Wavelet analysis has attracted attention for its ability to analyze rapidly
changing transient signals. Any application using the Fourier transform
can be formulated using wavelets to provide more accurately localized
temporal and frequency information. This paper gives an overview of
wavelet analysis and describes a software toolbox created by HP
Laboratories Japan to aid in the development of wavelet applications.

      

0&. (&3-#- &-) && 10&. ."),3 ), $/-. 106 ), 2'*& )( !.-  10&. -*.,/' ( -#!(& *,)--6
&.- "- ..,. '/" ..(.#)( ,(.&3 #( -#!(& *,)--6 #(! 10&.- , 0,3 /- /& ), *,)--#(! ()(-..#)(,3
#(! . "- ( -/-- /&&3 **&# #( '(3 **&#.#)(- -#!(&-
-/" - .,(-#(. -#!(& (&3-#- #'! (&3-#- )''/(#6
0&.- "0 ,. '/" 2#.'(. #( ." '."'.#-
.#)(- -3-.'- ( ).", -#!(& *,)--#(! **&#.#)(- . #-
)''/(#.3 *,"*- '), -) ."( #( (!#(,#(! /-
().  (1 ."),3 #( ." -(- .". '(3 ) ." #- (
." '."'.#& 0&)*'(. "- )&&)1  0,3 #(.,-.6
."(#+/- #(0)&0 #( 10&.- -/( )#(! +/,6
#(! *." " ,(. 0&)*'(.- (  0#1 - ,-)&06
./, '#,,), #&.,- . 1, 0&)* #(*((.&3 #(
#(! -)' ) ." # #/&.#- #(",(. #( )/,#, (&3-#- ),
0,#)/- -#!(& *,)--#(! **&#.#)(- ( "0 ( %()1(
2'*&  .3*#& +/-.#)( #- ")1 .) ,&. ." )/,#, )6
), -)' .#' ". #- (1 #- ." 0&)*'(. ) ,(.
 ##(.- .) ." !&)& ), &)& "0#), )  /(.#)( "
,-/&.- )( ." '."'.#& )/(.#)(- ) 10&.- .".
0&)*'(. ) 10&. (&3-#- (  )(-#, ( )/.6
*,)0#  /(# # ,'1),% ), ." -/$.
!,)1." ) ." #..&1))6&3 ."),3  #,-. */&#-" #(
#."#( ."#- ,'1),%  )'')( &#(% #- -.&#-" 6  1"#" -)/!".  (1 **,)" .) (-1, -)' )
.1( ." '(3 #0,-# # *,)&'- .". , ) #(.,-. .) ."- # #/&.#- !#( #. #- ." /(# 3#(! ,'1),% '
# ,(. #&- #(&/#(! &.,#& (!#(,#(! -#!(& *,)6 *)--#& 3 ,(. ,-/&.- #( 10&. ."),3 ,&. .) *,)6
--#(! . )'*,--#)( '."'.#& (&3-#- ",')(# &'- ) ",')(# (&3-#- &-) .) -#'#&, *,)&'- #( )*,6
(&3-#- )*,.), ."),3 ( *"3-#-  ,.&- +/(./' .), ."),3 && ." &,5(63!'/( ."),3 .". "-
#& ."),3 0&. ."),3 "- )' ( .#0 , ) !(,. '/" ) ." 2#.'(.
,-," #( ."- #&- ", , )**),./(#.#- ), /,.",
( &.,#& (!#(,#(! .", "0 ( #(*((. 0&6
0&)*'(. ) )." ." '."'.#& /(,-.(#(! )
)*'(.- #( ." (&3-#- ) ()(-..#)(,3 -#!(&- -*# #&&3
10&.- (  1# ,(! ) **&#.#)(- #( -#( (
#( ." ),' ) ." -"),.6.,' )/,#, .,(- ),'  0,#.#)( )
(!#(,#(!
1"#" && ." ), .,(- ),' 1- #,-. */&#-" #(
#% )/,#, (&3-#- 10&. (&3-#- &- 1#." 2*(-#)( 
 '$), 0( #( 10&. ."),3 1- ." #-)0,3
) /(.#)(- #( .,'- )  -. ) -#- /(.#)(- (&#% ) -'))." ').", 10&.- 1")- -. ) #-,. .,(-&.#)(-
)/,#, (&3-#- 10&. (&3-#- 2*(- /(.#)(- (). #( ( #&.#)(- ),'- ( ),.")(),'& -#- ),  1", 
.,'- ) .,#!)()'.,# *)&3()'#&- /. #( .,'- )   #- ." ,& (/',- (  #- ." -. ) && /(.#)(-  .".
1"#" , !(,. #( ." ),' ) .,(-&.#)(- ( #&.#)(- "0 )/( (,!3 .". #- /(.#)(- ), 1"#"
)  #2 /(.#)( && ."  
  " 10&.-


