EE330 Digital Signal Processing

Discrete Time Signals and Systems

Arbab Latif
Fall 2021

Resources:
Discrete Time Signal Processing, A. V. Oppenheim and R. W. Schaffer, 3rd Edition, 2010 (Chapter 2.5-2.9)
2 Last Week
• Signal
• Discrete Time Signals
• Discrete Time Systems
– Types / Properties
• Linear Time Invariant Systems
• Properties of LTI Systems
Linear Constant Coefficient
Difference Equation

• Section 2.5
4
Linear Constant Coefficient Difference Equation
(LCCDE) Representation:
• The relation between the input 𝑥[𝑛] and the output y[𝑛] of a
discrete-time (DT) system can be represented using a LCCDE
• It is DT equivalent of differential equations
• 𝑎0 𝑦 𝑛 + 𝑎1 𝑦 𝑛 − 1 + … + 𝑎𝑁 𝑦 𝑛 − 𝑁 = 𝑏0 𝑥 𝑛 + 𝑏1 𝑥 [𝑛 −
1ሿ + … + 𝑏𝑀 𝑥 𝑛 − 𝑀
• ∑𝑁 𝑎
𝑘=0 𝑘 𝑦 𝑛 − 𝑘 = ∑ 𝑀
𝑚=0 𝑏𝑚 𝑥 𝑛 − 𝑚
• The output at any time instant 𝑛 can be obtained as :
1
• 𝑦𝑛 = − { ∑𝑁 𝑎
𝑘=1 𝑘 𝑦 𝑛 − 𝑘 + ∑ 𝑀
𝑚=0 𝑏𝑚 𝑥 𝑛 − 𝑚 }
𝑎0
 LCCDE Representation of a Causal Moving Average
5
System:

• Moving Average System (an FIR System) is given by ?

1
• 𝑦𝑛 = ∑𝑀2
𝑥[𝑛 − 𝑘ሿ
𝑀1 +𝑀2 +1 𝑘=−𝑀1
• where 𝑀1 and 𝑀2 specify the use of future- and past-values of
𝑥[𝑛], respectively.
• For 𝑀1 = 0, it becomes a causal system.
• The LCCDE (∑𝑁 𝑀
𝑘=0 𝑎𝑘 𝑦 𝑛 − 𝑘 = ∑𝑚=0 𝑏𝑚 𝑥 𝑛 − 𝑚 ) of a causal
Moving Average System is obtained by setting ?
7 LCCDE Representation of Accumulator (Ex. 2.14):

• Accumulator system is defined by:

• 𝑦 𝑛 = ∑𝑛𝑘=−∞ 𝑥[𝑘ሿ
• The output for (n-1) can be specified as
• 𝑦 𝑛 − 1 = ∑𝑛−1𝑘=−∞ 𝑥[𝑘ሿ

• The LCCDE representing an Accumulator is

• 𝑦 𝑛 −𝑦 𝑛−1 =𝑥 𝑛
• 𝑖. 𝑒. 𝑁 = 1, 𝑎0 = 1, 𝑎1 = −1, and 𝑀 = 0, 𝑏0 = 1 .
9 LCCDE Representation of Accumulator (Ex. 2.14) (contd):

• The LCCDE representation of an Accumulator System gives

us a better understanding of how we could implement such a
system.
• 𝑦 𝑛 −𝑦 𝑛−1 =𝑥 𝑛
• 𝑜𝑟 𝑦 𝑛 =𝑥 𝑛 +𝑦 𝑛−1
• Note that its the recursive form of an Accumulator.
Frequency-domain Representation of
Discrete-time Signals and Systems

• Section 2.6
13 Eigen-functions for LTI Systems:
• Complex exponentials are eigenfunctions of LTI systems:
xn = e jn
• Let’s see what happens if we feed x[n] into an LTI system:
 
yn =  hkxn − k =  hk 
e j(n −k )

k = − k = −

• The eigenvalue is called the frequency response of the

system   − jk  jn
yn =   hk e j jn
e = He e ( )
 k = − 
• H e( )is a complex function of frequency
j

( ) =  hke

j − jk
He
k = −
16
Example 2.17: Frequency Response of the
Ideal Delay System:
• LCCDE Representation: y[n] = x[n − n d ]
• For input x[n] = e jn
• The output will be
j (n − nd ) j n − j nd
y[n] = e =e e
• The eigen-value is
j − j nd
H (e ) = e
H R (e j ) = cos( n d ) H (e j ) = 1
j
H I (e ) = − sin( n d ) H (e j ) = − n d

(Real & Imaginary part representation) (Magnitude & phase representation)

