Course : Digital Signal Processing

Course Code : 22UECC502

Credit :4
Hrs :52

By
Mr. Shrikanth Shirakol
e q u i s i te:
Assistant Professor, Dept. of E&CE Prer n d
a l s a
SDMCET, Dharwad
Sign s
System
Karnataka.
[email protected]
[email protected]
 Course Learning Objectives (CLOs):

• The course focuses on Discrete Fourier

Transforms, properties and their applications.
It also deals with the design of analog and
digital filters using various methods and their
realization.

Dept. of E&CE, SDMCET, Dharwad 2

 Course Outcomes (COs)
Mapping to POs(1-12)/ PSOs (13,14)
Description of the Course Outcome:
At the end of the course the student
Substantial Moderate Slight
will be able to:
Level (3) Level (2) Level (1)

Compute transforms and inverse

transforms of a signal and apply
CO-
the properties of DFT to solve 1,2 - 13
1
signal processing problems

Optimize the computation of

CO-
DFT and Compare with direct 2, 3, 13, 12 -
2
computations.
Design analog and digital IIR
CO- filters to satisfy the given
2, 3 12 -
3 specifications and realize the
hardware structure
Dept. of E&CE, SDMCET, Dharwad 3
 Mapping to POs(1-12)/ PSOs
Description of the Course Outcome: (13,14)
At the end of the course the student will be able
Slight
to: Substantial Moderate
Level
Level (3) Level (2)
(1)
Design digital FIR filters to satisfy the
CO-4 given specifications and realize the
2, 3 12 -
hardware structure.
Design of filters using pole-zero concept
and to understand the fundamentals of
CO-5 multirate signal processing and its 2, 3 12 -
applications.

Dept. of E&CE, SDMCET, Dharwad 4

•
Unit-I

Discrete Fourier Transform (DFT): Introduction, frequency domain sampling and

reconstruction of discrete time signals, Discrete Fourier Transform, DFT as a linear
transformation, relationship of DFT to other transforms, Properties of DFT, Linear
Filtering Methods based on DFT: use of DFT in linear filtering, filtering of long data
sequences, frequency analysis of signals using DFT 10 Hrs

Unit-II

Efficient Computation of the DFT: Radix-2 Fast Fourier Transform (FFT)

algorithms for the computation of DFT and IDFT: Decimation-In-Time (DIT) and
Decimation-In-Frequency (DIF) algorithms, comparison of direct computation and
FFT computation of DFT, applications of FFT algorithms: efficient computation of
DFT of two real sequences, efficient computation of DFT of 2N point real
sequences. Discrete Cosine Transform(DCT): Forward DCT, Inverse DCT, DCT as
an orthogonal transform. 10 Hrs

Dept. of E&CE, SDMCET, Dharwad 5

 Unit-III

Design of Analog IIR Filters: Introduction, design of IIR Butterworth and

Chebyshev type–I filters . Design of Digital IIR Filters: IIR filter design by
Approximation of Derivatives, Impulse Invariance Transformation, frequency
transformations in digital domain, structures for the realization of IIR filters.
12 Hrs

Unit-IV

Design of Digital FIR Filters: Symmetric and Anti-symmetric FIR filters,

Design of Linear phase FIR filters using windows method and frequency
sampling method, Design of FIR Differentiators, Design of Hilbert Transformers,
structures for the realization of FIR filters. 10 Hrs

Unit-V

Pole-Zero Placement Method for the Design of Simple Filters: Ideal filter
characteristics, Simple IIR & FIR digital filters, Notch filters, Comb filters, All-
pass filters, Digital Resonators. Multirate Digital Signal Processing:
Introduction, Decimation, Interpolation, sampling rate conversion, applications
of multirate signal processing. 10 Hrs
Dept. of E&CE, SDMCET, Dharwad 6
 Reference Books
1) Proakis & Monalakis, “Digital Signal Processing: Principles, Algorithms &
Applications”, 4/e, Pearson Education, New Delhi, 2007.

