GRAZ UNIVERSITY OF TECHNOLOGY

Fund. of Digital Communications

Chapter 2: Signals and Systems
Klaus Witrisal
witrisal@[Link]

Signal Processing and Speech Communication Laboratory

[Link]
Graz University of Technology

October 6, 2016

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 1/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

Outline
 2-1 Signal Spaces
 2-2 Linear Operators, Linear Systems, and a Little
Linear Algebra
 2-3 Frequency Domain Representation of Signals
 2-4 Matrix diagonalizations
 2-5 Bandpass signals

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 2/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces

 References: (Figures taken from these books.)
 Barry, Lee, Messerschmitt: “Digital
Communications”, 3rd Ed., Kluwer Academic
Publishers, 2004
 J. G. Proakis and M. Salehi, “Communication
System Engineering,” 2nd Ed., Prentice Hall, 2002
 M. Vetterli, et al., “Foundations of Signal
Processing,” Cambridge, 2014

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 3/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces

 Idea: represent signals as vectors (in linear vector
spaces)
 allows for geometric interpretations
 linear (vector-) algebra can be used for signal
processing algoithms

 applies for continuous-time and discrete-time signals

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 4/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Def: linear vector space
 set of vectors X , and scalars (in R or C) for which
the following operations are defined:
◮ vector addition and scalar multiplication
 and the following properties must hold:
◮ additive identity (zero vector), additive inverse,
multiplicative identity
◮ associative, commutative, and distributive laws
results are vectors in vector space
 ⇒ linearity follows in this case:

x, y ∈ X ; a, b ∈ R(or C) → ax + by ∈ X

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 5/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Elementary operations (in a 2D linear space)
a) sum of two vectors
b) multiplication of a vector by a scalar

[Barry 2004]

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 6/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

Properties of the elementary operations in a vector space:
 Given vectors x, y, z ∈ X and scalars a, b ∈ R(or C):
a) Commutativity: x+y =y+x
b) Associativity: (x + y) + z = x + (y + z) and
abx = a(bx)
c) Distributivity: a(x + y) = ax + ay and
(a + b)x = ax + bx
d) Additive identity: There exists an element 0 ∈ X s.t.
x + 0 = 0 + x = x for every x ∈ X
e) Additive inverse: There exists a unique element −x ∈ X s.t.
x + (−x) = (−x) + x = 0 for every x ∈ X
f) Multiplicative identity: For every x ∈ X , 1x = x

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 7/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Subspaces in 3D Euclidean space X = R3
a) line (1D) ax ∈ X1 X1 ⊂ X = R3
b) plane (2D) ax + by ∈ X2 X2 ⊂ X = R3

[Barry 2004]

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 8/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 In digital communications:
 mapping of information onto sets of M waveforms
 signal space of M waveforms

S = span{s1 (t), s2 (t), ..., sM (t)}

 set of all signals, that can be represented as linear

combinations of these M waveforms
⇒ a subspace of continuous-time signals
 elements of this subspace; s(t) ∈ S
M
X
s(t) =α1 s1 (t) + α2 s2 (t) + · · · + αM sM (t) = αi si (t) ∈ S
i=1
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 9/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 E.g.: Signal space spanned by two signals s1 (t), s2 (t)
a) (basis) signals s1 (t), s2 (t)
b) examples of signals in S = span{s1 (t), s2 (t)}

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 10/54

al Processing and Speech Communications Lab [Barry 2004]
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 inner product (scalar product) −→ inner product space
 needed f. geom. interpretations of:
distance, angle, length (of/between vectors)
 E.g.: inner product on X = CN
N
X
hx, yi = xi yi∗ = yH x
i=1

x = [x1 , x2 , ..., xN ]T
y = [y1 , y2 , ..., yN ]T
∗
yH = [y1∗ , y2∗ , ..., yN ]
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 11/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

Properties of the inner product:
 Def.: The inner product maps two vectors to a scalar,
i.e. a = hx, yi, where x, y ∈ X , a ∈ C(or R)
 The following holds:
Distributivity: hx + y, zi = hx, zi + hy, zi
Linearity in the first argument: hax, yi = ahx, yi
(Conjugate linearity in the second: hx, ayi = a∗ hx, yi)
Hermitean symmetry: hx, yi∗ = hy, xi
Positive definiteness: hx, xi > 0 for x 6= 0

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 12/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Def: (induced) norm (squared length, “energy”)
PN
kxk = hx, xi = i=1 |xi |2 = xH x
2

