.683 1;
p~i _ l ,
",'
x _ lin~ pac .(O , J, h n'ltn (psi)) ;
plot (x,psi)
F)gure 7,5 ..now, Ihe OOlpU1.
,
"
,
,
I
",
,
\ 1.5
" "' "
w. will now d,scuss ""'100d (or Ill<: C"'I5IlUclio" or sc.ling rUnCI'O'" using IIcc"".i ..
apfI<0,;n\.6on, U.ing "". nng fu""tion. <>bt1l.,ocd
In th,. way. the ,,",.odated wO\'ckt fu ....." ;""
cln be cO"Slrucl.d,
0-'
11>10 "ppruxim"iotJ metltod" b.d on tlte rccursiYC ["mml. ii(/) _ 2, h(k) .n1(21 - k ).
,-.
<l>( jro) ~ * 4>(j;) lIi~l) (7. 10)
(7.1 1)
122 In,lgh' into W",'oltc'<-r-r<:>m l1>>rr to I'r.e'-'co
A. i --> ... ,hi:: faetor on the left <l>( jrol2'* '1 ""m'erges 10 the ".Ioe <1>(0) .. I (;ince the
The hm" of RHS quan"'y e",," and is 'ho rour\cr ,nm,;furm or. "on,inuou' ,;caling fune"on.
Rewri'ing lhe 10" "'Iua"O". Wc h",'c an i"finll< produc,
I ~,,' I '-1,' I "
<l>(jw) ~ Ji. 1II~'- ) Ji. If(~ - )Ji.If(~'" ) (7. 13)
Ikrc. I/ k~' ) "juSl 'he F"""", ,ra",form or 'he filter "o<'fHd""I' IIt(t)) .nd
lb. Fouri.r 'nmfonn 0( ,h. up .. mplc'1l "'queoce IIt(!II, That I.,
H(~""') ~ Fourie, Ir""sform of 110(0). O. h(I). O. h(21. 0, h(N - 1)1 (1. 14)
AI.., ""le that
~ I/(ri"" ) ~ Fou,k, "",,,form of .fi 11,(0). O. 11( 1). o. 1t(2). O. hlN - I) 0.15)
,
Therefore, in lb. Iln", do"'''n lb. di"-"'INtn'e ,'.Iue' h"',n). n 'hodo) can be obtained in
,.,
Ihe ttn ' ''',"Uon step hy ;r" ld ""n"olu,ion of 'he dyodie u!".mplcd 'mpul", re"l'm",
(7.16)
Exa m p le: In 'hI> e .. mple we will de.1 "'tth 'he construe""n of 'ho ..:aling funct;,,"
",-('t>nlin~ to Doubeeh,e. N. 4.
TItc """ffie,ents hi!) are a. follow>:
h(O) ~
h(2)., .1-.jj
M
1- Jj
h(J) ~ ~
M
lJy ron>idcrin g the rac.o, 2'12 wc ob in .he i "'I",I><: sp<>nsc from Eq. (7.16) .s:
hoolnl [0.683 1.183 0.111 ...j).ISJ) (1.18)
h",lol (0.683 0 !.1M3 0 0317 0 ...{/.I!IJ) (7.19)
'""
A convolu'~", of ,he", 'wo ",quence . re.uh. in Ihe J ,,,,,e'o"me ,.Iuo, of 4(1) .... h" '(n) ",h,ch
i. '/IO"'n in n~ur<: 7.4.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
The sequen'" ho,ln) has ,he ...,nle n"m~,ic.1 v. I""S", ho,ln). bu, ,hre. ~e"",. are in." 'rled
bel"'''" <.ch of Ih. non lero ' olu,",. Tile '.'Ull h" 'ln) is ob,aine..;! by convol.-ing h'''(n) ",ilh
ho,(o). Figure 7.6 ' ' '>w'
'he h"'n) value.,.
,,
,
,
o ,
o. ,
,
, , , , , , , 10 I1
Alloough Eq. (7.17) .uggesl' Ih. <le,i.."o" of a ~ep fUnclion /"\1) from Ih. impul ..
respon><. il i~ SCcn th thc i.e",,,,'< appro'im'lio" of the "'aling funclion can alsa be pot,,,,)<d
hy line., in'erpol.,;.,,, "",wccn Ihc .'.1""5 of ,he ,mp"l .. re""" ... h"' (n).
The impul'" spoil" h"'(n) ha, (n - 1)(2'" - I) + 1 coofficicms. "'hore N is Ihe number
of eocfficknl.!i or hln). If w< <",mp'''' the runnin~ i"on ,,"h th< d iscTCI< ,,,ne >I.ps
,~ -- "
2" ,
(7.20)
". " ,ill gel Ihe fi~hllim. ",aling fo, Ill<: linea, inle'polation ",o"liOfted earlie,. It Ihen rail s in
the inl<,.,.1 0 ,;: ,,;: N _ I. as does.he scahng fune,;on l<) be "ppruxim"cd. Figur". 7.7 .nd 7.8
"now Ih~ i,orall1e "l'PfO"'U'"'''' for i = 3 .,,,t
j = 4.
,
'" Insig ht into \\ ",'. "C'<- -r<:>m l1>>r ~ to I'r .e'-'co
"
U
,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
"'
.
0.2
,
I. -,.
I
-0.2
~ ,
, W ~
...
..,,
"
-0.2
~ ,
, "
,
"' "
The plo. ha> al.,,,,,, (o.,wged willl (our i'erallon' (IOC. ngur. 7.9).
,., ,----;c------------,
u
..
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
"'
.,
.,
-<l.2
~-
.., , "
,
"
n r.UIII: 1.9 111< '''''''"''''''. plo< "r fI' l ,/1<, ro., 'I<"rul""',
SUMMARY
WaH,kl and ~a li ni funclion. in general arr no. 2;,'cn a, "p lieil fURmon,. Lilc fmelllb. lhey
are .'1'"".",..-<1 a, "".Ied .nd tr"",I.,.d ,w,ioos of i1>df. Computing ."' c' 'alu", "'qUt""
ilerl,io" and special algor,lIlm ,,!>Ceded fot the somc. In this Ch.p' c< Wc discus.scd Daubechie!l-
I.all"" " alll(Jl""ithm in de'.,1s ""d srn''''<"<i how ,he alll"";thm con he implemen'ed on Micm",ft
E,,,,,i. If on. " intCfC.'I,-<I only in plonin. Ihe functinn_. i' is cnOUllh to ""mpule Ih~ funcH""
,olu., .. dyadlc points. Thi> chapter discuS>cd 'wo m.tho,1> lII.,
can co.ily run'pule ["n(,ion
,- .Iue. 01 dy~d,c po,n,,_
EXERCISES
7. 1 Whot i,d,"''''tc fonn <>fDi l.,;"" <'<Iualion'! E,prc'" oJ> (~/16) on lenn, o f To- T, .nd
4>(0).