2) Sanjit K. Mitra, “Digital Signal Processing”, 2/e Tata Mc-Graw Hill, 2004.

3) Li Tan, “Digital Signal Processing Fundamentals and Applications”,

Elsevier,2003.

4) Emmanuel C. Ifeachor, Barrie W. Jervis, “Digital Signal Processing: A

Practical Approach”, Pearson Education,2/e, 1999.

5) P. P. Vaidyanathan, “Multirate Systems and Filter Banks”, Pearson, 1992.

Slides Ref: 1) Dr. Vijaya C, Prof & Head, Dept. of E&CE, SDMCET
2) Prof. Kotresh Marali. Asst. Prof., Dept. of E&CE,
SDMCET
Dept. of E&CE, SDMCET, Dharwad 7
 Unit-I
Discrete Fourier Transform (DFT):
• Introduction to Digital Signal Processing

• Frequency domain sampling and reconstruction of discrete time (DT)

signals

• Discrete Fourier Transform

• Properties of DFT

• Linear Filtering Methods based on DFT

• frequency analysis of signals using DFT

07Hrs
Dept. of E&CE, SDMCET, Dharwad 8
 Introduction to Digital Signal Processing

• What is Digital Signal Processing (DSP)?

• Why Digitalization ?

Dept. of E&CE, SDMCET, Dharwad 9

Introduction to Digital Signal Processing

Dept. of E&CE, SDMCET, Dharwad 10

 What is Digital Signal Processing?
• Digital Signal Processing (DSP) is used to manipulate real
world signals after they have been converted to digital form.

Analog Digital Signal Analog

ADC Processor DAC Signal
Signal

Figure-1: Block Diagram of Digital Signal Processing

Dept. of E&CE, SDMCET, Dharwad 11

Importance of Digitalization

Dept. of E&CE, SDMCET, Dharwad 12

 Why digitization?
• No two analog systems with same design will have same responses
– Analog components temperature dependent
– Differ from each other even with the same manufacturing
conditions
– Component aging changes the circuit response with time

• Change in system specifications requires change in entire hardware

• System size dependent on system specifications

• Difficult to reproduce systems in large quantities

– need for adjusting component values of each circuit

Dept. of E&CE, SDMCET, Dharwad 13

 Revisit to Signals
Types
TypesofofSignals
Signal

Periodic Aperiodic/
(F.S.) Non-Periodic (F.T.)

C.T Signal D.T. Signal C.T Signal D.T. Signal

C.T.F.S D.T.F.S C.T.F.T D.T.F.T

Dept. of E&CE, SDMCET, Dharwad 14

Note:
• Frequency analysis of D.T. Signals x[n] are usually
carried out using DSP i.e., “Digital Signal
Processors”.
• To perform frequency analysis on D.T., signals, they
are to be converted into an equivalent frequency
domain representation.
x[n] x[ejw]

Dept. of E&CE, SDMCET, Dharwad 15

 The Fourier Transform of Discrete-Time
Aperiodic Signals
• The F.T. of D.T. aperiodic signal x[n] is
continuous function of ꞷ and is periodic with
period 2π

Dept. of E&CE, SDMCET, Dharwad 16

Frequency Domain sampling and Reconstruction of
Discrete-Time Signals

Dept. of E&CE, SDMCET, Dharwad 17

Frequency Domain sampling and Reconstruction of
Discrete-Time Signals

Dept. of E&CE, SDMCET, Dharwad 18

 Frequency sampling
• Need for frequency domain sampling
• Processor used for signal processing task is digital
– Understands only numerals.
• Let x(n) be a discrete, non periodic signal of finite
energy. 
X ( )   x(n)e  jn (1)
• Its Fourier Transform: n 

• Frequency variable is a continuous variable and

hence FT, continuous function in . It is digital
frequency. Points on circumference of unit circle
in z plane. 2
• FT is periodic with period
 Frequency sampling
• To discretize FT, digital frequency is divided into N
equal parts.
• Frequency spacing between each sample points=
2
  radians
N
• Any sample is identified as k  2k radians
N

