Foundations of Signal Processing

Tables

Martin Vetterli
École Polytechnique Fédérale de Lausanne and University of California, Berkeley
Jelena Kovačević
Carnegie Mellon University
Vivek K Goyal
Massachusetts Institute of Technology

August 8, 2012

Copyright (c) 2012 Martin Vetterli, Jelena Kovačević, and Vivek K Goyal.
These materials are protected by copyright under the
Attribution-NonCommercial-NoDerivs 3.0 Unported License
from Creative Commons.
iiSignal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Cover photograph by Christiane Grimm, Geneva, Switzerland.

Experimental set up by Prof. Libero Zuppiroli and Philippe Bugnon,

Laboratory of Optoelectronics Molecular Materials, EPFL, Lausanne,
Switzerland.

The photograph captures an experiment first described by Isaac Newton

in Opticks in 1730, explaining how the white light can be split into its
color components and then resynthesized. It is a physical implementa-
tion of a white light decomposition into its Fourier components—the col-
ors of the rainbow, followed by a synthesis to recover the original. This
experiment graphically summarizes the major theme of the book—many
signals can be split into essential components, and the signal’s charac-
teristics can be better understood by looking at its components; the
process called analysis. These components can be used for processing
the signal; more importantly, the signal can be perfectly recovered from
those same components, through the process called synthesis.
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Fundamental subspaces associated with matrix A

The four fundamental subspaces associated with a real matrix A ∈ RM×N . The
matrix determines an orthogonal decomposition of RN into the row space of A and
the null space of A; and an orthogonal decomposition of RM into the column space
(range) of A and the left null space of A. The column and row spaces of A have
the same dimension, which equals the rank of A. (Figure inspired by the cover
of G. Strang. Introduction to Linear Algebra. http://math.mit.edu/linearalgebra/,
4th edition, 2009.)

RN RM
A

row space column space

R(AT ) R(A)

AT

0 • AT A • 0

null space
N (A) left
null space
N (AT )

R(AT ) ⊥ N (A)
dim(R(AT )) + dim(N (A)) = N R(A) ⊥ N (AT )
dim(R(A)) + dim(N (AT )) = M

Space Symbol Definition Dimension

M
Column space (range) R(A) {y ∈ C | y = Ax for some x ∈ CN } rank(A)
Left null space N (A∗ ) {y ∈ CM | A∗ y = 0} M − rank(A)
dim(R(A)) + dim(N (A∗ )) = M

Row space R(A∗ ) {x ∈ CN | x = A∗ y for some y ∈ CM } rank(A)

Null space (kernel) N (A) {x ∈ CN | Ax = 0} N − rank(A)
∗
dim(R(A )) + dim(N (A)) = N
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Discrete-time processing concepts

Concept Notation Infinite-length sequences Finite-length sequences

Shift δn−1 linear circular

Sequence xn n∈Z n ∈ {0, 1, . . . , N − 1}

vector

LSI system hn n∈Z n ∈ {0, 1, . . . , N − 1}

filter, impulse response
operator

X N−1
X
Convolution h∗x xk hn−k xk hN,(n−k) mod N
k∈Z k=0

Eigensequence v ejωn ej(2π/N)kn

satisfies h ∗ vλ = λvλ h ∗ vω = H(ejω ) vω h ∗ vk = Hk vk
invariant space Sλ Sω = {α ejωn } Sk = {α ej(2π/N)kn }
α ∈ C, ω ∈ R α ∈ C, k ∈ Z

Frequency response λ λω = H(ejω ) λk = Hk

X N−1
X
eigenvalue hn e−jωn hn e−j(2π/N)kn
n∈Z n=0

Fourier transform X DTFT DFT

X N−1
X
jω −jωn
spectrum X(e )= xn e Xk = xn e−j(2π/N)kn
n∈Z n=0
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Multirate discrete-time processing concepts

Concept Expression

Sampling factor 2
Input xn , X(z), X(ejω )

Downsampling by 2 yn = x2n
y = D2 x h i
Y (z) = (1/2) X(z 1/2 ) + X(−z 1/2 )
h i
Y (ejω ) = (1/2) X(ejω/2 ) + X(ej(ω−2π)/2 )
(
xn/2 , n even;
Upsampling by 2 yn =
0, otherwise.
y = U2 x
Y (z) = X(z 2 )
Y (ejω ) = X(ej2ω )