7.2 In ca", or
3~and ",",'.Ie' .y>lem. how ",,11 ,he db"",," diitui"" "4U";''' 1",>1< !ok" '!
Huw many matri"", arc ;n,'oh'cd?
7.4 U~mgD.ul>cchic!O-Laj;.ria, .Igonthm find the ,'01"" o( '1'( 112). '1'( I) and .;0( 114) (or
~.(.p D.ub(.",hies "....... let system.
7.5 Writ" I MATLAB code (or Cl sca<k algorithm (or ftndtng ,-.Iues of :t) a( dyatlk
ro mts ,
7.6 Implcmcm Daul>cchics----L,g.ri .. algorithm as an E.cci ,,"orh"..! functioo!.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
BiorUtogonal Wavelets
INTRODU CTION
"11 of us kMl ..... bout onl>ogon.lay. Onhogonallly of "0010'" and fU"CIl"nS allow us 10 "'J'f""""'
general ,,"",or. ,"d funcrion, In le,,,\> of set or eomplc onIIogon.1 ,'cc,ors and (W><'lions.
""'1leCn"cly. BUI boorlhogoo.hl)' "a """coP' which. e.g'""",, ",~ "01 much f.n" I.., "un. In
.i~n.1 procc.sing. cs.p<cially in ",",'elels. IIowe,'er. it i ro"pI whkh nobotly <an igllOk.
Uiunl>ogon.1 w',cicl!l .R: the working horse behind m.ny ronllnell',"1 app lications li ke finger
prim im.ge con'p""5Sion.
Tt i, &"'0" by
Uy ,.k ing rolu", of T', kl uS rorm ,'"",ors <I, .nd d, so Ih., d,,, (I. -11/3) and
d, = (0. 21./i ). (/, . ",1 <I, .., n'" Of' hOZOl\.L 00' ... ~)em."" of > ba,i, .," '~cy can ',"n ~'.
Wc <all Ill. b.,is 0<, (d ,. <1,1 os du. 1 b."s 0<, rorrcspooding 10 II>c prin,.1 b'>I~ 0<, (p,. p,l.
Le' us I>OW d"~01C ,he in .. , produc' (or dOl pro<' uc') of '''0 ,',<'0" p, .nd p, .s (p,. p,).
",
Since. T " ., I. (Idcn~.y m.lt,,) link (on>idorahon will >how "'at
(P . d.) = I ICbc'C~' (1.0)' (I. 1/,fj) '" 11
d, and Pl is orlllogonll 10 d,. More,w"r. (1',. 11 ,) = I .nd (Pl' 11,) = 1. Wh~ w. call Ihe h.,i<
<et lS(lflmal .nd duo!"1 Soon W~ will.<ce SOme '"Cl)' inl<"'Sling ",1.lIon< belwccn Ihe<c 'wo ""L<
lu "'11.'f)" Ihe"" .d!'-'Ct,,... Before that. le' 0.10 il> _"en .. on In n dlme",iu",,1 <pal".
Lel 11',.p,. . .. 1', ) be I so:l nf n Independem "cclors (each an " tupk) Ihal form IhI:
J>rim.1 hasi, set. The.<C ,'10r.; ore nUl orthogonol anlUng ,,".,,"<che< 001 'he)" . re ind<r<nocnl.
By ta~ing ,,"cso: ,'o<tors as row~ of I matrix. we fin(ilhe in'cr .. oflh.- malrix and In ~ e ils column
"CCturs b ekment< of du . 1 bosh .:1. Le, il be ld l 11,. d , l. Tbe<c ,'10r.; lollow II'Ic
foll""'n& t"'o relauonshlp"
1',1. d l 1..- .,-ery j., i
(1',. d') " () for ."cry j., j
(1',. d,) I for every i
Cun>llkr the t",,,dlmen,iun.1 empl" we haw JU." ." ,en. Le' U' try lu .'pre" ,"<'Ctor (I. 1)
using primal basis..,1 11'1' p,].
(I. I) '" aI', -j. bp,
.. here a and b arc IwO ""alar.,
,,'
, ~,
~" ,[ ~) ~
= .. ,
Th .. ,,(1. 1),,( 1 _ 11,fj)p, -j. 1J,fj,.,.
.Ji ~"
,
,--
"' .Ji
Can w~ gellhe'" coefficients by proje<;ting Ihe ,'",'or (I. I ) onlo 1' , and p,~ A liltl.
,,","kinll w,lI ,how Ih ...... will ",M get " bee.u", 1', .nd p, ..~ nUl unh"ional. Nuw.
Projecting (, 1I onlO 11, . Wo g~t
(1. 1)'(1.-1113),,1 1113,,0
ProJccting ((, I) 0"10 11,. "'. !!ct
(I. 1)'/0. 21,fj): 2113 ~b
Th .. j, lhe cru. of bionhollOOahl)'. Now. we .re ",ad)" for e.reodlo!! Ihl< ron"",,1 to funcli""
""ac.".
8.2 BIORTHOGONAL WAVELET SYSTEMS
In ""n) fi lleri": ""plic."o", ..e """~ filrers w"h ,)m"lCuinl coefrrdcnls 1o achieve Ime ..
ph.se. Non" of the onhogo"ol ..... lcI ,)slems."""pl IlIar. h .. S) mm"I,;".1 codfoci<nls. DUI
H= is lOO inadequ.t. r",. m,ny pr"",i.,1 app lications. 8,onh",,,.. 1 w",.ic, .<y>lenl t ... b<
<Ic.i,ned 1o to .... Ihis propcny. Th.1 i, O<lr mOli,ollon r",. <)e.<igning .""h ... .,elel ,)lifem. DUI
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
the prier i. Ih .. non_zero coeffiei.n'" in .... I)"sis fllt.rs lnd synthe>" nllers arr nm ,he ..",c.
,.,
,-" ..
,) - L h(k)..fi~2' - k)
,
and its Ir,n,I.I05 r",.on Ihc prim.1 ""aling funClion b.>I. and com."SJ'Onclinll ""CC he V..