• Thus discretized version of FT becomes

2kn
2k 2k   j
X (k  ) X ( )   x ( n )e N
(2)
N N n 
 DFT and IDFT Definition
DFT 2k
X (k )  X ( )  X (k )
N
 2kn
 j
X ( k )   x ( n )e N
k 0,1,2, ...., N - 1
n 
N1
X (k )  x(n)WNkn ; k 0,1...N  1
n 0
IDFT 1  j
2kn
x ( n) 
N
 X ( k )e
n 
N
; n 0,1,2,....., N - 1

N1
1
x ( n) 
N
 X
k 0
( k )W  kn
N ; n 0,1...N  1
 DFT-IDFT
• x(n) is a discrete signal of length N. Its N point
DFT is
N1
X (k )  x(n)WNkn ; k 0...N  1 (1)
n 0

• And inverse DFT is

N1
1
x ( n) 
N
 X
k 0
( k )W  kn
N ; n 0...N  1 (2)
 Answer the following
• What is the need for frequency domain sampling?

• How FT is sampled and reconstructed using these samples?

• State & prove all the properties of DFT

• Explain the procedure for linear filtering using DFT

• What is the difficulty in frequency analysis using DFT and how it

can be overcome?
 Properties of DFT
• linearity property
• Periodicity property
• Symmetry property
Circular symmetry of a sequence:
Periodic sequence formed using x(n) is

x p (n)   x(n  lN )
l 

Shifting this sequence by k units, resulting periodic

sequence be 
xp (n)   x(n  k  lN )
l 
Dept. of E&CE, SDMCET, Dharwad 25
Dept. of E&CE, SDMCET, Dharwad 26
Dept. of E&CE, SDMCET, Dharwad 27
 Symmetry property
• With x(n) xR (n)  jx I (n) & X (k )  X R (k )  jX I (k )
N1
 2kn 2kn 
X R (k )   xR (n) cos( )  xI (n) sin( )
n 0  N N 
N1
 2kn 2kn 
& X I (k )    xR (n) sin( )  xI (n) cos( )
n 0  N N 
• Similarly
1 N1
 2kn 2kn 
xR ( n)  
N
 X R (k ) cos(
k 0  N
)  X I (k ) sin(
N 
)

1 N  1 2kn 2kn 
xI (n)    X R (k ) sin( )  X I (k ) cos( )
N k 0  N N 
 Symmetry property
• Different cases are possible for a sequence.
• Case (1): Real and even sequence: x(N-n)=x(n)=xR(n)
N1
2kn
 X (k )  X R (k )  xR (n) cos( )
n 0 N
• DFT of real and even sequence is real and even
sequence

1 N1
2kn
• IDFT : xR ( n) 
N

k 0
X R (k ) cos(
N
)
 Symmetry property
• Case (2): Real and odd sequence: x(n)=-x(N-n)
N1
2kn
 X I (k )   xR (n) sin( )
n 0 N
N1
2kn
 X (k )  X I (k )  j  xR (n) sin( )
n 0 N
• DFT of real and odd sequence is imaginary and odd
sequence

• IDFT : j N1
2kn
xR ( n) 
N

k 0
X I (k ) sin(
N
)
 Symmetry property
• Case (3): imaginary sequence: x(n)=jxI(n)
N1
2kn N1
2kn
X R (k )  xI (n) sin( ) & X I (k )  xI (n) cos( )
n 0 N n 0 N
• If xI(n) is odd sequence, X(k) is real
• If xI(n) is even sequence, X(k) is imaginary
• Summary:
x(n)  x ER (n)  x OR (n)  jx EI (n)  jx OI (n)

X (k )  X RE (k )  X OR (k )  jX IE (k )  jX OI (k )
 Symmetry property
• Consider 4 or 5 point sequences depicting these 4
cases. Determine the corresponding DFT and verify
symmetry property.
Multiplication of 2 DFTs & circular convolution property