).#( #( ."#- 13 "0 -*#& -&#(! *,)*,.#- "3

.
.  
, &)&#4 #( .#' ( ,+/(3 *,'#..#(!  &)-, )(6


(.#)( .1( ." /(.#)( #(! ,*,-(. ( ."#,
) ##(.- ,., (/',#& -.#&#.3 #( ,)(-.,/.#)( ( "#- #-  '#( # ,( ,)' ." ), .,(- ),' ( ."
'(#*/&.#)( #- (-/, ), - () ),.")(),'& -#- (  !(,. ,)'
-'))." 10&.- "/- ." /(# 3#(! ,'1),% ,)/!".
" )$.#0 ) 10&. (&3-#- #- .)  #( ."- *)1, /& )/.  .., /(,-.(#(! (  (1 **,)" .". )0,6
10&. -#- /(.#)(- ( #(  ##(. '.")- ), ."#, )'- ." # #/&.#- #( ." -"),.6.,' )/,#, .,(- ),'
)'*/..#)( . (  -")1( .". 0,3 **&#.#)( /-#(! ." '.")-
-. )/,#, .,(- ),'  (  ),'/&. /-#(! 106
&.- .) *,)0# '), &)&#4 .'*),& ), -*.#& ( ,6 ( ." (2. -.#)( 1 !#0 ( )0,0#1 ) ." '#( ./,-
+/(3 #( ),'.#)( "/- #(-. )  ,+/(3 -*.,/' ) 10&. (&3-#- ( ."( ./,( .)  -) .1, .))&)2 .".

 ',  1&..6%, )/,(&  Hewlett-Packard Company 1994

HP Laboratories Japan has developed to help in the developĆ
ment of wavelet applications.  m,n(t) + a 0
*mń2
ǒ

t * nb 0
am
0
Ǔ
, (7)
For an excellent tutorial introduction to the subject, see Rioul
and Vetterli2 and the references therein (it lists 106 referĆ where m and n are integer values. The discrete wavelet
ences). Daubechies' book1 is a standard reference on the transform and its inverse transform are defined as follows:
subject.
ŕ
R
S m,n + Ȁ m,n(t)s(t)dt, (8)
  
 *R

This section gives a quick overview of the main formulas.

s(t) + k  ȍ ȍ Sm,nm,n(t), (9)
m n

   where k is a constant value for normalization.

Consider a complexĆvalued function  satisfying the followĆ The function m,n(t) provides sampling points on the scaleĆ
ing conditions: time plane: linear sampling in the time (bĆaxis) direction but
logarithmic in the scale (aĆaxis) direction.
ŕ
R
2
|(t)| dt t R (1) The most common situation is that a0 is chosen as:
*R
a0 = 21/v, (10)
ŕ
R 2
|
()|
c  + 2 d t R, (2) where v is an integer value, and that v pieces of m,n(t) are
| |
*R processed as one group, which is called a  The integer v
where
is the Fourier transform of . The first condition is the number of voices per octave; it defines a wellĆtempered
implies finite energy of the function , and the second conĆ scale in the sense of music. This is analogous to the use of a
dition, the admissibility condition, implies that if
() is set of narrowband filters in conventional Fourier analysis.
smooth then
(0) = 0. Wavelet analysis is not limited to dyadic scale analysis. By
The function  is the mother wavelet. using an appropriate number of voices per octave, wavelet
analysis can effectively perform the 1/3Ćoctave, 1/6Ćoctave,
 