(8.l)
,.,
,-" ~(I) - L irV112~(2I-t) (8.2)
OlIld lramlale,," for,,, the du.1 ..,.Irng fu"",ion hasi nd rorrc>pOndin2 ",acr b< v". This n'.'nS
th .. ,) .nd ilS 1n""I.I<O.'" "'" onhugonal .",ong the",,,,,!>.s bul onhoeonal It) lran,l.tes of
I) (e>cepl 0"" b.,i, runClion as in wcr",. <pace) ",ch Ih.,
, - k) .1 j,(, - m) for .,cry *"* m
",hieh in '0'" m"ans
Li~c ..,al,n. functions. ""cle' fonClion also follow Ihe "".nng relalion ,iven by
In onllo~onal ... .,dcl S)",.",. <Pi'1 I. onltogonaL 10 \I'(r) and i .. InIn,L.te>. In biur,hog"".1 >yst.m
our requ"cm<:Ol is Lbal ')
be ol1l>ogonal LO jO(') ,nu liS lran,I'IC,,"
SImIlarly 11(1) mu," be o<,hogooal 10 V(I) and ios "aoslales, "Ill c.n be "'ri"eo "'''
ilk) _HI'h(N _ l _ p)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
"h.",, P is a pro!",r inlexer sh,[o Ihal enSureS <,nhOllooallly of ~I) and \1'(1) Tbo< shift p de!",nds
on whelher N i. odd or ."en. N I, 'he number of ,.'" of ,be filler.
Tb" foil"". from follOWing r.ct>:
L O(t) ""ed I" be onhoxon~ 1 ,,, ;-(1)
2. O(t) is n"de up of 0(2t). and i(f) re m.de up of ~2t)'
J. Traml.te, " r 1I(2t), and ~2Il' are hklnbogonal
4. Number of liP' for 1Xr) C.n be nen or odd
If N is ."cn. ooe possiblll,y of pi"""men' or fiUer "<>om"ien,s for N = of ("nd Il<nce
' ..0'1.... o( \112t) IId ~2J)s I, as '"""" h~re:
No""h ..
Th" mcam 'ha!
J J
2t - 2)';:(21 - 2)J1 + 2h,g, 21 - 3)"<21 - 3)<11
o
o 0 8,~h, jl~ -hl
"
i.=h, 8l~ - 1Io
n.c"" ..,/a,h'" plattmoolS or fiil <""melen" ensure ~(I) J. ;;(1) or <"<lu,,'.lcn~y ,h.
double -",I(, onho~oo.l i ly o( filler rodlicien'.' an: .. ,I,r,ed.
If N i; odd. Onc f'OIiS,h, iny o f plocemen' or fll[c, coefficient> If'" N.~) Is ~> .!>own here:
~
X, "- h. i, " i, "- h, C, " h,
0
i. ~O ~ ~ h, ~ ~ X. ~ -110
An.oli>.:r f'OliSibilny is
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
~ 0 0 0
go .. 0 g, ".0 C,"0 Cl .." -h, R "
h, X, -h, i h, X, -110
The dem'a,ion of 'he rc,ul, rollow in lIIe .anle line as In onho,,,,,.1 "'3,,,1<,,,
Simila,ly.
since ~(I).L \I"(r). \11(1)" I I- I)' h(N - k - p) f{ll - l) (8.6)
(S.1)
In onllogonol wnclc< .)''''el1l 1"1') is ",-lII"11ooal '0 ""y ",aled ve"lOn of itsclf and Its
l",n,latcs. lh ~l is.
The inl"&,al is 'cro c' ''''p' when j '" 0 IUld " '" "'.
When j 0 IInd l " m. lIIe in'egral bcronle'
J\II(t) 'i(1) dl I
So In hK>nll<>gon.1 ''''ell'. lhe "'0'" ~ener.j equ""o" wnm:c'ing 1"11) and Ii'{I) ,~
Now. \\'. ",elude Ihe disc"",ion by PUl"nG all 'he rel>,lon' in b,oJ1ho.lloo.1 \\ av"lel ') "<111' In
0"" plac". W", ha,. ,he roll""'in~ relationship_"
L V "c Vocl',c V.
1. I'_,cl'ocl',c!',. -cf'..
.. lIel"<: I'; . 'pan(l"p(2',-k)) and I~ . ,p.nI2"~YI-k))
In orlhofOn.1 w",'clcl S)'5/CI1I we ha>'< VI onho~on' l to W,.
J. But in bionoollO",1 ~y'tem Vi "onho,on.' "nly to If'r Soo" larly. I~ "onhoK"".1
o~ly '0 Wr
4. W .1 rf', for oil *"*-}.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
/11) = ~>,(k)2"jI{2"- t)
SlII". 'ili) re "'" onhoil""'1. to lIel 'ik) "'e proje<lflt) on 10 Ibe ,lu.1 b..i, " , . (tl. Tb.!;,
w. ha,'e
jl{t ) .. L h(i).n~2J -k)
and
~t) .. L ii(k).n~21-i)
1M2' 1- il _ L j,(~).n;12(2' I - i ) - nf _ L ii(n).n;.:y" , - 2k - n) (8,10)
(8.11)
B)' suh~llIull"lI 'ariable m = 2. + n in Eqs. (8,10) "nd (8.11). "" gel
pOJI - k) E L, 1,{m _ U),JijJ(2 J't , - .. ) (8 . 12)
""pi, _ ~ I" L s(m - 2l ).fii\Zi+ l , - ml \8,131
"
Lelf{t) be lhc e"'menl or Vjo l' Then
SJ f k ) .. f [(')2j1'~:2'I-t)dl
"
~ L h(m - lk)'I+lf m) (8,14)
"""
Th< hmil for m <10",,00. o n whallnde~e" th{t) are <lcf"",d
Similar ly.
"
.. L g(m-2~)SI ,(m) (~.I~)
... 2l
For in' c'l""t3 1i,,,, or tbe ITsult. the n:odor <"n rofe o ch3pt<. on 0r1hog"".1 ... a,'cicLS. No ... , ..
100" . on 10 sy nlh~" ..
8.5 BIORTHOGONAL SYNTHESIS- FROM COARSE SCALE 1'0
FINE SCALE
In ,h .. ",c1l0" we "'l l try to 2"t rormu la ror ,.,,)nl '", coefficieot. rrom 'i anu d, coeffICients.
Le'/lt) be ~IC delnen. o r VI'"
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
" I
, 'll l )21" +f(2i I - t) +I
,
d,lk)2i".,-(2i I - t)
Sub>ti,u"ng 10 thc .bove cqu.lt"n. wc iCI
/1./) ..
.
L,'1.Ik).L hlm - 2t)~i., ... 1/1 + L d,V) I. s(m -2k)"., ... (I)
Lc. uS 00'" projcr. /1.'1 OOto nonn.ItLcd h,,,., i;j. ,.. (t) " 2'i' "" f(.2 i " t - n) 10 gc. If' ,(nl.
Tbi, gh'cs
'I. ,(n) " J[~ ~ hIm - '1 I k) H)'I+'''' (I) ] iJ"~ (I) dt
+ J[ ~(fN)~glm m'I.,. (IJ];j.,.