• Consider two sequences

DFT DFT
x1 (n)  X 1 (k ) & x2 (n)  X 2 (k )
N N

• Product of DFTs is equivalent to circular

convolution of those two time domain sequences

DFT
x1 (n) N x2 ( n )  X 1 (k ) X 2 (k )
N
 Time reversal property
DFT DFT
If x(n)  X (k ) then x((  n)) N  X ((  k )) N  X ( N  k )
N N

Circular time shift property

DFT DFT 2kl
 j
If x(n)  X (k ) then x(( n  l )) N  X (k )e N
N N

Circular frequency shift property

DFT 2ln DFT
j
If x(n)  X (k ) then x(n)e N
 X (( k  l )) N
N N
 Complex Conjugate Properties
DFT DFT
If x(n)  X (k ) then x (n)  X  ((  k )) N  X  ( N  k )

N N

Multiplication of two sequences

Consider two sequences
DFT DFT
x1 (n)  X 1 (k ) & x2 (n)  X 2 (k )
N N

Product of two sequences is equivalent to circular convolution

of DFTs of those two time domain sequences
1
DFT
x1 (n) x2 (n)  X 1 (k ) N X 2 (k )
N N
Parseval’s Theorem

𝑁 −1 𝑁 −1
1
∑ |𝑥 [ 𝑛 ]| = 𝑁 ∑ | 𝑋 [ 𝑘 ]|
2 2

𝑛=0 𝑘=0
 Linear Filtering using DFT
• Linear convolution of two time domain sequences
is equivalent to multiplication of FT of these
sequences. FT
x1 (n)  x2 (n)  X 1 ( ) X 2 ( )

• For a system with h(n) as impulse response and

FT
x(n) as input, y (n)  x(n)  h(n)  X ( ) H ( ) Y ( )
• With length of x(n) as L, length of h(n) as M, length
of y(n) will be ….
• N=L+M-1
• Sampling Y ( ) to have same length as y(n), there
will be N samples in Y(k).
 FT
y (n)  x(n)  h(n)  X ( ) H ( ) Y ( )
• What about length of X(k) and H(k)?
– (Sampled versions of Fourier transform of x(n) & h(n))
• If they are of length L & M respectively, how to
multiply unequal length DFTs and get Y(k) of
length N=L+M-1?
• What about length of X(k) and H(k)?
• Output sequence y(n) is of length N > M and L
 Procedure for Linear Filtering using DFT
• (1) Extend x(n) & h(n) to N
– Append M zeros at the end of x(n)
– Append L zeros at the end of h(n)
• (2) Perform circular convolution of these two
extended sequences
– Result is linear convolution of x(n) & h(n)
• OR
• Step (1) same
• (2) Determine N point DFT of x(n) and h(n)
– X(k) & H(k) with N points each
• (3) Determine IDFT {X(k)H(k)}
 Difference between Linear convolution and
circular convolution
• For circular convolution, length of sequences
should be equal
• For linear convolution, length of sequences should
be N=L+M-1=length of x(n) + length of h(n) - 1
– Extend sequences to length N and proceed in the same
way as circular convolution
 Linear filtering of long sequence
• (1) Overlap save method
• (2) Overlap add method

• In both methods,
– It is assumed that input is too long and divided into
small blocks of block size L
– It is assumed that length of h(n) is M

• Intermediate output will be of length N=L+M-1

 Overlap save method
L L L
.....

M-1 zeros
x1(n)
Total length of each
x2(n) smaller sequence =N
M-1 x3(n)
M-1
y1(n)

Discard M-1 y2(n)

samples
Discard M-1 y3(n)
samples
Discard M-1
samples
 L Overlap
L add method
L
.....

M-1 zeros
x1(n) M-1 zeros
x2(n) M-1 zeros
x3(n)
Total length of each
y1(n) smaller sequence =N

y2(n)
M-1samples added y3(n)

M-1samples added
 THANK YOU

Next: Unit 1- Part 2 : Frequency Analysis of signals using DFT