17
Example 2.18: Sinusoidal Response of LTI
Systems
• For input x[n] = A cos(o n +  )
A j jo n A − j − jo n
x[n] = e e + e e
2 2 (Inverse Euler’s Formula)
A j jo n A j
e e jw
e H (e jo n ) e jo n
2 H (e )
2
A − j − jo n A − j
e e jw
e H (e − jo n ) e − jo n
2 H (e )
2

• The general output will be

A
y[ n] = [H(e jo ) e j e jo n + H(e − jo ) e − j e − jo n ]
2
18 Example 2.18 (Contd)

• Assuming an ideal-delay system

j − j nd
H (e ) = e

• The output will be

A − jo nd j jo n
y[ n] = [e e e + e + jo nd e − j e − jo n ]
2
y[n] = A cos(o n +  − o n d )

y[n] = A cos[o (n − n d ) +  ]
Note: Output is the input delayed by nd
20
Examples of Frequency Responses:
1) Ideal Lowpass Filter
H lp (e j )
One period of the frequency response
1 Fundamental Period

− −c 0 c  

H lp (e j )
Periodicity of the frequency response
1

−2 −2 + c − −c 0 c  2 − c 2 

21
Examples of Frequency Responses:
2) Ideal Highpass Filter

H hp (e j ) Fundamental Period

− −c 0 c  
Representation of Sequences
by Fourier Transform

(Section 2.7)
26 Discrete-Time Fourier Transform (DTFT)

• Fourier analysis equation (Time → Freq)

The Fourier-Transform Equation

• Fourier synthesis equation (Freq → Time)

1
x[n] =
2 2
 X (e jw ) e jn dw

Inverse Fourier-Transform Equation

27
DTFT is in general a complex-valued
function
j j j
X (e ) = X R (e ) + j X I (e )
(Rectangular Form: Real & Imaginary part representation)

j j jX (e j )
X (e ) = X (e ) e
(Polar Form: Magnitude & phase representation)
29 Frequency Response & DTFT
• From the last section on Freq. Response
and the definition of the DTFT, it is easy to
see that
“ Frequency-response is the DTFT of the
impulse response” 
H(e jw ) =  h[n]e− jn
and consequently k =−

“Impulse response is the Inverse-DTFT of

the frequency response.”
1
h[n] =
2 
[2 ]
H(e jw ) e jn dw
31
Sufficient Condition for existence –
Absolute Summability:
• If 𝑥[𝑛ሿ is absolutely summable then 𝑋(𝑒 𝑗𝜔 ) exists.
• Stated otherwise, the DTFT exists for all
absolutely summable sequences, i.e. ,

• ∑∞
𝑛=−∞ 𝑥[𝑛ሿ < ∞

– Moreover for such sequences, the DTFT converges

uniformly to a continuous function of frequency.

• Corollaries:
– All stable LTI systems have continuous frequency
responses.
– All finite-duration sequences are stable and have
finite continuous DTFT.
32
Example 2.21 Absolute Summability for
a Suddenly-applied Exponential:
• Consider the sequence,
x[n] = a u[n]
n

Determine the DTFT when it exists.

----------------------------------------------------
  n

(
X (e jw ) =  a n e− jn =  ae− j )
n =0 n =0

1
X (e ) =
jw
− j
, if a  1
1 − ae

(use geometric series sum formula)

For plots , see Fig. 2.22 (Pg. 57)
34
Mean-square Convergence for ‘square
summable’ sequences:
• For square-summable sequences, we
are guaranteed to have mean-square
convergence, i.e., for
 M
j
X (e ) :=  x[n]e
n =−
− jn
,
j
X M (e ) := 
n =− M
x[n]e− jn

we can be sure that


lim

j j 2
X (e ) − X M (e ) d  = 0
M →  −

•
36
Example 2.22: Impulse Response of
Ideal Lowpass Filter:
• Frequency-response of Ideal lowpass
filter:
j
1,   c
H lp (e ) = 
0, c    
----------------------------------------------------
• Impulse Response:

1
hlp [n] =
2 −
 H (e j )e + j n d 

1 c + jn
= 
2 −c
e d

sin(c n)
= , −  n  .
n
37 Example 2.22: SpecialCase-
= /4 c

sin(c n)
h[ n] =
n
I-DTFT of Ideal Lowpass Filter
0.25

0.2
h[n]
0.15 Continuous Envelope

0.1
h[n]

0.05

0 n
-0.05

-0.1
-20 -15 -10 -5 0 5 10 15 20
n
c •Zero-crossings: Multiples of 4 (except at n=0)
h[0] =
 •Decays at the rate of 1/n (Not Abs. Summable)
38
Example 2.22: Gibbs Phenomenon (Non-
uniform & Discontinuous Convergence)
c =  / 2
1.5 1.5
1 1
H1(ej)

H3(ej)
0.5 0.5
0 0
-0.5 -0.5
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
/ /
1.5 1.5
1 1
H19(ej)
H7(ej)

0.5 0.5
0 0
-0.5 -0.5
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
/ /
40 Generalized Functions:

• DTFT of ‘some’ sequences which are neither

absolutely nor square summable can be
defined in certain special cases using the
“Theory of generalized functions”.
• Examples: Constants, Complex exponentials,
Unit step sequence, etc.
• No guarantee for Uniform/Mean-square
convergence. Only back substitution in
Inverse-DTFT equation works.