 or 1/12Ćoctave analyses that are used in acoustics.
If  satisfies the conditions described above, then the  The main focus of current research is on finding optimal
 
 of a real signal s(t) with respect to the wavelet wavelet basis functions and efficient algorithms for computĆ
function (t) is defined as: ing the corresponding wavelet transforms. The wavelet basis
function can be implemented as an FIR (finite impulse reĆ
ŕ
R
S(b, a) + 1
Ǹa
ǒ
Ȁ t * b
a s(t)dt, Ǔ (3) sponse) filter or an IIR (infinite impulse response) filter
*R depending on the particular properties needed.

where Ȁ denotes the complex conjugate of , and this is 
 
defined on the open (b,a) halfĆplane (b Ů , a u 0). The
parameter b corresponds to the time shift and the parameter
a corresponds to the scale of the analyzing wavelet.
If we define a,b(t) as
This section describes how to display complexĆvalued
functions such as equations 3 and 8 so that useful informaĆ
tion about the signal s(t) can be highlighted. There are two
aspects to consider.

 a,b(t) + a *1ń2 t * b
a , ǒ Ǔ (4)
The open (b,a) halfĆplane on which the wavelet transform is
defined can be mapped onto the full plane (b,-log(a)). This
representation is indispensable if we want to display, in a
which means rescaling by a and shifting by b, then equation
single picture, information with a wide range of scale parameĆ
3 can be written as a scalar or inner product of the real signal
ters. For example, for sound signals in the audible range, a
s(t) with the function a,b(t):
spread of ten octaves is common. A disadvantage of this

ŕ
R representation, on the other hand, is that straight lines on
S(b, a) + Ȁ a,b(t)s(t)dt. (5) the open (b,a) halfĆplane become exponential curves in the
*R logarithmic representation.
When function (t) satisfies the admissibility condition, Expressions 3 and 8 depend on the choice of the analyzing
equation 2, the original signal s(t) can be obtained from the wavelet . To obtain full quantitative information about the
wavelet transform S(b,a) by the following inverse formula: signal s(t) from its transform S(b,a), we need to know the
analyzing wavelet . There are, however, many features of
ŕ ŕ
R R
s(t) + 1 S(b, a) a,b(t) dadb . (6) the signal that are independent of the choice of . Such feaĆ
c a2 tures involve the phase of the complexĆvalued functions.
*R *R
Therefore, it is useful to represent separately the modulus
  
 and the phase of the complexĆvalued function S(b,a) to be
In the discrete domain, the scale and shift parameters are described.
discretized as a = a m0 and b = nb0, and the analyzing waveĆ Shown in Figs. 1 and 2 is an example of the wavelet transĆ
lets are also discretized as follows:
form of a localized pulse that approximates a delta function.

 Hewlett-Packard Company 1994 December 1994 HewlettĆPackard Journal 

 NJ
Scale a
1ąifĂ 0Ă v t v 1
f(t) + 0ąotherwise.

If we define the function y(t) as y(t) = f(2t) - f(2t-1), we

can obtain the following function:

ȡą1ąifĂ 0 t t v 1ń2
y(t) + ȥ*1ąifĂ 1ń2 ttv1
Ȣą0ąotherwise.
The function f(t) is the Haar scaling function, and y(t) is the
Haar wavelet. This function is orthogonal to its own translaĆ
tions and dilations, that is, the family
0
ym,n(t) = 2-m/2 y(2-mt - n),ąąm,n Ů , (14)
Time Shift b
where  is the real integers, constitutes an orthonormal
  Magnitude of the wavelet transform of a delta function.
basis for L2( ). Historically the Haar function was the origiĆ
nal wavelet. This wavelet is not continuous, and its Fourier
The horizontal axis is time in both the magnitude picture,
transform Y(w) decays only like |w|-1, corresponding to
Fig. 1, and the phase picture, Fig. 2, and the vertical axis is
bad frequency localization.
scale, with small scale at the top.
Meyer Wavelet. Yves Meyer constructed a smooth orthonormal
In Fig. 1, the magnitude increases toward the top of the picĆ
wavelet basis as follows. First of all, define the Fourier transĆ
ture. The modulus or magnitude, |S(b,a)|, is converted to
form F(w) of a scaling function f(t) as:
grayscale and is normalized to its maximum, that is, the plot
shows x, where:
ȡ1 ifĂ |w| v 2 p
3
x+ |S| v 1.
|S max|
(11) ȥcosƪ2 nǒ4p |w| * 1Ǔƫą ifĂ 3 p v |w| v 43 (15)
F(w) +ą p 3 2