 Def: angle (for real-valued vectors)

hx,yi
−1 ≤ kxk·kyk = cos(θ) ≤ 1
 distance: kx − yk

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 13/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Standard inner product spaces:
 DT signals (square-summable sequ.; X = ℓ2 (Z))
∞
X
hx[n], y[n]i = x[n]y ∗ [n]
n=−∞
X∞
kx[n]k2 = |x[n]|2 < ∞, ky[n]k2 < ∞
n=−∞

 CT signals (square-integrable functions; X = L2 (R))

Z ∞
hx(t), y(t)i = x(t)y ∗ (t)dt
−∞
Z ∞
kx(t)k2 = |x(t)|2 dt < ∞, ky(t)k2 < ∞
−∞
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 14/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Projection onto a subspace
 Def: subspace S ⊂ X
 for x ∈ X : projection of x onto S is unique element
x̂ ∈ S , that is “closest” to x

kx − x̂k = min kx − yk
y∈S
x̂ = arg min kx − yk
y∈S

 Projection Theorem:
 projection error x − x̂ must be orthogonal to the
subspace S

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 15/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Projection of signal r(t)
 Projection of x onto sub-
onto s1 (t), s2 (t)
space S

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 16/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 In digital communications: (review)
 mapping of information onto sets of M waveforms
 signal space of M waveforms

S = span{s1 (t), s2 (t), ..., sM (t)}

 set of all signals s(t) ∈ S , that can be represented as

linear combinations of these M waveforms, i.e.
M
X
s(t) = αi si (t)
i=1

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 17/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Orthonormal basis (of a signal space)
 minimum set of N orthonormal functions (N ≤ M )
that can be used to represent the elements s(t) ∈ S
(as linear combinations):
N
X
s(t) = si ψi (t)
i=1

 orthonormal (basis) functions {ψi (t)|i = 1, . . . , N }:

Z ∞ (
∗ 1 i=k
ψi (t)ψk (t)dt = δ[i−k] = → kψi (t)k = 1
−∞ 0 i 6= k

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 18/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Different bases for the signals s1 (t), s2 (t)

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 19/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Geometric representation of six waveforms in

three different bases
 geometric relations remain the same!

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 20/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 arbitrary signals ⇒ projection onto S , given an
orthonormal basis for S

r(t) = r̂(t) + e(t)

N
X
r̂(t) = ri ψi (t)
i=1

e(t) = r(t) − r̂(t) ... projection error

rj = hr(t), ψj (t)i ... coefficients
R∞
= −∞ r(t)ψj∗ (t)dt

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 21/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Gram-Schmidt orthogonalization
 find the N ≤ M orthonormal basis functions
1. ψ1 (t) = kss11 (t)k
(t)
, S1 = span{ψ1 (t)}
projection of s2 (t) onto S1
s2 (t)−ŝ2 (t)
2. ψ2 (t) = ks2 (t)−ŝ2 (t)k
, ŝ2 (t) = c21 ψ1 (t)
sk (t)−ŝk (t) Pk−1
k. ψk (t) = ksk (t)−ŝk (t)k , ŝk (t) = i=1 cki ψi (t)

no basis function, if sk (t) − ŝk (t) = 0

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 22/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-1 Signal Spaces (cont’d)

 Representation of M waveforms {sk (t)} through an
orthonormal basis ⇒ vectors

sk = [sk,1 , sk,2 , ..., sk,N ]T

Z ∞
where sk,i = sk (t)ψi∗ (t)dt = hsk (t), ψi (t)i
−∞

 operations on signals are equivalent to operations on

vectors (preservation of inner products; unitarity)

hsj (t), sk (t)i = hsj , sk i = sH

k sj

ksk (t)k2 = ksk k2 = sH

k sk

ksj (t) − sk (t)k2 = ksj − sk k2

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 23/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY
2-2 Linear Operators, Linear
Systems, and a Little Linear
Algebra
 Hilbert spaces: inner product spaces with properties
such as
 completeness: space contains all convergence
points of sequences
 separability: space contains a countable basis
⇒ standard spaces CN , ℓ2 (Z), L2 (R) are all separable
Hilbert spaces

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 24/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, ...