40
41
Example 2.19: Fourier Transform of
a Constant:
• Consider the sequence: 𝑥 𝑛 = 1, ∀𝑛 ∈ 𝑍
• DTFT?
--------------------------------------------------
• 𝑥 𝑛 is abs. summable or sq. summable? (No)

• Yet, it is possible and useful to define its DTFT to be the

periodic impulse train ?
• 𝑋 𝑒 𝑗𝜔 = ∑∞ 𝑟=−∞ 2𝜋𝛿(𝜔 + 2𝜋𝑟)

Thus, we get
X (e j ) = 2 ( ), for    .
42
Example 2.24 Fourier Transform of Complex
Exponential Sequences:

• Consider :
X (e j ) = 2   ( − 
r =−
o + 2 r )
• What is the Inverse-DTFT?
------------------------------------------------------
1 

j n
x[n] = 2 ( − 0 ) e d 
2 −
=   ( − 0 ) e d j n
− Sifting Property

   ( −  ) d  = e
jo n jo n
x[n] = e 0
−

jo n
e  2 ( − o ),     .
43 DTFT for sum of exponentials:
• Using the theory of generalized functions
(not covered here), we can define the DTFT
for
• 𝑥 𝑛 = ∑𝑘 𝑎𝑘 𝑒 𝑗𝜔𝑘 𝑛 , ∀𝑛 ∈ ℤ.

to be 
j
X (e ) =   2 a  ( − 
r =− k
k k + 2 r )

• Special Case: Sum of Harmonics (𝜔𝑘 = 𝑘𝜔0 )

– DTFT representation of periodic sequences in
terms of Discrete Fourier Series (DFS)
coefficients.
44 DTFT for Unit-step Sequence:
• Neither Absolutely nor square summable

• Using ‘theory of generalized functions’, the

DTFT of u[n] can be shown to be


1
U (e j ) = − j
+   ( + 2 r )
1− e r =−

(The proof of above eq. is out of the scope of this course)

45 Summary for Convergence:
Property Satisfied in the Time- Guaranteed results in the
domain Frequency-domain

Absolute Summability DTFT exists & converges uniformly

(e.g., All FIR Systems) to a continuous function of
frequency.

Square Summable (finite energy) but DTFT exists & only Mean-square
no absolutely summable. convergence is guaranteed. The
resulting DTFT may be
1 sin(o n)
e.g., x[n] = u[n − 1], discontinuous (e.g., Ideal lowpass
n n filter).
Sequences which are not finite DTFT may not exist at all. In few
energy cases, the DTFT may be expressed
(e.g., x[n]=1 or unit step sequence, or as sum of impulse trains
periodic signals) (generalized functions).
46
Sampling & Periodicity across Time
and Frequency:
Time Frequency

Continuous & Aperiodic in Time: A Aperiodic & Continuous CTFT: C

Example: P 1 O
− at
e u (t ) E
a + j
N
T
R
I I
Discrete & Aperiodic in Time: Periodic & Continuous DTFT: N
O
U
Example: D 1 O
I
1 − ae − j
U
a n u[n] C S

Continuous & Periodic in Time: Aperiodic (in general) & Discrete

P (impulses): D
cos(ot ) E 1 1
 (  −  o ) +  ( +  o )
I
R S
I
2 2 C
Discrete & Periodic in Time: O Periodic & Discrete: (Sum of R
E
D impulse trains)
cos(o n)
T
I 1 1 E
C  ( − o ) +  ( + o ),    .
2 2
47
Sampling & Periodicity across Time
and Frequency:
Time Frequency

Continuous & Aperiodic in Time: Aperiodic & Continuous CTFT:

Example: 1
− at
e u (t ) a + j
C A
O P
Discrete & Aperiodic in Time: N Periodic & Continuous DTFT: E
Example: T
I 1 R
I
D a n u[n] N
U
1 − ae − j O P
I O D E
S Continuous & Periodic in Time: U Aperiodic (in general) & Discrete I R
C S (impulses): C I
cos(ot )
R
1 1
E  (  −  o ) +  ( +  o ) O
T 2 2 D
E I
Discrete & Periodic in Time: Periodic & Discrete: (Sum of C
impulse trains)
cos(o n) 1 1
 ( − o ) +  ( + o ),    .
2 2
Symmetry Properties of The
Fourier Transform

