Some Formulae

for
Digital Signal Processing

Prof. Dr.-Ing. Walter Kellermann

Andreas Brendel, M.Sc.

Multimedia Communications and Signal Processing (LMS)

University of Erlangen-Nuremberg

WS 2019-20
1 Transforms
1.1 The z-Transform
Definition of the (two-sided) z-transform:
+∞
X
Z{x[k]} = X(z) = x[k]z −k (1)
k=−∞
1
I
x[k] = X(z)z k−1 dz (2)
2πj
C

Theorems:

Sequence z-Transform Region of Convergence (ROC)

x[k + k0 ] z k0 X(z) Rx− < |z| < Rx+

1


x[−k] X z 1/Rx+ < |z| < 1/Rx−
 
z
x[k] · z0k X z0 Rx− < |z/z0 | < Rx+

kx[k] −z dX(z)
dz Rx− < |z| < Rx+

x∗ [k] X ∗ (z ∗ ) Rx− < |z| < Rx+

x[k] ∗ y[k] X(z)Y (z) max {Rx− , Ry− } < |z| < min {Rx+ , Ry+ }

1 z
H

x[k] · y[k] 2πj X w Y (w)w−1 dw Rx− · Ry− < |z| < Rx+ · Ry+
Cxy

1.2 The Discrete-time Fourier Transform (DTFT)

Definition of the DTFT:
∞
X
X(ejΩ ) = F∗ {x[k]} = x[k]e−jΩk (3)
k=−∞
Z2π
1
x[k] = F∗−1 {X(ejΩ )} = X(ejΩ )ejΩk dΩ (4)
2π
0

DTFT of the rectangular window:


1 0 ≤ k < M,
wM [k] = (5)
0 k < 0, k ≥ M,
M−1
jΩ
X 1 − e−jΩM sin(ΩM/2) −jΩ(M−1)/2
WM (e ) = e−jΩk = = e . (6)
1 − e−jΩ sin(Ω/2)
k=0

DTFT of the unit impulse sequence:

∞ ∞  
X 2π X 2πλ
δ[k − λN ] ◦ — δ Ω− (7)
N N
λ=−∞ λ=−∞
Theorems:
Sequence DTFT
jΩk0
x[k + k0 ] e X(ejΩ )

x[−k] X(e−jΩ )

x[k] · ejΩ0 k X(ej(Ω−Ω0 ) )

jΩ
kx[k] j dX(e
dΩ
)

x∗ [k] X ∗ (e−jΩ )

x[k] ∗ y[k] X(ejΩ ) · Y (ejΩ )

2π
1
X(ejη )Y (ej(Ω−η) )dη
R
x[k] · y[k] 2π
0

1.3 The Discrete Fourier Transform (DFT)

Definition of the DFT:
M−1 M−1
2π kµ
X X
X[µ] = DFTM {x[k]} = x[k]e−j M kµ = x[k]wM (8)
k=0 k=0
M−1 M−1
1 X 2π 1 X
x[k] = IDFTM {X[µ]} = X[µ]ej M kµ = −kµ
X[µ]wM (9)
M µ=0 M µ=0

Theorems:

x̃[k] = x[k] for k = 0(1)M − 1,

x̃[k + λM ] = x̃[k] for λ ∈ Z

Sequence DFT
−µκ
x̃[k + κ] X[µ] · wM

x̃[−k] = x̃[M − k] X[−µ]

kλ
x[k] · wM X[µ + λ]

x∗ [k] X ∗ [−µ]

M−1
P
x1 [κ] · x̃2 [k − κ] X1 [µ] · X2 [µ]
κ=0
M−1
1
P
x1 [k] · x2 [k] M X1 [µ − κ] · X2 [κ]
κ=0
2 Causal LTI Systems
2.1 Causal LTI Systems in the Time Domain
N -th Order Difference Equation
N
X M
X
y[k] = − an y[k − n] + bm x[k − m] (10)
n=1 m=0

State Space Representation for Causal MIMO Systems

uN,1 [k + 1] = AN,N · uN,1 [k] + BN,L · xL,1 [k] (State equation) (11)
yK,1 [k] = CK,N · uN,1 [k] + DK,L · xL,1 [k] (Output equation) (12)

Transfer function
H(z) = C [zI − A]−1 B + D (13)

2.2 Causal LTI Systems in the z-Domain

M
X M
X
bm z −m
bM−m z m
Y (z)
H(z) := = m=0 = m=0
(14)
X(z) XN N
X
−n n
an z aN −n z
n=0 n=0

For Real-valued Systems:

H ∗ (z ∗ ) = H(z) (15)

2.3 Causal LTI Systems in the DTFT Domain - Frequency Response

m
X m
Y
bµ ejµΩ ejΩ − z0µ


µ=0 µ=1
H(ejΩ ) = |H(ejΩ )| · e−jb(Ω) = n = bm · n (16)
X Y
jνΩ jΩ


aν e e − z∞ν
ν=0 ν=1

The phase b(Ω):