Filtering with h y = D2 H x h i
& downsampling by 2 Y (z) = (1/2) 12 H(z 1/2 )X(z 1/2 ) + H(−z 1/2 )X(−z 1/2 )
h i
Y (ejω ) = (1/2) H(ejω/2 )X(ejω/2 ) + H(ej(ω−2π)/2 )X(ej(ω−2π)/2 )

Upsampling by 2 y = GU2 x
& filtering with g Y (z) = G(z)X(z 2 )
Y (ejω ) = G(ejω )X(ej2ω )

Sampling factor N
Input xn , X(z), X(ejω )

Downsampling by N yn = xNn
y = DN x
N−1
X
k 1/N
Y (z) = (1/N ) X(ωN z )
k=0
N−1
X
Y (ejω ) = (1/N ) X(ej(ω−2πk/N) )
k=0
(
xn/N , n/N ∈ Z;
Upsampling by N yn =
0, otherwise.
y = UN x
Y (z) = X(z N )
Y (ejω ) = X(ejNω )
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Kronecker delta sequence properties

Kronecker delta sequence

X
Normalization δn = 1
n∈Z
X X
Sifting xn0 −n δn = δn0 −n xn = xn0
n∈Z n∈Z
Shifting xn ∗n δn−n0 = xn−n0
Sampling xn δn = x0 δn
Restriction xn δn = 1{0} x

Dirac delta function properties

Dirac delta function

Z ∞
Normalization δ(t) dt = 1
Z−∞
∞ Z ∞
Sifting x(t0 − t) δ(t) dt = x(t) δ(t0 − t) dt = x(t0 )
−∞ −∞
Shifting x(t) ∗t δ(t − t0 ) = x(t − t0 )
Sampling x(t) δ(t) = x(0) δ(t)
Scaling δ(t/a) = |a| δ(t) for any nonzero a ∈ R
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Ideal filters with unit-norm impulse response

Ideal filters Time domain DTFT domain

r ( p
ω0 2π/ω0 , |ω| ≤ ω0 /2;
Ideal lowpass filter sinc(ω0 n/2)
2π 0, otherwise.
( √
√ N, |ω| ≤ π/N ;
Ideal N th-band filter (1/ N ) sinc(πn/N )
0, otherwise.
( √
√ 2, |ω| ≤ π/2;
Ideal halfband lowpass filter (1/ 2) sinc(πn/2)
0, otherwise.

Unit-norm box and sinc functions/sequences

Functions on the real line FT

√
x(t), t ∈ R, kxk = 1 X(ω), ω ∈ R, kXk = 2π
( √
1/ t0 , |t| ≤ t0 /2; √
Box t0 sinc (ωt0 /2)
0, otherwise.
r ( p
ω0 2π/ω0 , |ω| ≤ ω0 /2;
Sinc sinc(ω0 t/2)
2π 0, otherwise.

Periodic functions FS
√
x(t), t ∈ [−T /2, T /2), kxk = 1 Xk , k ∈ Z, kXk = 1/ T
( √ √
1/ t0 , |t| ≤ t0 /2; t0
Box sinc (πkt0 /T )
0, otherwise. T
r ( √
k0 sinc(πk0 t/T ) 1/ k0 T , |k| ≤ (k0 − 1)/2;
Sinc
T sinc(πt/T ) 0, otherwise.

Infinite-length sequences DTFT

√
xn , n ∈ Z, kxk = 1 X(ejω ), ω ∈ [−π, π), kXk = 2π
( √
1/ n0 , |n| ≤ (n0 − 1)/2; √ sinc (ωn0 /2)
Box n0
0, otherwise. sinc (ω/2)
r ( p
ω0 2π/ω0 , |ω| ≤ ω0 /2;
Sinc sinc(ω0 n/2)
2π 0, otherwise.

Finite-length sequences DFT

√
xn , n ∈ {0, 1, . . . , N − 1}, kxk = 1 Xk , k ∈ {0, 1, . . . , N − 1}, kXk = N
( √
1/ n0 , |n − N/2| ≥ (n0 − 1)/2; √ sinc (πn0 k/N )
Box n0
0, otherwise. sinc (πk/N )
r ( p
k0 sinc (πnk0 /N ) N/k0 , |k − N/2| ≥ (k0 − 1)/2;
Sinc
N sinc (πn/N ) 0, otherwise.
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Energy and power concepts for sequences