- (tldt
To undcr!'1and 'hc "'hole lboory bener. Ic' u, fir>J con,lder some bionhoBoo.' fil,crs and
COnstruc, corres.pondlng ",aling funcli"", and w.,clcl funetion .. Splinc ba.1ed biorlhog(mal
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
8.6.1 B-splines
B. ",linc< f","IlI sel of ",aJing fU""tinns satisfying the dil .. inn equalion wi,h binn,niaJ nllcr
coefficient.>. Ho".'.r. B-",h"o< olhcr 'hao Ihe lCrolh orde, B-,pho. (the flur fUl1-<,ion) arc 001
""hogoo.' to its OWn ,h,f... H<nce. '0
form wavelet .y>lcnl. bio"llo#"". ' ",. Iint funeti"", arc
u",d_
B.",hnc '" produced by coo,'ol, ing Haar ",.hng funcli"" " .. i'h ilself. To .~oid COtlru,ioo
in notalion (>phn<> are indexed by lit. unk, uf ' he polynunli.,). for the tinl. l><il1& kl uS write
110..- funCliOft q,(r) a,'
o ,; , ,.; I
Let N,,,I')" ~, _ kJ repro"'n" I""'_""C< of N,(f). wh,ch we know OIT ()"h()~on.1. Also
,,', kllOw
N I() " f>(2r) + ;<2' _ I)
Con",h'inll f>(') '" Ith i,,,,lf \1'0 lIet a IlCW fun"",", h.vinll "'nl ''''pe. Lel H_' de""te Ih..
fUn"non hy IV,t')
o ,.; r,;
l'
I
N,(r ) '"' (8. Ill)
2 - '. I S,s 2
The funcri"n i. <Jtnwn in Rgut< 8 . 1. Wc ca ll IhlS fu"",iun' ", 1,1>< uf unlcr 1. Tho ""ou'Y of this
sphoe fU'l<llon IS ,h31 tI ca" "" .xp..... d as I line .. combination of ",.led version of i,,,,,f .nd
iLS tranSIOlO . Flgu.e 8.2 illu"'3Io' lhe ("Oncop!.
That"
(8. 19)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
,
T"" scaling function """mei"",S arc 112( 1. 2. I) . NOlo {hat (I. 2. I) are binomial
"""md.nl' ,",'h,ch in lurn are obt.ined hy (xm,-olvmg ( I. I) with (I. I ). S,m i iorly. N.>I'I ..
obl.ine~ by cOfI'''''' ing t>o~ f"""lion ~I) (H ... scaling function) by ;,,,,If ] IIme'_ I1 is
"'luh'~lcnl to c'-"'wh ing ",:rl with N,ll). Thncfurc. Nill " gi,..,n by
,
,-I ' o ,.; , ,.; 1
N,(I) = ' ,
2( -21 + 6/-3) 1,.;,,.;2 (8.20)
,,
'2(1 - 6/ +9) 2,.;,,.;3
The tWO ""ak rd.lion (dilation "lo.,lon) fur ,his funetion is gi"~n by
The f,,'< conlponcnl. of "'i.
"'lua"on at< .""wn In Agur< 8.1 ",her< "'c orillin" funcbon
I~ !-ho"'n II.uh d'lsht'd lines and Ihe four ~aled and rr.n~I'led funell"""[I' !-h""," using !'Olrd
lincs,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
, ,
"
The ",,,ling cotm~ic"lS arc 1/4(1. J. J. I ), NOle. 8I\oin Ihall!. 1. 1. t ) .... billO.,ial
rodf",ienl' [com'ulw ( 1. 1. I) wi'" (I. I) '0 ge. (1.1.1. I )1. Nuw. we ".n gencrali,e 'he "".ling
",I."on foo- 'h" 'l'linc N,H).
- - - - -
Lel h(O). h(l). h(2). hO). h(4) be Ih. oormali,ed scaling funclion "",,(rlCknlS of jl(1).
(8,2])
Thi, i m.,~, Imp"".n' rda'~m ,h., dffide. 1'(1) ,
In Ih. o"h"OI1,1 n .... we haw
In Ihe 0I'1"OiOO.1 <... , Ih;, red""e, to Ihe weIHno,",'n bc' Ih.1 lbe lenglh Qf h h>s to be
."en. Equ"ion (8.2J) .1>0 impli.s that IIle d,ffercn<e bet,",'e.n Ihe I.nglh, o f hand h ,""," be-
.,'cn. Tb u.. Iheir 101l&lh mu't be both .'cn or l>oIh odd.
Lel us 11)' 10 'i,uollZe !iq . (8,2.1) and (8.14) using ngure 8,4.
~O) ~3)
iI, _ 0
iill)
'", '",
T" T T T, T iI,.4
N, _ I
N, 3
, , , , N,-N, _3
" W- To W- iI,-N, _ )
I"IGUIU: IU Re"",,,, po>ition,nl of >caJ,n. Folie, "",If";'n",
Tbc posilloning of sc.ling function codficients is 'cry impo,,""I. This decide. Ihe
run"rainlS "'., wc arc gOin, '0 pUI on ,lie """m";.nts. If Eq, (S.2~ ) is nol ,,".fied. W~ w;1I nO!
gd a r
.,jblc >OIulioll for our h(t) """fiicienlS. Many applica1ion, r<quin: sj,mnttrk sellinB
fun,-"lIon <""ft"'"ien'<, So le, n< rUI thi> coo"",inl fiISl, For .ymmcu)". wc ""cd h(O) = h(4 ) and
- -
h(l) ~ h(J). So tbe fi~ure can be ",dra"'n as s_" ill Fi~ure 8.5.
W< now .pply rondllions oo """fflcients, For f~1)Jl _ I. ,",'~ h.... Ih< condition
,,,
Bo<>nhoFi "'",'ok"
'"
hill ~2) h ( 1\ h(O)
T T T, T, T
0 ,
, ,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
0 0
)Ji J; 2Ji
"" iil1)
""
ii(1)
""
T, T, T,
0
T, T
0 , , o
2Ji J; lJi
l = -I w,lI al'" i!"'e Ihe .. me condilion .. g,,'.n in Eq. (8.27). SI""" we assumed het) are
~y",me'rk No mor< ""ndi,ions Can b< deriwd from Eq. (g,B).
NOIe Ihal we ho,'.
only lIIree unkllown, and we alreldy h..,. Ihr<e <:qU3Iio ,lS. 11"1 11 cln
b< <""Iy ,'cri lied ,h.,
only 'wo .., 'ndc!"'ntJcn'. Wc nero IJnc "IOn- eqll'''i"" '0 gel ""'qlle
","Illlon.
S",,,. f(1) .1 '1'\1). we h.v.