 Linear operator:
 generalizes finite-dimensional matrices
 used to perform operations with vector in-/outputs

 Def. of a linear operator (from H0 to H1 ):

 a function A : H0 → H1
 for all vectors x, y ∈ H0 and α ∈ C (or R),
 the following properties hold:
(i) additivity: A(x + y) = Ax + Ay
(ii) scalability: A(αx) = α(Ax)
H0 . . . domain; H1 . . . codomain
 Def.: linear operator on H0 : a function A : H0 → H0
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 25/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, ...

 a linear operator from CN to CM is the same as an
M × N matrix
 thus, concepts from linear algebra can be borrowed:

 e.g. the range of a lin. operator A : H0 → H1 is a

subspace of H1 :

R(A) = {Ax ∈ H1 |x ∈ H0 }

 e.g. the null space of a lin. operator A : H0 → H1 is a

subspace of H0 that A maps to 0:

N (A) = {x ∈ H0 |Ax = 0}

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 26/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, ...

 Def.: inverse
 an operator A : H0 → H1 is invertible if a linear
operator B : H1 → H0 exists s.t.:
(a) BAx = x for every x ∈ H0 and
(b) ABy = y for every y ∈ H1
 then B = A−1 is the inverse of A

 if only (a) holds, then B is the left inverse of A

 if only (b) holds, then B is the right inverse of A

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 27/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, ...

 Unitary operators preserve the geometry of vectors
(lengths and angles):
 Def.: A linear operator A : H0 → H1 that
(i) is invertible
(ii) preserves inner products

hAx, AyiH1 = hx, yiH0

for every for every x, y ∈ H0

 Prop. (ii) implies preservation of norms:
kAxk2 = kxk2 (Parseval theorem)
 Props. (i) and (ii) imply A−1 = AH
◮ (AH : H1 → H0 is the adjoint of A : H0 → H1 )
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 28/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, ...

 Orthogonal projection via pseudoinverse:
 projection onto R(A); least-squares approximation

 given A : H0 → H1 ; R(A) is a subspace of H1

 for any vector y ∈ H1 , how to find best approxim.
ŷ ∈ R(A) of y?
◮ find unique solution for ŷ = Ax that minimizes
ky − ŷk2
◮ projection theorem yields

hy − Ax, ai i = 0 ∀i; ai . . . i-th column of A

AH (y − Ax) = 0
AH Ax = AH y . . . normal equations
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 29/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, ...

 Orthogonal projection via pseudoinverse (cont’d):
 normal equations yield

x = (AH A)−1 AH y = By

B ... pseudoinverse of A
BA = I ... i.e. B is a left inverse of A
 furthermore:

ŷ = Ax = A(AH A)−1 AH y = Py

P = AB ... orthogonal projection onto R(A)

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 30/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, Linear

Systems, ...
 Eigenvectors and eigenvalues
 Def.: an eigenvector of an operator A : H → H is a
non-zero vector v ∈ H s.t.

Av = λv

for some λ ∈ C
λ . . . eigenvalue
(λ, v) . . . eigenpair

 for H ⊆ ℓ2 (Z), v is called eigensequence

 for H ⊆ L2 (R), v is called eigensignal

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 31/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, Linear

Systems, ...
 continuous-time (CT), linear systems
x T y

 T : L2 (R) → L2 (R) is a linear operator on L2 (R)

 i.e. y = Tx; x, y are CT functions x(t), y(t)

 in particular, we are interested in linear time-invariant

(LTI) systems
 Def. (LTI System):
y = Hx ⇒ y′ = Hx′ , where x′ (t) = x(t−τ ) and y ′ (t) = y(t−τ )

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 32/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-2 Linear Operators, Linear

Systems, ...
 Eigenfunctions of LTI systems (complex exponentials)

v(t) = ej2πf t

f ... frequency in Hz (cycles per second)

 system response becomes

(Hv)(t) = λf v(t) = H(f )ej2πf t

λf ... eigenvalue for v(t) at frequency f

H(f ) = λf ... frequency response of the LTI system

 this motivates frequency domain signal representations

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 33/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 Why? – develop intuition for many processing steps in
digital communications:
 sinusoidal signals are eigensignals of linear systems
 understanding the system response of LTI systems
 frequency occupation of communications signals
 separation of RF signals in frequency domain

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 34/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 Fourier series (complex-valued form)
 x(t) is periodic with period T , i.e. x(t) = x(t + T )
 functions with one period in (e.g.) L2 ([− 12 T, 12 T ))
∞
X
jk(2π/T )t 2π 1
x(t) = ck e , ω0 = , f0 = fundamental freq.
T T
k=−∞

 x(t) is represented in an orthogonal basis of complex

exponentials ejk(2π/T )t
 computation of the Fourier coefficients
D E 1 Z 12 T
ck = x(t), ejk(2π/T )t = x(t)e−jk(2π/T )t dt
T −1T
2
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 35/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 Fourier series properties
 given: Fourier series pair of a periodic signal
F
x(t) = x(t+T ) ←→ {ck }; ck ∈ C, ∀k ∈ Z; (i.e. {ck } ∈ ℓ2 (Z))