The phase of S(b,a) is given by a grayscale picture in which Ȣ0 otherwise,

a phase of 0 corresponds to white and a phase of 2p to
where n is a smooth function satisfying the following
black. This convention is quite useful in interpreting the reĆ
conditions:
sulting picture. When the phase reaches 2p, it is wrapped
around to the value 0. The lines where the density drops
abruptly to zero are clearly visible on the picture and play
an important role in the interpretation as a visible line of
n(t) + NJ 0ąifĂ t v 0
1ąifĂ tĂ w 1 (16)

constant phase. In Fig. 2, one can see the lines of constant

with the additional property
phase pointing to the location of the delta function.
n(t) + n(1-t) = 1. (17)
 
  
This function F is plotted in Fig. 3.
Haar Wavelet. The Haar wavelet is the simplest kind of waveĆ
let function. Suppose that f(t) is a box function satisfying In this case, the wavelet function y can be found easily from
the following: F. First, we find the Fourier transform of y:

Scale a
Y(w) + e iwń2 ȍ F(w ) 2p(2l ) 1))Fǒwń2Ǔ (18)
lŮ

+ e iwń2[F(w ) 2p) ) F(w * 2p)]Fǒwń2Ǔ. (19)

The function Y is plotted in Fig. 4.
Now since Y is compactly supported (its duration is finite
and nonzero) and Y Ů Ck where k is arbitrary and may be
R (i.e., Y has at least k derivatives), y can be obtained by
taking the inverse Fourier transform. Fig. 5 shows a graph of
the Meyer wavelet y(t) Ů C4.
Morlet Wavelet. This particular function was most often used
by R. KronlandĆMartinet and J. Morlet. Its Fourier transform
is a shifted Gaussian, adjusted slightly so that Y(0) = 0:
Time Shift b
ǒ Ǔ2 2 2
  Phase of the wavelet transform of a delta function. Y(w) + e * w*w0 ń2 * e *w ń2e *w 0ń2 (20)

y(t) + ǒe *iw 0t * e *w Ǔe*t2ń2.

2ń2
(21)

 December 1994 HewlettĆPackard Journal  Hewlett-Packard Company 1994

 1.2 1.2

1.0

0.8
0.5

0.6
(t)

 (t)
0.0
0.4

0.2
–0.5

0.0

–0.2 –1.0
–6 –3 0 3 6 –6 –3 0 3 6
t
t

  The Meyer wavelet.

  Fourier transform of the scaling function for the Meyer basis.
and tan-1[Imhs, m,ni/Rehs, m,ni], where the brackets indiĆ
Often 0 is chosen so that the ratio of the highest maximum cate the scalar or inner product of the signal waveform s
of  to the second highest maximum is approximately 1/2, with the basis function m,n, that is,
that is,

0 = (2/ln 2)1/2 ] 5.3364…
In practice one often takes 0 = 5. For this value of 0, the
second term in equation 20 is so small that it can be neĆ
glected in practice. Consequently, the Morlet wavelet can
be considered as a modulated Gaussian waveform. Its real
and imaginary parts for 0 = 5 are shown in Figs. 6 and 7,
respectively.
The Morlet wavelet is complex, even though most applicaĆ
tions in which it is used involve only real signals. The waveĆ
let transform of a real signal with this complex wavelet is
plotted in modulusĆphase form, that is, one plots |hs, m,ni|
ŕ
R
(22) ǂs, m,nǃ + s(t)Ȁ m,n(t)dt.
*R
The phase plot is particularly suited for the detection of
singularities.
Daubechies Wavelet. Except for the Haar basis, all of the exĆ
amples of orthonormal wavelet bases consist of infinitely
supported functions. Ingrid Daubechies constructed an orĆ
thonormal wavelet in which  is compactly supported. The
way to ensure compact support for the wavelet  is to
choose a scaling function
with compact support.