,.,
\1'\1) "
..
L (-I) j,(N-t-p)~21-*)
Tha, I> \1'1') fUOI""on <I<""OId, on h(k) and "'" quIre \1'1') ,,"ch ,h., f !"(') d' ~ O. ThIS
rcquire, '~>I I (- I)' h(N - *- p) ~ 0_ If you rememlk:r. ,hi, i. our "n;,hlni momcOl'
ror-<iilion.
'0'
Thu,. we h3>'.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
h(4) - h(3) + ';(2) - h{l) + h(O) ~ 0 => 2';{O) - 2h(lj + h(2) ~ 0 (8 .28)
, 0 0
R>rk" I. Ihll)+h(O).O
Th"_",, e<.juations Ca" Ik: eo"'y solwd U""lI E>cel ""rc.d_"'",,' ",Iv ...
n~ u r<: 8.7 >hOl"S Ilk: corresponding ", aling funClion. """"ti<:r fu""rion alld lII<:ir duals.
Table 8,1 >how. Ihe con,piclc se, of fillers. Note Ihe rel3th'c posilioni"g of fIlter roo:'flcicnts.
From rlk: lable il can bo ca5ily ""on 111.,. flhc", h >I,d 8 .re double->hlfl onboronal, Similarly.
h ""d 1I arc d"uble>hirt onbug"",ls.
, ,'
,
o ,
-, "
fi'
o , fi , fi
j ,,
fi --, fi ,, fi
-, ,
.f2 ,fi!
o , fi ,
Jl -, ,,
fi --, , .n -,, Jl -,
fi ,
, '"
0.'
0.' ,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
0.2
o --"----.:.--:---c.-------'!
()..'5 I..'l 2 '~"--~----,c---~,--~
,
" ,- - ----c,---- ---,
,
0
, , "" "'" .0 '" ,
, , , 0.'
,
SUMMARY
As W~ ~oow. > basis that spa,,' I spooce </0<, 001 h..~ 10 be ,,"hog""'1. In order to ga,,, g""'er
ncxtb,.il)" in 'he <08>11\1011"0 or ",",okt b.""s. wc rc"'n~d to rcluing the unoogon.lity
,,,,,dition and allowed """ .,rthogon.1 w",'<let ba",_," For cumplc. il" well_known ,h.,the Ha.,
Wl\'dct is Ihe only ~"""" wO\'dtt ,ha, is compac1ly ,ul'flOned. orthogon.1 ond symmetric. In
,nany 'pp!!",,';on,. lb. sy"uncll)' of the filler cocffidcnLS .. Mien dcs.i,ablc "lICe" r.",'IS in
hne.r phase of I"" ,ran,fer fu"en"n_ So. in onlcr to ronslrucl ",ore f."nhes of compaClly
"'ppor".d. 'yrn,nctri,- wovdCL> in this ch'p'er, wc did fO"'go the "'quircmcnl of ort/Iugunali'y
.nd. in panicu''', ....'. inlroouc<:d 'he so-called hlorth''II n,,~ 1 " 'a ... le,,. Th. ", Wc <kri'-ed
"'prcs>ion ror ",orlh<>go"o, .nalys;s and ,)"'heSlS, In 'he la....clion "'. sho",'e;j how
bio"hogon.1 w",<ict> e.n I>c eon,ltUctcd [wm B. ...,li ne wO\'ckl bbc<. Mure '!>uUI lite
con"JUClion of blonooion.1 w",'de. will I>c .'plained In Chaf'lcr 9,
EXERCISES
I _ I - _
f'ro,c Ihat h... (n) ~ "2lk(n) + h(n - 1)1. h... (M) = "2lh(M) + h(M - 1)1 Ikfine" I
I k(k) ~ I ii(k) = fi
L, I;(k)h(k - 21).0,
3.3 How m~ny 'anbltinl! momenlS ba'e tlte primary and ~almg functions in lile
biorlbogon.' coin .. family (in lit .. family ",.hng fUbe'ions .,,,,, ha,. vlnbhi"g
momcnl,) deftned by:
Designing WaveJels-
Frequency Domain Allpl'Oach
INTRODUCTION
In 'he pn:vious ehap,er. "c d<:lcrmil"dl so"'c ba>;e properties of !lie ,.:eurs;on coeff,cient. l/rIll I
to make "'" docomp",,,ionl...,con,,,uc';on "OR. Wc "",11 "'''''., lh"s< cood"km' he", and .1",
"'ph, . .. rhon, in "'I'm, of H(w) and I/,(z) . Along the "'ay. we .,ill an,,,,e, ,he q uest;"" of how
'0 fmd 'he Ig(l)) from IIr(t)l . Onc ,urh 1"',,,h,lny wa< Ili,-cn in Ch'p'er 4. bu, IKIW ". Can
g"'" 'he complcl~ am"or, Then ...... ,11 discu" wh .. olhcrdc,,,.hlc prop<nic. wc may wanl
Our ,,"",'ck,s 10 h,,'c. and how 'hey aro e'pR'<cd in 'mo' of jh(t). I/(ro) and 11,1'). T"Il<th.
'h" w,lI all"", U' to "rite down the '<I "r "'iu.ti"". wc "<>cd 10 ""h'c to find ",'.,-det< wilh
pankular p.oponi"" We wil l aC1.ally ",I". Ihese equation' for "I><: imponam s.pccia' cose. In
"'i, ch'pter, w< wil l al", disc"," an ,mJ>O".n' d." of ",... Ie, dcs.ign procedure <'oiled
I.. .... n... r ......'~.'oI.' d"'l~n ,
T he followinll <ood)'",n, are ",,<es.","y for 1h(kll and 18(t) 1 to protluce a .-,lId ""'ling fu""lion
and ", . .. del ~y>lc,": Some of ,he"" ",,,,1<> w< h~ alr<ady Sn in Chapta 4.
I. Unll ar.. under ><lling funclion
L h(t) w ,fi (9.lsl
'"
l . If conditions 9.1(0) and 9.I(b) an: ,",,;.Bed. ,lien
LI
hil!)_ L
h(lt + 1). ~
.J2
(9.10)
(9. Id)
Sine. fl<IJ and 1jI'(r) arc rdated 11 lh~ rcqulI'l:d n'ndiu,,". can be ilnJ'O"'-..! On (h{.lJI for the
des;B" of w",'clet ')'51Cm "'jlh deSIred J'fOP"ni ...
R""alllhal H(ro) is Ill<: Fourle, lransform of (hltJ). Let Cl"') IIc Ihe FOII,ie, nansfum of 18(k).