Deterministic sequences WSS discrete-time stochastic processes

Energy spectral density Power spectral density

X
A(ejω ) = |X(ejω )|2 A(ejω ) = E[ xn x∗n−k ] e−jωk
k∈Z
Energy Power
1 π
Z
1 π Z
E = A(ejω ) dω P = A(ejω ) dω
2π −πX 2π −π
E = a0 = |xn |2 P = a0 = E[ |xn |2 ]
n∈Z

Orthogonality concepts for sequences

Deterministic sequences WSS discrete-time stochastic processes

∗
Time cx,y,k = hxn , yn−k in = 0 cx,y,k = E[ xn yn−k ]=0
Frequency Cx,y (ejω ) = X(ejω ) Y ∗ (ejω ) = 0 jω
Cx,y (e ) = 0
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Deterministic discrete-time autocorrelation/crosscorrelation

properties

Domain Autocorrelation/Crosscorrelation Properties

Sequences x n , yn
xk x∗k−n an = a∗−n
P
Time an
Pk∈Z
cn ∗
k∈Z xk yk−n cx,y,n = c∗y,x,−n

DTFT A(ejω ) |X(ejω )|2 A(ejω ) = A∗ (ejω )

C(ejω ) X(ejω ) Y ∗ (ejω ) Cx,y (ejω ) = Cy,x
∗
(ejω )

z-transform A(z) X(z) X∗ (z −1 ) A(z) = A∗ (z −1 )

C(z) X(z) Y∗ (z −1 ) Cx,y (z) = Cy,x∗ (z −1 )

DFT Ak |Xk |2 Ak = A∗−k mod N

Ck Xk Yk∗ ∗
Ck = Cy,x,−k mod N

Real sequences x n , yn
P
Time an x x an = a−n
Pk∈Z k k−n
cn k∈Z xk yk−n cx,y,n = cy,x,−n

DTFT A(ejω ) |X(ejω )|2 A(ejω ) = A(e−jω )

C(ejω ) X(ejω ) Y (e−jω ) Cx,y (ejω ) = Cy,x (e−jω )

z-transform A(z) X(z) X(z −1 ) A(z) = A(z −1 )

C(z) X(z) Y (z −1 ) Cx,y (z) = Cy,x (z −1 )

DFT Ak |Xk |2 Ak = A−k mod N

Ck Xk Y k Ck = Cy,x,−k mod N

h iT
Vector of sequences x0,n x1,n
" #
a0,n c0,1,n An = A∗−n
Time An
c1,0,n a1,n An = AT−n
" #
A0 (ejω ) C0,1 (ejω ) A(ejω ) = A∗ (ejω )
DTFT A(ejω )
C1,0 (ejω ) A1 (ejω ) A(ejω ) = AT (e−jω )
" #
A0 (z) C0,1 (z) A(z) = A∗ (z −1 )
z-transform A(z)
C1,0 (z) A1 (z) A(z) = AT (z −1 )
" #
A0,k C0,1,k Ak = A∗−k mod N
DFT Ak
C1,0,k A1,k Ak = AT−k mod N
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Discrete/continuous Fourier transform summary

Transform Forward/Inverse Duality/Periodicity

Z ∞
X(ω) = x(t) e−jωt dt
Fourier −∞
1
Z ∞
transform
x(t) = X(ω) ejωt dω
2π −∞

Z T /2
1
Xk = x(t) e−j(2π/T )kt dt
Fourier T −T /2 dual with DTFT
series x(t + T ) = x(t)
X
x(t) = Xk ej(2π/T )kt
k∈Z

X
X(ejω ) = xn e−jωn
Discrete-time n∈ZZ dual with Fourier series
1 π
Fourier transform xn = X(e jω
)e jωn
dω X(ej(ω+2π) ) = X(ejω )
2π −π

N−1
X
−j(2π/N)kn
Xk = xn e
Discrete n=0
Fourier transform N−1
1 X
xn = Xk ej(2π/N)kn
N k=0
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Discrete Fourier transform (DFT) properties