1.2 1.0

1.0

0.5
0.8

0.6
Re (t)

0.0

()

0.4

0.2 –0.5

0.0

–1.0
–0.2 –4 –2 0 2 4
–10 –5 0 5 10 t


  Real part of the Morlet wavelet for 0 = 5.

  Fourier transform of the Meyer wavelet.

 Hewlett-Packard Company 1994 December 1994 HewlettĆPackard Journal 

 1.0 1.5

0.5 1.0
Im (t)

(t)
0.0 0.5

–0.5 0.0

–1.0 –0.5
–4 –2 0 2 4 0 1 2 3
t t

  Imaginary part of the Morlet wavelet for w0 = 5.   The Daubechies scaling function for N = 2.

First of all, find a progression {ak;k Ů } satisfying the Figs. 8 and 9 show graphs of the Daubechies scaling function
following four conditions for all integer N w 2: f and the corresponding wavelet y for the value of N = 2.

ak = 0ąif k t 0 or k u 2N (23)

   

R
ȍ a ka k)2m + d 0m for all integer m (24)
The wavelet analysis software developed by HP Laboratories
k+*R
Japan is implemented as a toolbox in the Khoros system.
R
ȍ a k + Ǹ2 (25)
The Khoros system is an integrated software development
environment for information processing and visualization,
k+*R
based on the X Window System. It is distributed in the pubĆ
R lic domain and has been ported to the HPĆUX* operating
ȍ bkkm = 0,ą0 v m v N-1, (26) system.3
k+*R
Khoros components include a visual programming language,
where bk = (-1)ka-k+1.
code generators for extending the visual language and adding
If N = 1, then a0 = a1 = 1, corresponding to the Haar basis. new application packages to the system, an interactive user
interface editor, an interactive image display package, an
We can find a compactly supported scaling function f(t)
from the above progression {ak}. The function f(t) is one 2
solution of a functional equation:
R
f(t) + ȍ a k Ǹ2 f(2t * k). (27)
k+*R 1
It is continuous and compactly supported and satisfies

ŕf(t)dt = 1 for integer N and the corresponding progression

(t)

0
{ak}. The support of f(t) is [0,2N-1].
Furthermore, if bk is defined as the condition 26, the function
y(t) satisfying a functional equation
R
ȍ
–1
y(t) + b k Ǹ2 f(2t * k) (28)
k+*R

is compactly supported and fulfills the following:

• ŕy(t)t dt = 0 for all integers 0 v m v N-1.

m
–2
0 1
t
2 3

• f(t), y(t) Ů Cl(N) for Holder spaces Cl(N), where l(N) is an

integer parameter and the elements of Cl(N) are functions   The Daubechies wavelet for N = 2.
that have l(N) derivatives.

 December 1994 HewlettĆPackard Journal  Hewlett-Packard Company 1994

 s(t)