(Section 2.8)
50 Symmetries for Real-valued Sequences:

• Even Sequences:
– Symmetric about vertical axis (Mirror image)
– 𝑥𝑒 𝑛 = 𝑥𝑒 −𝑛 , ∀ 𝑛 ∈ 𝑍

– Examples: cos(o n), n .2

• Odd Sequences:
– Symmetric about the origin (180 degree rot.)
xo [n] = − xo [−n],  n 
– Examples:
sin(o n), n .
51 Symmetries for Complex-valued Sequences:

• Conjugate-Symmetric Sequences:
– Aka ‘Hermitian’
xe [n] = xe*[−n]
– Example:
j n
x[n] = e
• Conjugate-AntiSymmetric Sequences
– Aka ‘Skew-Hermitian’
xo [n] = − x [−n]
*
o
– Examples:
x[n] = sin(o n) + jcos(o n)
55 Symmetry Properties (General):
Signal DTFT X( e𝑗𝜔 )
𝑥[nሿ

1 x*[n] X * (e− j )

x*[− n] X * (e j )
2

3 Re{x[n]} X e (e j )

jIm{x[n]} X o (e j )
4

5 xe [n] X R (e j ) = Re{X(e j )}

6 xo [n] jX I (e j ) = j Im{X(e j )}
56
Symmetry Properties for Real
sequences:
Signal DTFT
𝑥[nሿ X( e𝑗𝜔 )

For any real 7 X (e j ) = X * (e− j ) ( DTFT is Conj. Symm. )

sequence
x[n]
8 X R (e j ) = X R (e− j ) ( Real part is even )

( Imag. part is Odd )

9 X I (e j ) = − X I (e− j )

10 X (e j ) = X (e− j ) ( Mag. is even )

j − j
11 X (e ) = −X (e )
( Phase is odd )

12 xe [n] ( even part of x[n] ) X R (e j )

13 xo [n] ( odd part of x[n] ) jX I (e j )

Fourier Transform Theorems

(Section 2.9)
72 Properties of DTFT:
Relationship in Time Relationship in Frequency
Linearity Property:
1 ax[ n] + by[n] aX (e j ) + bY (e j )
Shift in time: Phase-factor in freq:
2
𝑥 𝑛 − 𝑛𝑑 , 𝑛𝑑 ∈ ℤ X (e j ) e − jnd
Mult. by Exp.: jo n
Shift in freq:
3 x[n]e X (e j ( −o ) )
Time-reversal:
4 x[ − n] X(e − j ) (for x[n] real,X
*
(e j ) )
Multiplication by n: Differentiation in Freq dX (e j )
5 nx[ n ] j
d
Convolution in Time: Multiplication in freq:
6 x[ n]  y[n] X (e j ) Y(e j )
Multiplication in Time: Conv: 1 
7 x[ n] y[n]
2 
−
X (e j )Y (e j ( − ) ) d 
73 Parseval’s Theorem (Plancherel Theorem):

• Energy in time-domain is equal to the

energy in frequency-domain.
 
1
 
j 2
x[n] = X (e ) d
2
8
n =− 2 −

• Inner product is equal in both domains:

 
1
9 
n =−
x[n] y [n] =
*

2 
−
X (e j ) Y* (e j )d
82 Fourier Transform Pairs:
Sequence DTFT
(Periodic with period 𝟐𝝅)

 [n] 1
(Use DTFT Eq.)
− j no
 [n − no ] e
(Example 2.14)

1, −  n    2 ( + 2 k)
k =− (Example 2.19)
1
a u[n], a  1
n

1 − ae − j (Example 2.17)

1
u[ n] − j
+   ( + 2 k)
1− e k =− (Eq. 2.148)
83 Fourier Transform Pairs (Contd.):
Sequence DTFT
(Periodic with period 2*pi)
1
(n + 1) a u[n], ( a  1)
n
(1 − ae − j ) 2
(Use Diff. in Freq. Prop.)
sin( p (n + 1)) 1
r n
u[n], ( r  1)
sin( p ) 1 − 2r cos( p ) e− j + r 2e− j 2
(Use Mod. Prop.,&DTFT of sin(.))

sin(c n) j

1,   c
X (e ) = 
n (Ex. 2.18) 0, c    


1, 0  n  M sin( (M + 1) / 2) − j M /2
x[n] =  e
0, otherwise sin( / 2)
(MA Ex. 2.16)
84 Fourier Transform Pairs (Contd.):

Sequence DTFT
(Periodic with period 2*pi)


e jo n
 2 ( − 
k =−
o + 2 k )
(Example 2.20)

cos(o n +  ) 

 [
k =−
e j
 ( − o + 2 k ) +  e − j
 ( + o + 2 k )]
(Use Euler’s Identity & above prop.)