OVERVIEW OF DIGITAL SIGNAL PROCESSING

1 • Presenter: Engr. Ahsan Khalid Ghauri

• Venue: Electrical Engineering Department

• University of Engineering and Technology, Lahore (Narowal Campus)

• Date: October 22nd, 2018

• Technical Talk Arranged by IEEE UET Narowal Chapter

 WELCOME TO THE SEMINAR
2
• Welcome to All Participants
 OUTLINE OF SEMINAR
3

• Introduction to Signals

• Introduction to Systems

• Digital Filter Design

• Multirate Signal Processing

4 WHAT IS A SIGNAL?

• Voice
• Radar pulse
• Electrocardiogram (ECG)
• Image
• Video
 SIGNALS THAT ARE A FUNCTION OF TIME
5

Signal Type Representation Amplitude

Continuous in time 𝑥(𝑡) Continuous
Discrete in time 𝑥(𝑛𝑇) Continuous
Digital 𝑥𝑄 (𝑛𝑡) Discrete
 RELATION BETWEEN VARIOUS SIGNAL TYPES
6

𝑥(𝑡) 𝑥(𝑛𝑇) 𝑥𝑄 (𝑛𝑡)

Digital Signal
Analog Signal Sampled Signal Quantized Signal Processor
7 SAMPLING

• Analog signal is continuous in both time and amplitude.

• As a result of sampling, the continuous in time signal becomes discrete in time.

• No effect on the variation of amplitudes.

8 QUANTIZATION

• The amplitude of the signal are assigned a discrete amplitude value.

• Finite set of output values are indexed.

 WHAT IS DIGITAL SIGNAL PROCESSING?
9

• Input signal output signal

Analog to Digital Digital Signal Digital to Analog

conversion Processor conversion
10 ADVANTAGE OF DIGITAL REPRESENTATION

• Signal is robust against noise.

• Ease of storage of data and transmission of data.

• Communication is almost entirely digital

• Lower cost
 DSP APPLICATION AREAS
11

• Speech

• Audio

• Image

• Video
 BASIC DISCRETE-TIME SIGNALS
12
Signal Name Mathematical Representation Graph

Unit Impulse 1, 𝑛 = 0, 1
(Unit sample) δ[𝑛] = ቊ
0, 𝑛≠0

Unit Step 1, 𝑛 ≥ 0,
u[𝑛] = ቊ
0, 𝑛 < 0

Real Exponential 𝑥 𝑛 = α𝑛

Sine Wave 𝑥 𝑛 = 𝐴𝑐𝑜𝑠(𝜔0 𝑛 + Φ)

 REPRESENTATION OF A DISCRETE TIME SIGNAL AS
13 AN IMPULSE

𝑥 𝑛 = ෍ 𝑥 𝑘 𝛿[𝑛 − 𝑘]
𝑘=−∞
 DISCRETE-TIME LTI SYSTEM
14

δ𝑛 LTI System h𝑛
𝑥𝑛 y𝑛
• Linearity

• Time Invariance
15 CONVOLUTION

𝑦 𝑛 =𝑥 𝑛 ∗ℎ 𝑛

∞ ∞

𝑦[𝑛] = ෍ 𝑥 𝑘 ℎ[𝑛 − 𝑘] = ෍ ℎ 𝑘 𝑥[𝑛 − 𝑘]

𝑘=−∞ 𝑘=−∞
 DECONVOLUTION
16
• If the impulse response and output of the system are known.

• y[n]=h[0]x[n] + σ𝑛𝑚=1 ℎ 𝑚 𝑥[𝑛 − 𝑚]

• When n=0, y[0]=x[0]h[0]

𝑦[0]
• 𝑥[0] =
ℎ[0]

𝑦[𝑛]−σ𝑛
𝑚=1 ℎ[𝑚]𝑥[𝑛−𝑚]
• 𝑥 𝑛 =
ℎ(0)
17 BLIND DECONVOLUTION

Impulse response of the system is unknown.

 CORRELATION
18

• Autocorrelation
𝑟11 = σ𝑁−1
𝑘=0 𝑥1 𝑘 𝑥2 [𝑘]

Cross correlation
𝑟12 = σ𝑁−1
𝑘=0 𝑥1 𝑘 𝑥2 𝑘
 Z-TRANSFORM
19

𝑋 𝑧 = ෍ 𝑥[𝑛]𝑧 −𝑛
𝑛=−∞
 DISCRETE TIME FOURIER TRANSFORM (DTFT)
20

𝑋(𝑒 𝑗𝜔 )=σ∞
𝑛=−∞ 𝑥[𝑛]𝑒
−𝑗𝜔𝑛

1 𝑗𝜔 )
𝑥𝑛 = ‫׬‬ 𝑋(𝑒 𝑒 𝑗𝜔𝑛 𝑑𝜔
2𝜋 2𝜋
 DISCRETE FOURIER TRANSFORM (DFT)
21
A periodic signal with N samples

The periodic signal can be represented as a discrete sum of sinusoids.