   *.53/*% 7*4) $/.34".4

t '2&15&.$9

&84&.3*6& ,*#2"29 /' *-"(& 02/$&33*.( .5-&2*$", ".",93*3
".% 3*(.", 02/$&33*.( 2/54*.&3 ".% 
 0,/44*.( 0"$+"(&3 *. 4)& '/,,/7*.( 4)2&& &8"-0,&3 /' 39.4)&3*:&% %"4" ".",93*3
)& )/2/3 3934&- ",3/ 3500/243 4)& 4//,#/8 50%"4& -&4)/%
'/2 .&7 2/54*.&3 $2&"4&% #9 "./4)&2 0&23/. /2 %&6&,/0&% /.
"./4)&2 -"$)*.&  4//,#/8 $/.42*#54&% #9  "#/2"4/2*&3
"0". 4)&  //,#/8 $/.4"*.3 7"6&,&4 "00,*$"4*/. %&6&,;
/0-&.4 4//,3 *-"(& %"4" $/-02&33*/. 54*,*4*&3 ".% /4)&2
54*,*4*&3
3 '/2 4)& .5-#&2 /' 6/*$&3 %*3$533&% &"2,*&2 7& 53& 6 
Example 1. )& '*234 &8"-0,& (*6&3 4)& ".",93*3 /' 47/ 3*.5;
3/*%3 *(  3)/73 " 3*.53/*% 7*4) " 3*.(,& $/.34".4 '2&;
15&.$9 ".% *(3  ".%  2&02&3&.4 *43 7"6&,&4 42".3'/2-
)& )/2*:/.4", "8*3 *3 *. 4*-& *. #/4) 4)& -"(.*45%& 0*$452&
*(  ".% 4)& 0)"3& 0*$452& *(   )& 6&24*$", "8*3 *3
3$",& 3-",, 3$",& "4 4)& 4/0 &24"*. '&"452&3 /' 4)& 3*(.", "2&
&6*%&.4 )/2*:/.4", 342*03 /' $/.34".4 -"(.*45%& ".% 6&24*$",

   ,*.&3 *. 34&0 7*4) 4)& 0)"3& /' 4)& 3*(.",
*( 
3)/73 " 3*.53/*% 7*4) ,*.&"2,9 *.$2&"3*.( '2&15&.$9
)& '/,,/7*.( &8"-0,&3 *,,5342"4& 4)& "%6".4"(&3 /' 4)& 4*-&; )& 7"6&,&4 42".3'/2- ".",93*3 2&35,43 '/2 4)*3 3*(.", "2&
3$",& 2&3/,54*/. 02/0&24*&3 /' 4)& 7"6&,&4 42".3'/2- ".% " 3)/7. *. *(3  ".%  ,&"2,9 6*3*#,& *3 4)& 507"2% 3,/0&
2&,"4&% $/.$&04 4)&   
 '/2 4)& ".",93*3 /' $/22&30/.%*.( 4/ 4)& *.$2&"3& /' '2&15&.$9
6"2*/53 *.054 3*(.",3 *.$,5%*.( " %&,4" '5.$4*/. " 34&0 /2
#/8 '5.$4*/. " %*''&2&.4*",,9 %*3$/.4*.5/53 '5.$4*/. " 2"-0 Example 2. )& 3&$/.% &8"-0,& *3 4)& ".",93*3 /' 4)& 350&2;
'5.$4*/. 3*.53/*%", '5.$4*/.3 " $)*20 3*(.", ".% " 35- /' 0/3*4*/. /' 47/ %&,4" '5.$4*/.3 ".% 47/ 3*.53/*%3 "3 3)/7.
(,*%*.( 4/.&3 *. *(   .& %&,4" '5.$4*/. *3 ,"2(&2 4)". 4)& 3*.53/*%",
3*(.",3 ".% *3 6*3*#,& *. *(   #54 4)& /4)&2 *3 -5$) 3-",,&2

 
)*3 3&$4*/. (*6&3 3&6&2", "00,*$"4*/. &8"-0,&3 /' 7"6&,&4;
#"3&% 3*(.", ".",93*3 *.$,5%*.( #/4) 34"4*/."29 ".% ./.34";
4*/."29 3*(.", ".",93*3 )&3& 2&35,43 7&2& /#4"*.&% 7*4) "
/2,&4 7"6&,&4 4)"4 *3 " $/-0,&8 3*.53/*% 7*.%/7&% 7*4) "
"533*". &.6&,/0& &802&33&% "3 '/,,/73
ǒ

4 + & *$4 &80 *  4 
Ǔ  
".% %/&3 ./4 "00&"2
*(3  ".%  3)/7 4)& 7"6&,&4 42".3'/2- 2&02&3&.4"4*/.3
!& $". &"3*,9 3&& 4)& 47/ 0&"+3 "4 3-",,&2 3$",& 4)"4 $/22&;
30/.% 4/ 4)& %*3$/.4*.5*4*&3 $/.4"*.&% *. 4)& *.054 3*(.",
Example 3. )*3 &8"-0,& 3)/73 4)& ".",93*3 /' " 35- /' 4)2&&
3*.53/*%3 7*4) %*''&2&.4 34"24*.( 4*-&3 )& *.054 3*(.", 3)/7.
*. *(  *3 ./4 %*3$/.4*.5/53 #54 *43 '*234 %&2*6"4*6& *3