(the Fourier series reconstruction converges to x(t))

 Hermitean symmetry: for x(t) ∈ R: ck = c∗−k
 Parseval theorem (signal power)
Z T ∞
X
1
P = |x(t)|2 dt = |ck |2
T 0 k=−∞

(normalized) power of a spectral component is |ck |2

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 36/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 (LTI) system response in frequency domain
x(t) H(f ) y(t)

∞
X ∞
X
x(t) = ck ejk(2πf0 )t → y(t) = ck H(kf0 )ejk(2πf0 )t
k=−∞ k=−∞

H(f ) . . . frequency response of the LTI system

F
x(t) = x(t + T ) ←→ {ck }
F
y(t) = y(t + T ) ←→ {dk = ck H(kf0 )}
F
 Exploits linearity: αx(t) + βy(t) ←→ {αck + βdk }
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 37/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 E.g. periodic pulse train represented as Fourier series
 passed through first-order lowpass filter
Zeitsignal
Fourier series coefficients
0.3
1 ck ↔ x(t)

k
coefficients c , d
0.2

k
dk ↔ y(t)
.5 0.1

0
0
-0.1
-4 -2 0 2 4 6 8 -20 -15 -10 -5 0 5 10 15 20
time t index k (frequency k f )
0
Koeffizienten der Fourier Reihe
truncated reconstruction of the time-domain signals
.3
1.5
x(t)
signals x(t), y(t)

.2
1 y(t)
.1 0.5
0 0

1 -0.5
-20 -15 -10 -5 0 5 10 15 20 -4 -2 0 2 4 6 8
index k (frequency k/T0)
time t

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 38/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 nonperiodic signals:
 Fourier transform
Z ∞
X(f ) = x(t)e−j2πf t dt
−∞

 inverse Fourier transform

Z ∞
x(t) = X(f )ej2πf t df
−∞

 Fourier transform pair: (if both exist ∀t, ∀f ∈ R)

F
x(t) ←→ X(f ); x(t) ∈ L2 (R), X(f ) ∈ L2 (R)
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 39/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 Fourier transform properties
 given: Fourier transform pair
F
x(t) ←→ X(f )

 Hermitean symmetry: for x(t) ∈ R: X(f ) = X ∗ (−f )

 Parseval theorem (signal energy)
Z ∞ Z ∞
E= |x(t)|2 dt = |X(f )|2 df
−∞ −∞

|X(f )|2 . . . energy density at frequency f

 Note: E < ∞ for x(t) ∈ L2 (R)
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 40/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 Properties of the Fourier transform (cont’d)

linearity ax(t) + by(t) ↔ aX(f ) + bY (f )

 
time scaling x(at) ↔ 1
|a|
X fa
convolution x(t) ∗ y(t) ↔ X(f )Y (f )
x(t)y(t) ↔ X(f ) ∗ Y (f )
time shift x(t − τ ) ↔ X(f )e−j2πf τ
frequency shift ej2πf0 t x(t) ↔ X(f − f0 )
1 1
x(t) + x∗ (t) X(f ) + X ∗ (−f )



real part Re{x(t)} = 2
↔ 2
dm x(t)
↔ (j2πf )m X(f )
 dtm m
−jt dm X(f )
2π
x(t) ↔ df m
Rt 1 1
R∞
−∞ x(τ )dτ ↔ j2πf
X(f ) + 2
δ(f ) −∞ x(τ )dτ
1
x(t) cos(2πf0 t) ↔ 2
(X(f − f0 ) + X(f + f0 ))
X(t) ↔ x(−f )

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 41/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 Fourier transform pairs

ej2πf0 t ↔ δ(f − f0 )
δ(t − T ) ↔ e−j2πf T
1
cos(2πf0 t) ↔ 2
[δ(f − f0 ) + δ(f + f0 )]
1
sin(2πf0 t) ↔ 2j
[δ(f − f0 ) − δ(f + f0 )]
1
sinc(F t) ↔ F
rect (f, F/2)
rect(t, T /2) ↔ T sinc (f T )
1
e−αt u(t) ↔ j2πf +α
; Re {α} >0
δ(f ) 1
u(t) ↔ 2
+ j2πf
1 j

2
δ(t) + πt
↔ u(f )
P+∞ 1 P+∞ 1


k=−∞ δ(t − kT ) ↔ T m=−∞ δ f− T
m
1
jt
↔ −πsgn(f )