ሖ
𝑁−1
ሖ = ෍ 𝑥[𝑛]𝑒
𝑋[𝑘] ሖ −𝑗2𝜋𝑘𝑛/𝑁
𝑛=0

𝑁−1ሖ
1
ሖ = ෍ 𝑋[𝑘]𝑒
𝑥[𝑛] ሖ 𝑗2𝜋𝑘𝑛/𝑁
𝑁
𝑘=0
 DIGITAL FILTERS
22
A Discrete Time LTI system

• 𝑦 𝑛 = 𝑥 𝑛 ∗ ℎ[𝑛]

• Causal ℎ 𝑛 = 0, 𝑛<0

• Bounded σ∞
𝑛=−∞ ℎ 𝑛 < ∞
 IDEAL FILTER RESPONSES
23

|𝐻𝐿𝑃 𝜔 | |𝐻𝐻𝑃 𝜔 |
1

ω𝑐 π
|𝐻𝐵𝑃 𝜔 | |𝐻𝐵𝑆 𝜔 |
 DIGITAL FILTER IMPLEMENTATION
24

• 𝑦 𝑛 − σ∞ ∞
𝑘=1 𝑎𝑘 𝑦 𝑛 − 𝑘 = σ𝑚=0 𝑏𝑚 𝑥 𝑛 − 𝑚

• 𝐻 𝑧 =

• M zeros and N poles.

• All poles inside the unit circle.
 FIR FILTER
25
𝑦 𝑛 = σ𝑀
𝑟=0 𝑏𝑟 𝑥[𝑛 − 𝑟]= 𝑏0 + 𝑏1 𝑥 𝑛 − 1 + ⋯ + 𝑏𝑀 𝑥 𝑛 − 𝑀

𝑏𝑛 , 0≤𝑛≤𝑀
ℎ[𝑛] = ቊ
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

H(z)= σ𝑀
𝑛=0 𝑏𝑛 𝑧
−𝑛
 WINDOW DESIGN IN FIR FILTERS
26

• ℎ 𝑛 = ℎ𝐼 [𝑛] ∙ 𝑤[𝑛]

• H 𝑒 𝑗ω𝑛 = 𝐻𝐼 𝑒 𝑗ω𝑛 ∗ 𝑊(𝑒 𝑗ω𝑛 )

 COMMON FIR WINDOWS
27

Window Name Function

Rectangular 𝑏𝑛 , 0≤𝑛≤𝑀
w[𝑛] = ቊ
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Blackman 2π𝑛 4π𝑛
w 𝑛 = 0.42 − 0.5 cos( ) + 0.08𝐶𝑜𝑠( )
𝑀 𝑀
Hamming 2π𝑛
𝑤 𝑛 = 0.54 − 0.46 cos( )
𝑀
Hanning 2π𝑛
𝑤 𝑛 = 0.5 − 0.5 cos( )
𝑀
 IIR FILTER
28
• 𝑦 𝑛 = σ𝑁 𝑀
𝑘=1 𝑎𝑘 𝑦 𝑛 − 𝑘 + σ𝑚=0 𝑏𝑚 𝑥 𝑛 − 𝑚

• For M<N,

σ𝑀
𝑚=0 𝑏𝑚 𝑧
−𝑚
• H(z)=
1−σ𝑁 𝑎𝑘 𝑧 −𝑘

σ𝑀
𝑚=0 𝑏𝑚 𝑧
−𝑚 𝐴𝑘
• H(z)= σ𝑁 σ𝑁
𝑘=1
1− 𝑘=1 𝑁𝑎𝑘 𝑧 −𝑘 1−𝑑𝑘 𝑧 −1

• ℎ[𝑛]= σ𝑁 𝑛
𝑘=1 𝐴𝑘 (𝑑𝑘 ) 𝑢[𝑛]
 IIR FILTER DESIGN METHODS
29

Butterworth designs yield maximally flat amplitude.

Bessel designs yield maximally flat group delay.

Chebyshev designs yield equi-ripple in either passband or stopband.

Elliptic designs yield equi-ripple in both passband and stopband.

 SAMPLING WITH AN IMPULSE TRAIN
30

• 𝑥𝑝 𝑡 = 𝑥 𝑡 𝑝 𝑡
𝑝 𝑡

• 𝑝 𝑡 = σ∞
𝑛=−∞ 𝛿(𝑡 − 𝑛𝑇)
X
𝑥 𝑡 𝑥𝑝 (𝑡)
 SPECTRUM OF SAMPLED SIGNAL
31
𝑋(𝑗ω)

𝜔𝑀 ω

P(𝑗ω)

𝜔𝑆
 SAMPLING WITH AN IMPULSE TRAIN- EFFECT
32 ON THE FREQUENCY DOMAIN

𝜔𝑠 > 2𝜔𝑚
𝑋𝑝 (𝑗𝜔)

𝜔𝑀 𝜔𝑠 − 𝜔𝑀 𝜔𝑠

𝜔𝑠 < 2𝜔𝑚 𝑋𝑝 (𝑗𝜔)

𝜔𝑠 − 𝜔𝑀 𝜔𝑀 𝜔𝑆
 MULTIRATE SIGNAL PROCESSING- DECIMATION
33

Every Nth sample is kept.

The rest of the N-1 samples are removed.

Signal is compressed in the time-domain.

 MULTIRATE SIGNAL PROCESSING- INTERPOLATION
34

N-1 zeros are inserted in between the samples.

Signal is expanded in the time domain.

35 THANK YOU!
 QUESTIONS
36