Appendix: Tables and formulae

Multidimensional Fourier transform: integral relations

Space variable: x = (x1 , . . . , xd ) ∈ Rd

Frequency variable: ω = (ω1 , . . . , ωd ) ∈ Rd

Relation

Z
Direct transform fˆ(ω) = F{f }(ω) = f (x)e−jhω,xi dx1 · · · dxd
Rd

Z
1
−1
{fˆ} = fˆ(ω)ejhω,xi dω1 · · · dωd = f (x)
a.e.
Inverse transform F
(2π)d Rd

Z Z
1
Parseval f (x)g ∗ (x)dx1 · · · dxd = fˆ(ω)ĝ ∗ (ω)dω1 · · · dωd
Rd (2π)d Rd

Z Z
1
Energy conservation |f (x)|2 dx1 · · · dxd = |fˆ(ω)|2 dω1 · · · dωd
Rd (2π)d Rd

Z
Integral f (x)dx1 · · · dxd = fˆ(0)
Rd

∂ n1 · · · ∂ nd fˆ(ω)
Z
Moments xn1 1 · · · xnd d f (x)dx1 · · · dxd = j n1 +···+nd
Rd ∂ω1n1 · · · ∂ωdnd
ω=0
Z
F
Convolution: (f ∗ h)(x) f (y)h(x − y)dy1 · · · dyp ←→ fˆ(ω) · ĥ(ω)
y∈Rd

1
Tables and formulae Michael Unser

Multidimensional Fourier transform: properties

x ∈ Rd fˆ(ω) = f (x)e−jhω,xi dx1 · · · dxd

R
Operation f (x), Rd

Linearity a1 f1 (x) + a2 f2 (x) a1 fˆ1 (ω) + a2 fˆ2 (ω)

d
Y d
Y
Separability f (x) = fi (xi ) fˆ(ω) = fˆi (ωi )
i=1 i=1

Duality fˆ(x) (2π)d f (−ω)

 ∗
Real-valued signal f (x) real fˆ(ω) = fˆ(−ω) (Hermitian symmetry)

Reflection f (−x) fˆ(−ω)

Shift f (x − x0 ) e−jhω,x0 i fˆ(ω)

Modulation ejhω0 ,xi f (x) fˆ(ω − ω 0 )

Scaling f (x/a) |a|d fˆ(aω)

|detA|−1 fˆ (A−1 )T ω


Affine transformation f (Ax)

Convolution (f ∗ g)(x) fˆ(ω) · ĝ(ω)

∂ n f (x)
Differentiation (jωi )n fˆ(ω)
∂xni

1
Multiplication f (x) · g(x) (fˆ ∗ ĝ)(ω)
(2π)d

∂ n fˆ(ω)
Multiplication by monomial xni f (x) jn
∂ωin

2
Tables and formulae Michael Unser

Useful 1D Fourier-transform pairs

Z +∞
f (x), x∈R fˆ(ω) = f (x)e−jωx dx
−∞

Finite-energy functions

 1, |x| ≤ 1  ω  sin(ω/2)
2
rect(x) = β 0 (x) = sinc =
 0, otherwise 2π (ω/2)

 1 − |x|, |x| ≤ 1  ω   sin(ω/2) 2
tri(x) = β 1 (x) = sinc2 =
 0, |x| > 1 2π ω/2
ω
sincn+1


β n (x) = β 0 ∗ β n−1 (x)
2π
sin(πx) ω
sinc(x) = rect
πx 2π

1 2 2
√ e−x /2 e−ω /2
2π
2a
e−a|x| with a > 0
a2 + ω2

Generalized functions in S 0

δ(x) 1

δ(x − x0 ) e−jωx0
X X
δ(x − k) 2π δ(ω − 2πn)
k∈Z n∈Z

1 (2π) · δ(ω)

ejω0 x (2π) · δ(ω − ω0 )

cos(ω0 x) π [δ(ω + ω0 ) + δ(ω − ω0 )]

sin(ω0 x) jπ [δ(ω + ω0 ) − δ(ω − ω0 )]

3
Tables and formulae Michael Unser

Multidimensional Fourier-transform pairs

Z
f (x), x = (x1 , . . . , xd ) ∈ Rd fˆ(ω) = f (x) e−jhω,xi dx1 · · · dxd
Rd

Finite-energy functions
d
Y d
Y
f (x) = fi (xi ) fˆi (ωi ) with fˆi = F1D {fi }
i=1 i=1
d ω d ω 
M M i
Y Y
rect(x) = rect(xi ) sinc = sinc
i=1
2π i=1
2π
 p 
ω12 + ω22
p 1
 1, x21 + x22 ≤ 2
J1
circle(x1 , x2 ) = 2π p (J1 (x): Bessel first kind)
 0, otherwise (circle) ω12 + ω22
ω
sinc(x) rect
2π
2 2
(2π)−d/2 e−kxk /2
, x ∈ Rd e−kωk /2

