ELC 4351: Digital Signal Processing

Liang Dong

Electrical and Computer Engineering

Baylor University
liang [email protected]

March 16, 2017

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 1 / 25
Frequency Analysis of Signals

1 Frequency Analysis of Continuous-Time Signals

Frequency Analysis of Continuous-Time Signals
Power Density Spectrum of Periodic Signals
The Fourier Transform for Continuous-Time Aperiodic Signals
Energy Density Spectrum of Aperiodic Signals
2 Frequency Analysis of Discrete-Time Signals
The Fourier Series of Discrete-Time Periodic Signals
Power Density Spectrum of Periodic Signals
The Fourier Transform of Discrete-Time Aperiodic Signals
Convergence of the Fourier Transform
Energy Density Spectrum of Aperiodic Signals
Relationship of the Fourier Transform to the z-Transform
Frequency-Domain Classification of Signals: The Concept of
Bandwidth

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 2 / 25
The Fourier Series for Continuous-Time Periodic Signals

A linear combination of harmonics (harmonically related complex

exponentials):

Synthesis Equation
∞
X
x(t) = ck e j2πkF0 t
k=−∞

Analysis Equation
Z
1
ck = x(t)e −j2πkF0 t dt
Tp Tp

where, the fundamental period is Tp = 1/F0 .

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 3 / 25
The Fourier Series for Continuous-Time Periodic Signals

A linear combination of cosine functions, if signal x(t) is real:

Synthesis Equation
∞
X
x(t) = a0 + (ak cos2πkF0 t − bk sin2πkF0 t)
k=1

where

a0 = c 0
ak = 2|ck |cosθk
bk = 2|ck |sinθk
ck = |ck |e jθk

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 4 / 25
The Fourier Series for Continuous-Time Periodic Signals

The Dirichlet conditions guarantee that x(t) and its Fourier series
representation are equal at any value of t:
1 x(t) has a finite number of discontinuities in any period.
2 x(t) contains a finite number of maxima and minima during any
period.
R
3 x(t) is absolutely integrable in any period, i.e. Tp |x(t)|dt < ∞.

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 5 / 25
Power Density Spectrum of Periodic Signals

A periodic signal has a finite average power

Z
1
Px = |x(t)|2 dt
Tp Tp
∞
1X 2
Z
1
= x(t)x ∗ (t)dt Px = a02 + (ak + bk2 )
Tp Tp 2
k=1
Z ∞
1 X
= x(t) ck∗ e −j2πkF0 t dt
Tp Tp
k=−∞
∞
" Z #
X 1
= ck∗ x(t)e −j2πkF0 t dt
T p Tp
k=−∞
X∞
= |ck |2 (Parseval0 s Relation)
k=−∞

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 6 / 25
The Fourier Transform for Continuous-Time Aperiodic
Signals

Going from periodic signal to aperiodic signal, we make the period

Tp → ∞.

x(t) = lim xp (t)

Tp →∞
X∞
xp (t) = ck e j2πkF0 t , F0 = 1/Tp
k=−∞
Z Tp /2
1
ck = x(t)e −j2πkF0 t dt
Tp −Tp /2
Z ∞
1
= x(t)e −j2πkF0 t dt
Tp −∞
| {z }
X (F )

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 7 / 25
The Fourier Transform for Continuous-Time Aperiodic
Signals
We write F , kF0 = k/Tp and ∆F , F0 = 1/Tp .
As Tp → ∞, ∆F = dF . Therefore
∞
1 X
xp (t) = X (F )e j2πkF0 t
Tp
k=−∞
∞
X
= X (k∆F )e j2πkF0 t ∆F
k=−∞
x(t) = lim xp (t)
Tp →∞
∞
X
= lim X (k∆F )e j2πkF0 t ∆F
∆F →0
k=−∞
Z ∞
= X (F )e j2πFt dF
−∞

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 8 / 25
The Fourier Transform for Continuous-Time Aperiodic
Signals

Synthesis Equation (Inverse Transform)

Z ∞
x(t) = X (F )e j2πFt dF
−∞

Analysis Equation (Direct Transform)

Z ∞
X (F ) = x(t)e −j2πFt dt
−∞

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 9 / 25
Energy Density Spectrum of Aperiodic Signals
R∞ 2
Signal Energy: Ex = −∞ |x(t)| dt

Z ∞
Ex = x(t)x ∗ (t)dt
−∞
Z ∞ Z ∞ 
∗ −j2πFt
= x(t)dt X (F )e dF
−∞ −∞
Z ∞ Z ∞ 
∗ −j2πFt
= X (F )dF x(t)e dt
−∞ −∞
Z ∞
= X ∗ (F )X (F )dF
Z−∞
∞
= |X (F )|2 dF
−∞

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 10 / 25
Energy Density Spectrum of Aperiodic Signals

Parseval’s Relation
Z ∞ Z ∞
2
Ex = |x(t)| dt = |X (F )|2 dF
−∞ −∞

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 11 / 25
Energy Density Spectrum of Aperiodic Signals

Energy Density Spectrum:

Sxx (F ) , |X (F )|2

Therefore, Sxx (F ) ≥ 0, for all F .

If signal x(t) is real, |X (−F )| = |X (F )| and ∠X (−F ) = −∠X (F ). It

follows that
Sxx (−F ) = Sxx (F )

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 12 / 25
The Fourier Series of Discrete-Time Periodic Signals

x(n) is periodic with period N. That is, x(n) = x(n + N) for all n.