Thus. \le h",'c
G(fi,j. ~ L g(k~-""
Lel Hj.;:.) and Gj.;:) rCI""_", nl "'""",lh'dy 7... lransf"nn of (hit)! .nd (g( t). Th. "ultsclip<::"
{o dl"inllui'h be'''' .. n Fouri<r 'nd Ztran,form_ I",,"ad of doli"'n: ewry{h'"i from ." ,,,,rch. we
COn ju,{ (mnsl ... i{ rrolO /I(m) "'iog Table 9.1.
,. ,
,. ,
,. 1/t
-,
.'.'" -11<
Tho fon<:li"" 11 , is in &<ncral nO! Ih. complo_' conjug .. c of /I,. h i,"" d<~n~<l in Ihe ,.blt.
0.- <quII'akn,ly
11, (::) _ /1,(11::)
Note ho",e'cr. Ill>! many ,ulh"", use only :: with 1::1 = I. in ",hieh case i', is eOlOpl.,
conj",a,. of If, .
".., condnioo' in le",,' of H, .00 G, ,re:
11,(1) .. (9.3a)
~~ ~~-
(9.3f)
Again tile ghen IT","It> Un be p",,cd b.scd On ccn,," 1'e.",1l< ",e arc gO'"B to d,scuss in
S<'Ction 9.4.
Whon (t _ n) "odd. 'he sum in _<qu.", br.d e, ,'.nb"", .nd hon,.., 'he wh,,'e RHS "lIII;'he .
When ik - n) i, ",'en. say it - n, = 21. ("..ilb 1 E N) ,""" k = n + 21 ,od we "blain
' "''
A. pcr d"ublc.'' '"f, nrthogon.hty, L h(" + 21) h(n) .. .5(1) ODd he""" RHS i. MnzerO, ",dy
",he" 1 = O. Thus.
II/ (w)ll + IH(<I + H) 11 s I
-- ,[ L L ,0--.0'" 1 0 _~))
11(",)0(",) + 11(", + It"j G(",+..-) ,, -
, ,. h(t)g(n)r- ( [! + (- I) I
When (.I - n) is od<l. 111,- sun, in >quare brll<~<' ,->ni,he. and hene<: llIe whole RHS ,'an;s"",.
When (t - nl I' e'''n. ~y <k - n) " 21. (w.th I .. NI Ihen k" n + 21 alld we obllin
H(m)G(Il'1 + 11(,,/+ If) G(hH If ) ~ i ;(~ hi" + 21) ~(")) e -" ... (I + (- 1)" 1)
-.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Sin,"" I h(n +2J)g(n) = O for.ny I. RHS i. idcnlic.lly .om for.1I ", Therefore.
- lI(m)G(roj + lI(m + K)G(ro + Jr) _ 11
3. H(1f) .. 0 ro, lba"" w",'clot .' y",cm.
,,, 1/(",) -7!'
I ,V_ I
L h(k"'-~ "
>12
I /<-, I
1/(0) ~ 7!' LhIt) ~ r:;.fi .. 1
,,2 0 v~
Since III(W)I' + III(ro + If)l' .. I. 11/(0)1' + 11f(,.)I' .. I, This imp l,.s that l/(1r) _ 0.
W. 'hen say ,hat Ill''') h". ,cro .. m" If. 1I(1t) " 0 ma~c, 11("'1 litt le fl at ., (oJ" " as
>1\0",-" in f'gure . Wc can male 11(",) more na,.,
(0= 1< by pUlling con",raints Il'lIr) " 0, /1"(1<1
.. O. ,'e.Mure "'" flatn ."''' ", .. fr. srnoo"'., bcromcs lhe .... Img funelion . D.ubcchics fillers
arc in facl ma,im.lly 1101 fol f ', Ilowc'w. there i, a IImu to the number of mnstraWs th.t ".
Cln pUI 10 fI{m). For an N.llp orthogonal ",'avdcI .y:w:m. Nil coo'lJllinL> arc 'ctjuir.d (or
""hOB",,.lu)'. Onc <l<ilree of f",.,do", f<>r , .."0 al id ~ If_ The rcmaininx Nf!. - I <1<&= of
freedom may be umll.,d for more n.tnc>s at (0 '" , ..
H(O) " I . nd 1I(1f). 0 ba,ically imp""" Ihat h(")". k,w P"-'" fillor.
4. p Zeros al '" ~ "IS "G"1".I<ni to H ,(l) dl.i"bk by (, + I)'.
ll>c ptoof is lri,-i.l, Sub>liluling ~ ., ~~ .. in (~ + 11 ' wc ob,a,"
(;;+I) .. (~- -+ I)
fll '" ~ If. (~ ..... + I)-' has Srn zern of onlor p. Ho"".
II, (~) (: + I Y' Q(:)
.,001: __ 1
Whonp_!
Ht(z) '" (z ... 11' Q(z)
", IIOIZ ", -1
1.11,(;:) is d,"i"blc by (;:: ... I)' is C<!";"alonl 10 f l'"I"(I)dl~O. for '" = O. I. 2... P - I.
'-f
We "'ill prove 'hi' by >.howinJ !hot bu!h conditi"", pu, .. me cOR"",in" on ..."Iin& lil'cT
roeffo.:;"nls.
Form_O.
,-,
Fur m ;>0 O.
.-.
fl"(r)dl=O=> L (- I )'h(k)=O
,-,
f r \l'(r) JI =0 => L (- I)' k- h(k)"O. '" = 1.2, .... IP - 1)
.~
I " _, I " I
H,( - I)~ ~ .:.. /I{tK - 1) _7-" .:.. (_ 1) hit)
.J2 .J2 I
We "gain have p "'Iuation, "hieh are idenll"a l 1n !he equ'"nn' <!erh'ed for ".ni,h,ng momenls
of .,(1).
Wc <'.n nOW ""swc' the fOllowi ng q ue>l,on" A.",unllng thA' the 18(k)1 ",tisH", '1.1(0). 9,I(b)
n.
land. therefore utom.II". ll y 9 , I (e what choice of Ig(t) will m. ke the nom.ininK conduion.'
"lid ?
In terms (If H( m)
Define the matrix
lflm) If (m + Jr)]
M(m) (9.4 )
[ ClIVi G(<I+X)
and its rornpl ex ''OnjU&.tc t"'nspose
"
M (wI =
[
-
I/ (w)
I/(w+ lfl
-l
G(ml
G(w+ If)
(9_S)
So<> M " _ IM)' i, 'he <n'.. "" of M lA m.,ri, which ,,",,,fie, M " _ ( MJ' " M -I i, c. lled
IInl h..y.) Onc ""n",quence of thi, i"
.
M (mI M(wl= 0 [' ',]
which le.ds 10 I few more rel.tkms. (PI" eumple.