Generalized functions in S 0 (Rd )

d
Y
δ(x) = δ(xi ) 1
i=1

δ(x − x0 ) e−jhω,x0 i

X X
δ(x − k) (2π)d δ(ω − 2πn)
k∈Zd n∈Zd

1 (2π)d · δ(ω)

ejhω0 ,xi (2π)d · δ(ω − ω 0 )

1 cd d−1
, x ∈ Rd , d > 1 d−1
with cd = (4π) 2 Γ( d−1
2 ) [c2 = 2π ]
kxk kωk

Definition: F{f } = fˆ ∈ S 0 (Rd ) is the generalized Fourier transform of f ∈ S 0 (Rd ) iff

1
hf, φi = hfˆ, φ̂i, for all φ ∈ S(Rd ) (Schwartz’ class of test functions).
(2π)d

4
Tables and formulae Michael Unser

Radial functions: Fourier-transform pairs

Z
f (x) = F −1 {fˆ}(x) fˆ(ω) = F{f }(ω) = f (x)e−jhω,xi dx1 · · · dxd
Rd

Radial function
q q
f (x) = f (kxk), kxk = x21 + · · · + x2d fˆ(ω) = fˆ(kωk), kωk = ω12 + · · · + ωd2

Gaussian
2 2
e−kxk /2
, x ∈ Rd (2π)d/2 e−kωk /2

Thin-plate splines
d
cs,d 2s+d π 2 Γ( s+d
2 )
kxks , x ∈ Rd , s ∈ C, −(s + d), s ∈
/ 2N with cs,d = Γ(− 2s )
kωks+d

c0n,d d
kxk2n log kxk, x ∈ Rd , n ∈ N with c0n,d = (−1)n+1 22n−1+d π 2 Γ(n+ d2 ) n!
kωk2n+d

Hyperquadrics
d  d2 +s
2s+1 (2π) 2

2 2 s d a
(a + kxk ) , a > 0, s ∈ C\{0}, x ∈ R K d +s (akωk)
Γ(−s) kωk 2

Kν (x) ≥ 0: Modified Bessel function of second kind

bν,d d
kxkν Kν (kxk), x ∈ Rd , ν ∈ R + d with bν,d = 2ν+d−1 π 2 Γ(ν + d2 )
(1 + kωk2 )ν+ 2

5
Tables and formulae Michael Unser

Multidimensional z -transform: properties

Operation Discrete signal z -transform with z = (z1 , . . . , zd )

X
Definition x[k], k = (k1 , . . . , kd ) ∈ Zd X(z) = x[k] z1−k1 · · · zd−kd
k∈Zd

Linearity a1 x1 [k] + a2 x2 [k] a1 X1 (z) + a2 X2 (z)

Separability x[k] = x1 [k1 ] × · · · × xd [kd ] X(z) = X1 (z1 ) × · · · × Xd (zd )

Delay x[k − k0 ] z −k0 X(z) = z1−k01 · · · zd−k0d X(z)

Reflection xT [k] = x[−k] X(z1−1 , . . . , zd−1 )

X
Convolution (h ∗ x)[k] = h[k1 ]x[k − k1 ] Y (z) = H(z) · X(z)
k1 ∈Zd

6
Tables and formulae Michael Unser

1D z -transform pairs

X
a[k] X(z) = a[k] z −k ROC
k∈Z

 impulse at k = k0
Unit (or Kronecker)
 1, k = k0
δ[k − k0 ] = z −k0 z ∈ C\{0}
 0, otherwise

Unit step 
 1, k≥0 +∞
X 1
u[k] = z −k = |z| > 1
 0, otherwise 1 − z −1
k=0

Rectangularpulse of size m
 1, 0 ≤ k < m m−1
X 1 − z −m
um [k] = z −k = z ∈ C\{0}
 0, otherwise 1 − z −1
k=0

Exponentials
1
ak · u[k] (causal) |a| < |z|
1 − az −1

1 − a2
a|k| (symmetric) |a| < |z| < |a|−1
(1 − az −1 )(1 − az)

Modulated exponentials (causal)

1 − (r cos ω0 )z −1
rk cos(ω0 k) · u[k] |r| < |z|
1 − (2r cos ω0 )z −1 + r2 z −2

1 − (r sin ω0 )z −1
rk sin(ω0 k) · u[k] |r| < |z|
1 − (2r cos ω0 )z −1 + r2 z −2