A linear combination of N harmonically related exponents:

Synthesis Equation
N−1
X
x(n) = ck e j2πkn/N
k=0

Analysis Equation
N−1
1 X
ck = x(n)e −j2πkn/N
N
n=0

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 13 / 25
The Fourier Series of Discrete-Time Periodic Signals

The Fourier series coefficients {ck } is a periodic sequence with

fundamental period N (when extended outside the range [0, N − 1]).
N−1
1 X
ck+N = x(n)e −j2π(k+N)n/N
N
n=0
N−1
1 X
= x(n)e −j2πkn/N
N
n=0
= ck

The spectrum of x(n) is a periodic sequence with period N.

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 14 / 25
The Fourier Series of Discrete-Time Periodic Signals

A linear combination of cousin functions, if signal x(n) is real:

Synthesis Equation
L
X
x(n) = a0 + 2 (ak cos(2πkn/N) − bk sin(2πkn/N))
k=1

where

a0 = c 0
ak = 2|ck | cos θk
bk = 2|ck | sin θk

N/2 if N is even
L =
(N − 1)/2 if N is odd

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 15 / 25
Power Density Spectrum of Periodic Signals
The average power of a discrete-time periodic signal with period N:
N−1
1 X
Px = |x(n)|2
N
n=0
N−1
1 X
= x(n)x ∗ (n)
N
n=0
N−1 N−1
!
1 X X
= x(n) ck∗ e −j2πkn/N
N
n=0 k=0
N−1 N−1
" #
X 1 X
= ck∗ x(n)e −j2πkn/N
N
k=0 n=0
N−1
X
= |ck |2
k=0

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 16 / 25
Power Density Spectrum of Periodic Signals

Energy over a signal period:

N−1
X N−1
X
2
EN = |x(n)| = N |ck |2
n=0 k=0

If x(n) is real, ck∗ = c−k . Equivalently, |c−k | = |ck | and −∠c−k = ∠ck .

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 17 / 25
The Fourier Transform of Discrete-Time Aperiodic Signals
Analysis Equation
∞
X
X (ω) = x(n)e −jωn , ω ∈ [−π, π) or ω ∈ [0, 2π)
n=−∞

Synthesis Equation
Z π
1
x(n) = X (ω)e jωn dω
2π −π

X (ω) is periodic with period 2π:

∞
X
X (ω + 2πk) = x(n)e −j(ω+2πk)n
n=−∞
X∞
= x(n)e −jωn = X (ω)
n=−∞
Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 18 / 25
Convergence of the Fourier Transform

N
X
XN (ω) = x(n)e −jωn
n=−N

Uniform convergence:

lim {sup |X (ω) − XN (ω)|} = 0, for all ω

N→∞ ω
P∞
Uniform convergence is guaranteed if n=−∞ |x(n)| < ∞.

Mean-square convergence:
Z π
lim |X (ω) − XN (ω)|2 dω = 0, for all ω
N→∞ −π

P∞ 2
Mean-square convergence is for finite-energy signals n=−∞ |x(n)| < ∞.

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 19 / 25
Energy Density Spectrum of Aperiodic Signals

The energy of a discrete-time signal x(n):

∞
X
Ex = |x(n)|2
n=−∞
X∞
= x(n)x ∗ (n)
n=−∞
∞ Z π
 
X 1 ∗ −jωn
= x(n) X (ω)e dω
n=−∞
2π −π
Z π " ∞ #
1 X
= X ∗ (ω) x(n)e −jωn dω
2π −π n=−∞
Z π
1
= |X (ω)|2 dω
2π −π

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 20 / 25
Energy Density Spectrum of Aperiodic Signals

Energy Density Spectrum:

Sxx (ω) , |X (ω)|2

If x(n) is real, X ∗ (ω) = X (−ω). Equivalently, |X (−ω)| = |X (ω)| and

∠X (−ω) = −∠X (ω). It follows that

Sxx (−ω) = Sxx (ω)

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 21 / 25
Relationship of the Fourier Transform to the z-Transform

z-Transform
∞
X
X (z) = x(n)z −n ; ROC: r2 < |z| < r1
n=−∞

z in polar form: z = re jω . We have

∞
X
X (z) = [x(n)r −n ]e −jωn
n=−∞

If X (z) converges for |z| = 1,

∞
X
X (z) |z=e jω = X (ω) = x(n)e −jωn
n=−∞

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 22 / 25
Relationship of the Fourier Transform to the z-Transform

∞
X
X (z) |z=e jω = X (ω) = x(n)e −jωn
n=−∞

Fourier transform can be viewed as the z-transform of the sequence

evaluated on the unit circle.

If X (z) does not converge in the region |z| = 1, the Fourier transform
X (ω) does not exist.

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 23 / 25
Frequency-Domain Classification of Signals: The Concept
of Bandwidth

Power (energy)density spectrum

 low-frequency
concentration high-frequency
bandpass


Bandwidth — a quantitative measure

Suppose a continuous-time signal has 90% of its power (energy) density
spectrum in range F1 < F < F2 . The 90% bandwidth of the signal is
F2 − F1 .
Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 24 / 25
Frequency-Domain Classification of Signals: The Concept
of Bandwidth

F1 +F2
Narrowband: F2 − F1  2 (median frequency)
Wideband: Otherwise

X (F ) = 0 for |F | > B
Bandlimited:
X (ω) = 0 for ω0 < |ω| < π

No signal can be time-limited and band-limited simultaneously.

(Reciprocal relationship)

Liang Dong (Baylor University) Frequency Analysis of Signals I March 16, 2017 25 / 25