DFT properties Time domain DFT domain

Basic properties
Linearity αxn + βyn αXk + βYk
Circular shift x(n−n0 ) mod N kn0
WN Xk
in time
Circular shift −k0 n
WN xn X(k−k0 ) mod N
in frequency
Circular time reversal x−n mod N X−k mod N
Circular convolution (h ⊛ x)n H k Xk
in time
Circular convolution hn xn 1
(H ⊛ X)k
in frequency N
N−1
Circular deterministic an =
X
xk x∗(k−n) mod N Ak = |Xk |2
autocorrelation
k=0
N−1
Circular deterministic cn =
X
xk ∗
y(k−n) Ck = Xk Yk∗
crosscorrelation mod N
k=0
N−1 N−1
X 1 X 1
Parseval’s equality kxk2 = |xn |2 = |Xk |2 = kXk2
n=0
N k=0 N
Related sequences
Conjugate x∗n ∗
X−k mod N
Conjugate, time reversed x∗−n mod N Xk∗
∗
Real part ℜ(xn ) (Xk + X−k mod N )/2
∗
Imaginary part ℑ(xn ) (Xk − X−k mod N )/(2j)
Conjugate-symmetric part (xn + x∗−n mod N )/2 ℜ(Xk )
Conjugate-antisymmetric part (xn − x∗−n mod N )/(2j) ℑ(Xk )
Symmetries for real x
∗
X conjugate symmetric Xk = X−k mod N
Real part of X even ℜ(Xk ) = ℜ(X−k mod N )
Imaginary part of X odd ℑ(Xk ) = −ℑ(X−k mod N )
Magnitude of X even |Xk | = |X−k mod N |
Phase of X odd arg Xk = − arg X−k mod N
Common transform pairs
Kronecker delta sequence δn 1
kn0
Shifted Kronecker delta sequence δ(n−n0 ) mod N WN
Constant sequence 1 N δk s
Geometric sequence αn (1 − αWN kN k
)/(1 − αWN )
( q
k0 −1
r
k0 sinc (πnk0 /N ) N N
Periodic sinc sequence k
, k− 2 ≥ 2 ;
0
(ideal lowpass filter) N sinc (πn/N ) 0, otherwise.
(
n0 −1
√1 , n− N ≥ ; √ sinc (πn0 k/N )
Box sequence n 0 2 2 n0
0, otherwise. sinc (πk/N )
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Fourier series properties

FS properties Time domain FS domain

Basic properties
Linearity αx(t) + βy(t) αXk + βYk
Shift in time x(t − t0 ) e−j(2π/T )kt0 Xk
Shift in frequency ej(2π/T )k0 t x(t) X(k − k0 )
Time reversal x(−t) X−k
Differentiation dn x(t)/dtn (j2πk/T )n Xk
Z t
Integration x(τ ) dτ (T /(j2πk)) Xk , X0 = 0
−T /2
Circular convolution in time (h ⊛ x)(t) T H k Xk
Convolution in frequency h(t) x(t) (H ∗ X)k
Z T /2
Circular deterministic a(t) = x(τ ) x∗ (τ − t) dτ Ak = T |Xk |2
autocorrelation −T /2
Z T /2
Circular deterministic c(t) = x(τ ) y ∗ (τ − t) dτ Ck = T Xk Yk∗
crosscorrelation −T /2
Z T /2 X
Parseval’s equality kxk2 = |x(t)|2 dt = T |Xk |2 = T kXk2
−T /2 k∈Z
Related functions
Conjugate x∗ (t) ∗
X−k
Conjugate, time reversed x∗ (−t) Xk∗
∗
Real part ℜ(x(t)) (Xk + X−k )/2
∗
Imaginary part ℑ(x(t)) (Xk − X−k )/(2j)
Conjugate-symmetric part (x(t) + x∗ (−t))/2 ℜ(Xk )
Conjugate-antisymmetric part (x(t) − x∗ (−t))/(2j) ℑ(Xk )
Symmetries for real x
∗
X conjugate symmetric Xk = X−k
Real part of X even ℜ(Xk ) = ℜ(X−k )
Imaginary part of X odd ℑ(Xk ) = −ℑ(X−k )
Magnitude of X even |Xk | = |X−k |
Phase of X odd arg Xk = − arg X−k
Common transform pairs X
Dirac comb δ(t − nT ) 1/T
n∈Z
r  √
Periodic sinc function k0 sinc(πk0 t/T ) 1/ k0 T , |k| ≤ (k0 − 1)/2;
(ideal lowpass filter) T sinc(πt/T ) 0, otherwise.
 √ √
Box function 1/ t0 , |t| ≤ t0 /2; t0
(one period) sinc (πkt0 /T )
0, t0 /2 < |t| ≤ T /2. T
 
Square wave −1, t ∈ [−1/2, 0); 2j/(πk), k odd;
(one period with T = 1) 1, t ∈ [0, 1/2). 0, k even.

1/4, k = 0;
Triangle wave

1
(one period with T = 1) 2
− |t|, |t| ≤ 1/2. 1/(πk)2 , k odd;
0, k 6= 0 even.