A more inlpo1tanl cO<l>equcnce is that Wc un oow 1';<1 di=1 rd . ltons bell.'ccn 11 aOO G.
"Slog the In" <:'-';Io" formul. for 2 )( 2 matrices
[ T' [ ~]
lJ b I d
(9.8)
ea - (lJd - br)-('
"
g(l)-a(-I) h(N- I -l) (9.18)
The <'11";"" <>f t.(ro) ., ~;.. lea.Js 10 g(l) (-I)' h(] - t) .... h;"h ;, ",ha. SOme . nLhor. p",fcr. If
I/o(t) 1 = 1"(0), II( I). h(L) I. "'he'" L = N - I. lhe mo", usu.1 choice i_ "'(a,) = e-JU> '" Itlch
le.d. w
g(k)., (-I)' h(L _ k) .. (-I)' h(N _ I _ k) (9.20)
M (!)='
H.W 11.(-
.
<,,] (9,21)
, [ G, (d GL (-;:)
M , - ' W~ [ -H,(Z)
--
G,I!) ] (9,n)
11, ( -;:) G,l-I)
Tbe dc.cnninan. of M ,(z) is:
ol.(d = H,(z) G,(-l l - H,(-;:) G,(!i (9,23)
For ootnparing Ih. in"~""'n formu l. for M,I!I ", ,, It /01, 'w. "'. <kft,," "" before
G ,!z) .. - "'(z) H, (- 4) (9.24 )
t.(;:)t.(;:). J (9.m
In sum",ary. if /(,(<') ha. been foond. Ihen all Mlilable G,\:l an: of Ih. form:
(9.26)
",he.. t.(ll;~ any function ""ll!
tl.(-z) = ----I'.(Z) (9,21)
t.(z)tl.(z) ~ I (9.28)
Co",'~""'ly. e"cry such &(~) wi ll produtt a .ui,abk G"~).
As a ~pec,al ClSI.'. if (h(t)]. bl(t)( arc finite ~uen""'. H,W. G,~z) and Ilo(zf . re
polynomi als. ,1{:) i. an odd polYHOmial and Is nc,'er ,cm. ""ttI" pos;;ibly a, : '" 0 (bee,"",
AI, '(<:),. denned e,'ery,,'hofC "'''''1'' I"""hly ., ;: .. 0). "" 4(;:) mu" to.: of 'he f' J<m,
4(:) '" (/ z" ,
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Le' !'OIIfio. 'ra",fo.m of Ill') "nd \I'll) to.: re>p<:cti"cly .;\(,,,) ."~ ",(ro). If .,(If hos AI ,.",>IIing
mOments, ' ht-n
(9.JO)
(9.31)
(9.32)
Now. wc e" . lo.le fo, ro .. O.
We how Ihot ;;':0) c$. 0 (.inre '\"eroge ,,1"" o f o/>I.t) '.' nm zero). Th" me.", .h... for ;.to)
10 !>e cqo.1 10 O. G(O) "'0'" !>e <"<1..,1 '0 zero. Again ~O) c$. O. G(O) .. 0 and VIOl" 0 in'ply
lIIal G'(O) = O. Simil .. I~. ~O) c$. O. G(O) = o. G,(O) = 0 al><l Ii-"(O ) = 0 impl~ thal G~(OI "' 0
.1><1 so on, W. ,hu, deduce .hat G .."d it' titsl AI - I de,i''''''e" mo" ,'anish orlln.
W..... n' )fldll;"" 0" H. S!' we "'" ,be ha,;e ",lat;o",.I" r G(M)" -A(r<'H"".";-;"J
d,,(uel thal M ,.",-.it;oll mom",," are equi,. I.n' 10
H{tr)., H'(tr)., ItV , ~,,). 0 (9.33)
The lllcury of e""'pln fnnelk"" "",c, Ill., Ill;, ;, cqoi\"'i<nI
"
ttl
U (w) ..
[ l
I +e- J
2 L(",) (9. ,4)
(935)
UIO) .. I (9.36)
SI<p 1: C",W<f( SC<.'Ood Eq. (9.3(.1><1 lIIinl Eq. (9.31) ""od i,;"o' 00'0. <'Oodi,ion On L
(9,lS)
5/.1' J: Show"'.,
1">I,+"tl] (9,39)
5/.1' J: I~n<l lOe polynomial P of 11>< lowe>l ",,"'iblc dct:"'e "'at "'",'c s the "'luJlion and
""n,wl 11 bock '" I I~llI)I '
5/.1' 6: Foelor lL(llI)I' 10 f,nd Uro), and put .,'c'Y""ng b><.i.. loge,hcr,
Examp le:
sup I:
= (I+ r
;-;-.1+;'" IL(ro)I' " [cos' (;) r lL(ro)I'
"-,,, ~
y -s;n ' ( ~)
'"10 'he equ,"O"
. ~.
Th .. ",u>l b< ,alid (01" all ". For I y I < I we can npand (I - y)'v ;n,o a po,,'.r ..,,;n. Slnee
I' " of de~"", M - I nd 'he ,"",ood term "" lhe ri~h' h.nd .. de ...." wllh 1",""0' of yM. ,he
shOl"I<:'" ~.) mu" b< ~uaJ '0 I,!"SI M ICnn> in 'he power series (or (I - _rv.
"
'"
Ply)- ""'[M
L .. -') ,
. .... k
y (9.44)
Onc 'hlng "'C did nol "'cnlion before i, that "'" IlI"st . Iso h'c PI)') ;0, 0 fur )' E 10, 11.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
"nee P is the square of Land}' _ 'in' (,"12) ta~." on ,..1 ... In 10. 11. 8y hlCky colncide"", _
IIIi> Is "1I.r,~-d ""r~ for .11 \,.Ioe. of M.
"h)bc 'hi> is goOO poin' to ... I~ abuu, the numbcr of coefficient> In oh",", Iri .
poIY'IO", .. I, "od regular polynomial,_ SU[lJ>O"" U IV) ha. ,Ie&ree K. Then IL( ~'W'. "'rUlen as sum
of ""non'ial" ha cm" !loin: fron, -K to K. Wc coll.l'''' ~,< npone""'!> 1010 co.incs by
group,nl! ""si'h'. and n"g'"". PO,",'.I!< togethor Ih." "' re bock ", summ,,'lI from 0 tu K.
NothIng ch",,&c> during the nnt few ,{cps. so I' also has lhe degn... K. ",'hich "'t nOW know
to be M _ I. Th;s 'ra",I .. os in'" an H or dell"'" (2M _ I) or lM """fflcient< Ih(k) I. Thus. 11..
I)aub""hle. " a. ..'" ,,-Itb M "~"k hl,,~ n.on " " (,a. 2M """me"'n...