Im H(ejΩ )

b(Ω) = −arg H(ejΩ ) = −arctan = arg N (ejΩ ) − arg Z(ejΩ )
  
jΩ
(17)
Re {H(e )}

The group delay τg (Ω):

d b(Ω)
τg (Ω) = (18)
dΩ
Analytical Computation of the Frequency Response
Transfer function:
m
Q
(z − z0µ )
µ=1
H(z) = bm · Qn (19)
(z − z∞ν )
ν=1

Magnitude:
m q
1 + ρ20µ − 2ρ0µ cos (Ω − ψ0µ )
Q
µ=1
|H(ejΩ )| = |bm | · Q
n p (20)
1 + ρ2∞ν − 2ρ∞ν cos (Ω − ψ∞ν )
ν=1

Phase: n m
π X X
b(Ω) = [1 − sign bm ] · sign [sin Ω] + b∞ν (Ω) + b0µ (Ω) (21)
2 ν=1 µ=1
Group delay:
n
X m
X
τg (Ω) = τg∞ν (Ω) + τg0µ (Ω) (22)
ν=1 µ=1

Allpass Systems:
Frequency Response
ejnΩ · N (e−jΩ )
A(ejΩ ) = (23)
N (ejΩ )
Phase
n  
X ρν sin(Ω − ψν )
bA (Ω) = nΩ + 2 arctan (24)
ν=1
1 − ρν cos(Ω − ψν )
Group delay
n
X 1 − ρ2ν
τA (Ω) = (25)
ν=1
1 + ρ2ν − 2ρν cos(Ω − ψν )

(ρν · ejψν , ν = 1, . . . , n describe the singularities of N (ejΩ ))

3 Design of Discrete-time Causal LTI Systems

3.1 Design of Recursive Systems
Normalized form of bilinear transform:
sT
w = = u + jv (26)
2
z−1 1+w
w = , z= (27)
z+1 1−w
 
Ω
v = tan (28)
2

Magnitude square of the transfer function for the continuous-time system:

1
|G(jv)|2 = (C is a real constant) (29)
1 + C 2 |K(jv)|2

Tolerances for the characteristic function:

p p
2δP − δP2 1 − δS2
∆1 = , ∆2 = (30)
1 − δP δS

Design of Normalized Lowpass Filters

Power filters (Butterworth filters):
Characteristic function:
|K̃(jv ′ )|2 = (v ′ )2n (31)
Required filter order:
∆2 ∆2
log log
∆1 ∆1
n≥ = vS (32)
log vS′ log
vP
Chebyshev filters of Type I (Chebyshev approximation in passband):
Characteristic function:

|K̃(jv ′ )|2 = Tn2 (v ′ ) (33)


cos[narccosx] for |x| ≤ 1
Tn (x) = , n ∈ N0 (34)
cosh[narcoshx] for |x| > 1
(Chebyshev polynomial of first kind)
Required filter order: s" #
∆2 ∆2 ∆ 2 ∆22
arcosh arcosh ln + −1
∆1 ∆1 ∆1 ∆21
n≥ = vS =
arcoshvS′
" s #
arcosh vS vS2
vP ln + −1
vP vP2

Chebyshev filters of Type II (Chebyshev approximation in stopband):

Characteristic function:
Tn (vs′ )
K̃(jv ′ ) =  ′ (35)
v
Tn vs′

4 Sampling Rate Conversion and Multirate Systems

4.1 Sampling Rate Conversion
The impulse train in the time and frequency domain
Time domain:
 R−1
1 for k = n · R + κ 1 X j 2π
rR,κ [k] = = e R ρ(k−κ) (36)
0 else R ρ=0

Frequency domain:
∞
X
RR,κ (ejΩ ) = rR,κ [k]e−jΩk
k=−∞
R−1 ∞
1 X −j 2π
e−j (Ω− R ρ)k
2π
X
= e R ρκ
R ρ=0
k=−∞
R−1  
1 X 2π 2πρ
= e−j R ρκ · 2πδ Ω − (37)
R ρ=0
R

4.2 The Polyphase Decomposition

Time domain:
R−1
1 X
hD [k] = hD,R,ν [k] (38)
R ν=0
R−1 
1 X j 2π ρ(k−ν) hD [k] for k = λR + ν, λ ∈ Z
hD,R,ν [k] = hD [k] · e R = (39)
R ρ=0 0 else

z-domain:
R−1
X
HD (z) = HD,R,ν (z) (40)
ν=0
R−1
1 X 2π 2π
HD,R,ν (z) = ′
HD,R,ν (z R )z −ν = HD (ze−j R ρ ) · e−j R ρν (41)
R ρ=0