7)&2& $ *3 " $/.34".4 6",5& /'  3/ 4)"4 4)& '5.$4*/.
4

3"4*3'*&3 4)& "%-*33*#*,*49 $/.%*4*/.
Scale a

   "(.*45%& /' 4)& 7"6&,&4

42".3'/2- /' " $/.34".4;'2&15&.$9
0 3*.53/*%
Time Shift b

 Hewlett-Packard Company 1994 &$&-#&2  &7,&44;"$+"2% /52.", 

 Scale a


 
   
   "!
0 
Time Shift b
s(t)


    !
t  !
Scale a


     
    
0 !  !
Time Shift b
Scale a


 
   
    
0 !  !
Time Shift b

     "

   Hewlett-Packard Company 1994
s(t)


     
t  
Scale a


     
      
0    
Time Shift b
Scale a


 
   
      
0    
Time Shift b
s(t)


    
t   

 Hewlett-Packard Company 1994    

  
 Scale a

  &!,- ' ,  /.$,

,*&+'*% ' , * +!&-+'!+ /!,
0 !*&, +,*,!& ,!%+
Time Shift b
Scale a

   + ' ,  /.$,

,*&+'*% ' , * +!&-+'!+ /!,
0 !*&, +,*,!& ,!%+
Time Shift b

', ,  *)-&!+ & ,  !&&!&+ ' ,  '%('&&,+   


* .*1 $*$1 .!+!$ !& ,  /.$, ,*&+'*% (!,-*+   /.$, ,*&+'*% + ,  , ' !++,!& ,  ,!%3
+$ ($& !&,' ,!%3!&.*!&, $$+ /!, & +(, *,!' 3
!+  &  $'/3*)-&1 +!&-+'! +,*,+ !*+, '$$'/
(&&, '& ,  +$ (*%,*  !+ (*'(*,1 !+ !%('*,&,
1  %!-%3*)-&1 &  ! 3*)-&1 +!&-+'!
!& +(,*$ (*'++!& ' +!&$+ -, '+ &', , 1&%!
Real Data Analysis.  !+ +,!'& + '/+ ,  *+-$,+ ' *$ , +(,*-% !+($1+ / * ,!% , !+ .& 1 +%$$ !&*3
&$1+!+  !+ , (*'.! 1 ,  # ,.&+ &+,*-%&, %&,+ *$,!. ,' ,  $$ /!, 
!.!+!'& !+ ,  ,*&+%!,,* ,-*&3'& , '  -$3& ,*&+3
  *(*+&,,!'& ' +!&$+ %1 &!, ! ,  $$ + ( !+
!.* , , /+ ,#& ,  &,* *)-&1 ' 
  2
&', $ ,!%3!&.*!&, , *'- '-,  !+ ,!%3(&&,
/!, ,  %+-*%&, +(& +, ,' 
 #2 & ', *
"-+,%&, &  (*'*% (,!.$1 -  , &!)-
/'*+ ,  , !+ !$,* ,' ((*'0!%,$1  3#2 &3
$$ ,    
 + & (*'('+ , -++
/!,    ,!% !&,*.$ ,/& ('!&,+ !+ 
+
'$!)- $$+ (, ,' ,  $'$ +,*-,-* (*%!,,!&
  !&(-, +!&$ !+ ($',, !& !  & , !+ + ,  ,*&+3 +(*,!'& ' ,  +!&$ '%('&&,+
'*% /+ (*'*% '* ,  .$- ' .   & ,  %&!3
,- & ( + ' ,  /.$, ,*&+'*% * + '/& !& !+
 &  *+(,!.$1

  *&+%!,,* ,-*&3'& ,

 %*  /$,,3#* '-*&$  Hewlett-Packard Company 1994