Sawtooth wave 0, k = 0;
(one period with T = 1) 2t, |t| ≤ 1/2.
j(−1)k /(πk), k 6= 0.
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

z-transform properties

z-transform properties Time domain z domain ROC

ROC properties General seq. 0 ≤ r1 < |z| < r2 ≤ ∞
Finite-length seq. all z, except possibly 0, ∞
Right-sided seq. |z| > largest pole
Left-sided seq. |z| < smallest pole
BIBO stable ⊃ |z| = 1
Basic properties
Linearity αxn + βyn αX(z) + βY (z) ⊃ ROCx ∩ ROCy
Shift in time xn−n0 z −n0 X(z) ROCx
Scaling in time
N−1
1 X k 1/N
Downsampling xNn X(WN z ) (ROCx )1/N
N k=0

xn/N , n/N ∈ Z;
Upsampling X(z N ) (ROCx )N
0, otherwise.
Scaling in z αn xn X(α−1 z) |α| ROCx
Time reversal x−n X(z −1 ) 1/ROCx
Differentiation nk xn (−1)k z k ∂ k X(z)/∂z k ROCx
X
mk = n xn = (−1) ∂ X(z)/∂z k
k k k
z=1
n∈Z
Convolution in time (h ∗ x)n H(z) X(z) ⊃ ROCh ∩ ROCx
Deterministic an =
X
xk x∗k−n A(z) = X(z) X∗ (z −1
) ROCx ∩ 1/ROCx
autocorrelation
k∈Z
Deterministic cn =
X
∗
xk yk−n C(z) = X(z) Y∗ (z −1 ) 1/ROCx ∩ ROCy
crosscorrelation
k∈Z
Related sequences
Conjugate x∗n X ∗ (z ∗ ) ROCx
Conjugate, time reversed x∗−n X∗ (z −1 ) 1/ROCx
Real part ℜ(xn ) (X(z) + X ∗ (z ∗ ))/2 ROCx
Imaginary part ℑ(xn ) (X(z) − X ∗ (z ∗ ))/(2j) ROCx
Symmetries for real x
X conjugate symmetric X(z) = X ∗ (z ∗ )
Common transform pairs
Kronecker delta sequence δn 1 all z
Shifted Kronecker delta sequence δn−n0 z −n0 all z, except possibly 0, ∞
Geometric sequence αn un 1/(1 − αz −1 ) |z| > |α|
−αn u−n−1 |z| < |α|
Arithmetic-geometric sequence n αn un (αz −1 )/(1 − αz −1 )2 |z| > |α|
−n αn u−n−1 |z| < |α|
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

DTFT: Discrete-time Fourier transform properties

DTFT properties Time domain DTFT domain

Basic properties
Linearity αxn + βyn αX(ejω ) + βY (ejω )
Shift in time xn−n0 e−jωn0 X(ejω )
Shift in frequency ejω0 n xn X(ej(ω−ω0 ) )
Scaling in time N−1
1 X
Downsampling xNn X(ej(ω−2πk)/N )
N k=0
Upsampling xn/N , n = ℓN ; 0, otherwise X(ejNω )
Time reversal x−n X(e−jω )
∂ k X(ejω )
Differentiation in freq. (−jn)k xn
∂ω k
X ∂X(ejω )
Moments mk = nk xn = (−j)k
n∈Z
∂ω ω=0
Convolution in time (h ∗ x)n H(ejω ) X(ejω )
1
Circular convolution in frequency hn xn (H ⊛ X)(ejω )
X 2π
Deterministic autocorrelation an = xk x∗k−n A(ejω ) = |X(ejω )|2
k∈Z
X
∗
Deterministic crosscorrelation cn = xk yk−n C(ejω ) = X(ejω ) Y ∗ (ejω )
k∈Z
X 1
Z π 1
Parseval’s equality kxk2 = |xn |2 = |X(ejω )|2 dω = kXk2
n∈Z
2π −π 2π
Related sequences
Conjugate x∗n X ∗ (e−jω )
Conjugate, time reversed x∗−n X ∗ (ejω )
Real part ℜ(xn ) (X(ejω ) + X ∗ (e−jω ))/2
Imaginary part ℑ(xn ) (X(ejω ) − X ∗ (e−jω ))/(2j)
Conjugate-symmetric part (xn + x∗−n )/2 ℜ(X(ejω ))
Conjugate-antisymmetric part (xn − x∗−n )/(2j) ℑ(X(ejω ))
Symmetries for real x
X conjugate symmetric X(ejω ) = X ∗ (e−jω )
Real part of X even ℜ(X(ejω )) = ℜ(X(e−jω ))
Imaginary part of X odd ℑ(X(ejω )) = −ℑ(X(e−jω ))
Magnitude of X even |X(ejω )| = |X(e−jω )|
Phase of X odd arg X(ejω ) = − arg X(e−jω )
Common transform pairs
Kronecker delta sequence δn 1
Shifted Kronecker delta sequence δn−n0 e−jωn
∞
0
X
Constant sequence 1 2π δ(ω − 2πk)
k=−∞
Geometric sequence αn un 1/(1 − αe−jω ), |α| < 1
Arithmetic-geometric sequence n αn un αe−jω /(1 − αe−jω )2 , |α| < 1
r  p
Sinc sequence ω0 2π/ω0 , |ω| ≤ ω0 /2;
(ideal lowpass filter) sinc(ω0 n/2)
 2π √ 0, otherwise.
1/ n0 , |n| ≤ (n0 − 1)/2; √ sinc (ωn0 /2)
Box sequence n0
0, otherwise. sinc (ω/2)
Signal Processing: Foundations c 2012 Vetterli, Kovačević & Goyal