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 42/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 (LTI) system response in frequency domain
x(t) H(f ) y(t)

F F
x(t) ←→ X(f ) → y(t) ←→ Y (f ) = X(f )H(f )
H(f ) . . . frequency response of the LTI system
 System response in time domain:
 make use of convolution property:
F
y(t) = x(t) ∗ h(t) ←→ Y (f ) = X(f )H(f )

h(t) . . . impulse response of the LTI system (see below)

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 43/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 convolution integral (Faltungsintegral)
Z ∞
x(t) ∗ h(t) = x(λ)h(t − λ) dλ
−∞
Z ∞
= h(λ)x(t − λ) dλ
−∞

 convolution property of Fourier transform

F
y(t) = x(t) ∗ h(t) ←→ Y (f ) = X(f )H(f )

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 44/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 Dirac pulse, δ -pulse, Dirac distribution
 defined by integral (sampling, sifting property –
Ausblendeeigenschaft):
Z ∞
x(t)δ(t) dt = x(0)
−∞
Z ∞
δ(t) dt = 1
−∞

 Fourier transform:
F
δ(t) ←→ 1
F
δ(t − T ) ←→ e−j2πf T
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 45/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-3 Frequency Domain

Representation of Signals
 impulse response (of an LTI system):

x(t) = δ(t)
→ y(t) = δ(t) ∗ h(t)
Z ∞
= δ(λ)h(t − λ) dλ = h(t)
−∞

 time shifting property:

y(t) = δ(t − T ) ∗ x(t)

Z ∞
= δ(λ − T )x(t − λ) dλ = x(t − T )
−∞

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 46/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-4 Matrix factorizations

 LTI operator H on CN (circular convolution)
e
H = F−1 HF

F ... DFT matrix

F−1 ... inverse DFT F−1 = N1 FH
e
H ... frequency response (diagonal matrix)
 arbitrary lin. operator A on CN (full rank!)

A = VΛV−1 ... spectral theorem; EV decomposition

V ... Matrix of (N independent) eigenvectors

Λ ... eigenvalues (diagonal matrix)
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 47/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-4 Matrix factorizations

(cont’d)
 arbitrary lin. operator A : CN → CM
 singular value decomposition (SVD)
A = UΣVH
V ... (Unitary) matrix of N right singular vectors
Λ ... Singular values (diagonal matrix; sorted)
U ... (Unitary) matrix of M left singular vectors
 for A having rank r, there are r non-zero singular
values

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 48/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-5 Bandpass signals

 Definition:
 A bandpass (or narrowband) signal x(t) has an
f -domain representation X(f ) that is nonzero in a
(usually small) neighborhood of some (usually high)
frequency f0 ,

 i.e.: X(f ) = 0 for |f − f0 | ≥ W , where W ≪ fo

f0 . . . center frequency
W . . . bandwidth

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 49/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-5 Bandpass signals

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 50/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-5 Bandpass signals (cont’d)

 Bandpass signal
 Monochromatic signal  time-varying phasor
 has bandwidth W = 0
xl (t) = V (t)ejθ(t)
x(t) = A cos(2πf0 t + θ)
V (t) envelope
 represented by phasor θ(t) phase
xl (t) lowpass representation of x(t)
jθ
X = Ae

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 51/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-5 Bandpass signals (cont’d)

Relation between x(t) and xl (t):
 x(t) . . . bandpass signal
n o
x(t) = Re xl (t)ej2πf0 t

 z(t) . . . analytic signal

corresponding to x(t); pre-
envelope of x(t)
z(t) = xl (t)ej2πf0 t

 xl (t) . . . lowpass representa-

tion of x(t)
xl (t) = xc (t) + jxs (t)
Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 52/54
al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-5 Bandpass signals (cont’d)

 Generation and demodulation of bandpass signals (in
practice)

x(t) = xc (t) cos(2πf0 t) − xs (t) sin(2πf0 t)

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 53/54

al Processing and Speech Communications Lab
 GRAZ UNIVERSITY OF TECHNOLOGY

2-5 Bandpass signals (cont’d)

 Transmission of bandpass signals through LTI systems
 given: x(t), h(t), y(t) . . . input (bandpass), linear
system, output
 we find:
1
Yl (f ) = Xl (f )Hl (f )
2
1
yl (t) = xl (t) ∗ hl (t)
2
 system response can be computed using lowpass
equivalent representations of x(t), h(t), y(t)

Fund. of Digital CommunicationsChapter 2: Signals and Systems – p. 54/54

al Processing and Speech Communications Lab