S{<I' 6: Now. >1'0 h.Y< P and wc ho,'c to find our way ba<,~ . o L Tbi" P"""'''' is called >p<oelrnl
f""." , b .. , h", . D,ul>e<:h,e, "",d 'he followinll me'hod, b..",d "" a 'heorem of R ,e...
Do .11 'he "'!II'"'ion, bac~ w.rds unlt l wc bave I.(rojl' in ,<nn. of compk" npon<nlt.l.
'g.in, To m'~e " fe <a,,"r, let l: '" ~-jW, '" I .(w~' 'um, in'o.;; w.,
"n' a flul}'" omial in;; of
<kg= 2,11 - 2. Find all tbe root.'; of this polynomial. The><: roots turn ou' to COme In group" of
four. con<i",nll of <,.
11", ~ >Ild I ff,. If ;;, " .1 or un 'he unll ",rck . ,h.rr are onl)' 'wo r<~'"
'" ~ruup ,""'ead of four.
w. ""'~'" one pair l:;. In; from each group of four or On" roo, from each pair, If ".
multiply !he Wrm, \l: - ,,) for >11 k tOllether. wc get the orig,"" polynon,;>1 ( ccpt for
conmnl). If ..... e mu lllply IIIe leml. (: - ~J. ooly for the seh.'(!o;(I roolS. "'e gel Ihe square rool
!h., wc W>llt. cHep' fOT' COn"'n' (!h", '" ,he COn'cn' of 'he RkS>. mc<><cm). Wc Un !hen replaee
<: by e-'" ag''". Wc ar. done.
Th). whok fH"C<''' is quite a m.thematical .ou, d. for,' . Wc ill"", ..,. it by <kn"ing me
D.ubeen, ",,,-dets ",,,h I and 2 l'.nbh,nll moment<. Wc can 5Iart wi.h S'cp 5. th. o.he, ""P'
arc ooly m:"~ss.ary for the proof. Th<"" 'wo c .... can be do"" by """d. For noon: " ""isliing
,"O""'n'S. 'hcs< equations mu'" be soh'ed numeric.lly or by u>ing ,) ,ooolie alg"rbra softWaIT
padag .
P(y) "'
"'hieh ltamlalCs in'" l.(roH' " I. ano Ob'lOU>ly .(Q.)", I. This lead, .0:
1+,,-1 1+[jW
II(w) ~
[ 2
1" L(w) [ 2 ) I
3
I _Jl I ,..~-jM
if(w)" I
r;;:
,,2
hit) t ""
2
Thord.,.-. h(O) .. -E. h(l) " i' ,",'n;<h ar. ,be Ha" "".lI ng function coeff"''''''I>.
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
~-'"
,
- - - +2+-
~
,
",-,(-; +2--iJ
1'11<: rorres.pontlin~ pol) nomi., is _ ; .. 2, _ ~ ... hkh hi . f'OOI' 2 :t Ji . W. choo.. on<:
of tIIem. li~. 2 + Jj 1'11<: Ri"", lhI:of<m sa)', um L\w) """uk! be
h(O) = 1+ ./3
"""iF
h(l) = 3 + ./3
4J2
3-,fj
h(2) .. 4J2
1- ,fj
h(3) ..
"""iF
9.8 PARAMETR IC DESIGN OF ORTHOGONAL AND
BIORTHOGONAL WAVELETS
In N .. p Da ~ be,hie. ,,.,cl., syMern design. Ihere "ill al,,o), be Nil ""n",,,,inlS Iha< im"","
o" ""gooal"y of ".n,I., of ",alin~ fU"'lions. Onc d.~= of f",.<Iom I. u,ili,cd for n,a ~ing lbe
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
>re. under wayel., fun'tlon uro. Th~ rema'n'"ll Nn. - I degree m freedom" "",d for making
tile ",.Iing (.. ",lion IIllUinlOlly >n'ooth, W~ .an .. se lbe (Nil - 1) degre" of f"",do,n for lIlaking
paramNnc w.,'et",,_ Or '" ,~her word, (Nn. _ I) ,..",bles con be f",dy cit"sen and 'hen make
"'",'clel .)'>Iern ,hot ..,isnes III the on""gonality ""n"'raints. In Ihe follOWing "'Cl ion " 'e ... ill
describe how '0 dc.iBn SOIch .... ,.clet 'Y>1cms.
I h(n)~ I. h(n)~*
. " .. ._ 2
L
...... .-
g(n )+ I g(n) ~O (9.48)
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Equ."on, (9.47_9 .49) h,,o. koy .,,(0 ,n 'he gonem,oo o f pam"",,,k wno",,,. Eq u,""'"s (9. ~ 7)
and (9.49) ..... <q".illltnt '0
( 11,(;:)+ 11,(;:))1,. , ~.Ji. (9.50)
'-I
Let A, be fhe unitary m.trk .... i ~ I. 2 .... n. Let ifS prooucl be U.
,
V. A, XA , X ... XA _ n A,_
uu'.[n,,)[n,,)'
,., ,.,
Who'" 0 dcno.<> con,plox. conjugal. find Ir""SJX>S< .
Hence Ihe produCI of any number of onilOf)' m .. nu. h again onilary, Th" IIlIrhe~ Ih ... onil&rJ
Dr.M.H.Moradi ;Biomedical Engineering Faculty ;
Sine.: ... " "'p"ral. c,en and odd In<l<:" of MA).lhe filler lenglh will redu"" 10 half. 50 I. ' -Ifie'
frum 0 to Hr! _ l.
LOI lIJ,z) be tile Z transform of the "'que""e Ih(2* + Ill.
II"(;I.)~ L iI(2k+ Ili:' ... "'O.I ..... Nf2-
11.(:) . h( l ) + zh(3). for N. 2
S,milarly. we h,,'" '- tran,form for the ""'Iuen"" l.~(l) a.,
P{l) ~
H,(Z) HoW]
(953)
[ G,lz ) G.W
Now. to pro' e '"at Ihe po(ypha"" n'anh Eq, (9_53) i, U""'ry. "'e h.ve 10 pro"c the foll"""''"11
rood,,,on ..
(9.5-1)
(9.56)
Simil",ly. we ~a v<: -
L h(lt + I) h(2m + Ilz'-"
1/"(, 11/. (z) - L,
- - -
E<ju,tlon (9 ..59) ists 001), 'f 1 = 0 ontl is equol to ottc.
HcrK'C 111,)11+ 1 1I. (z~' _ I
Simil..- rn ..~'<lu'" "an be r"lIowru to pro'" E<j<. (9.55) and (9.56).
The (>IUof 'hot pol) ph ..", rn.<rh ". unltMY follo",_, from E<j'_ (9_54). (9_55) ""d (9 _56). That
"