Fourier transform properties

FT properties Time domain FT domain

Basic properties
Linearity αx(t) + βy(t) αX(ω) + βY (ω)
Shift in time x(t − t0 ) e−jωt0 X(ω)
Shift in frequency ejω0 t x(t) X(ω − ω0 )
Scaling in time and frequency x(αt) (1/α)X (ω/α)
Time reversal x(−t) X(−ω)
Differentiation in time dn x(t)/dtn (jω)n X(ω)
Differentiation in frequency (−jt)n x(t) dn X(ω)/dω n
Z t
Integration in time x(τ ) dτ X(ω)/jω, X(0) = 0
−∞ Z
∞ dk X(ω)
Moments mk = tk x(t) dt = (j)k
−∞ dω k ω=0
Convolution in time (h ∗ x)(t) H(ω) X(ω)
1
Convolution in frequency h(t) x(t) (H ∗ X)(ω)
Z ∞ 2π
∗
Deterministic autocorrelation a(t) = x(τ ) x (τ − t) dτ A(ω) = |X(ω)|2
Z −∞
∞
Deterministic crosscorrelation c(t) = x(τ ) y ∗ (τ − t) dτ C(ω) = X(ω)Y ∗ (ω)
−∞
Z ∞ 1
Z ∞
1
Parseval’s equality kxk2 = |x(t)|2 dt = |X(ω)|2 dω = kXk2
−∞ 2π −∞ 2π
Related functions
Conjugate x∗ (t) X ∗ (−ω)
Conjugate, time reversed x∗ (−t) X ∗ (ω)
Real part ℜ(x(t)) (X(ω) + X ∗ (−ω))/2
Imaginary part ℑ(x(t)) (X(ω) − X ∗ (−ω))/(2j)
Conjugate-symmetric part (x(t) + x∗ (−t))/2 ℜ(X(ω))
Conjugate-antisymmetric part (x(t) − x∗ (−t))/(2j) ℑ(X(ω))
Symmetries for real x
X conjugate symmetric X(ω) = X ∗ (−ω)
Real part of X even ℜ(X(ω)) = ℜ(X(−ω))
Imaginary part of X odd ℑ(X(ω)) = −ℑ(X(−ω))
Magnitude of X even |X(ω)| = |X(−ω)|
Phase of X odd arg X(ω) = − arg X(−ω)
Common transform pairs
Dirac delta function δ(t) 1
Shifted Dirac delta function X− t0 )
δ(t e−jω0 t X
Dirac comb δ(t − nT ) (2π/T ) δ(ω − (2π/T )k)
n∈Z k∈Z
Constant function 1 2π δ(ω)
Exponential function e−α|t| (2α)/(ω 2 + α2 )
−αt2 −ω 2 /α
p
Gaussian function er  π/α
p e
Sinc function ω0 2π/ω0 , |ω| ≤ ω0 /2;
(ideal lowpass filter) sinc(ω0 t/2)
 2π √ 0, otherwise.
1/ t0 , |t| ≤ t0 /2; √
Box function t0 sinc (ωt0 /2)
 0, otherwise.
1 − |t|, |t| < 1;
Triangle function (sinc (ω/2))2
0, otherwise.