Ece Filter Design

Lecture 22: Design of FIR / IIR Filters
Foundations of Digital Signal Processing
Outline
• Designing FIR Filters with Windows
• Designing FIR Filters with Frequency Selection
• Designing FIR Filters with Equi-ripples
• Designing IIR Filters with Discrete Differentiation
• Designing IIR Filters with Impulse Invariance
• Designing IIR Filters with the Bilinear Transform
• Related Analog Filters
Foundations of Digital Signal Processing Lecture 22: Designing FIR / IIR Filters 1
News
 Homework #9
 Due on Thursday
 Submit via canvas
 Coding Assignment #6
 Due on next Monday
 Submit via canvas
News
 Exam #2 – Great Job!
 Mean: 86.3
 Median: 87.5
Outline
Causality and Linear Phase
Question: Consider a length-M symmetric, causal filter.
What condition must be satisfied?
What condition must be satisfied?
 𝑥𝑥 𝑛𝑛 = ±𝑥𝑥 −𝑛𝑛 + 𝑁𝑁 − 1 = ±𝑥𝑥 𝑁𝑁 − 1 − 𝑛𝑛

 Positive: Even symmetry
 Negative: Odd symmetry
What is the phase response?
What is the phase response? Assume M is even.
 Even Symmetry
𝑋𝑋 𝑧𝑧 = 𝑎𝑎0 + 𝑎𝑎1𝑧𝑧 −1 + 𝑎𝑎2𝑧𝑧 −2 + ⋯ + 𝑎𝑎1𝑧𝑧 − 𝑀𝑀−2 + 𝑎𝑎0𝑧𝑧 − 𝑀𝑀−1
 Odd Symmetry
𝑋𝑋 𝑧𝑧 = 𝑎𝑎0 + 𝑎𝑎1𝑧𝑧−1 + 𝑎𝑎2𝑧𝑧−2 + ⋯ − 𝑎𝑎1𝑧𝑧 − 𝑀𝑀−2 − 𝑎𝑎0𝑧𝑧 − 𝑀𝑀−1
 Even Symmetry
𝑀𝑀−1 𝑀𝑀−1 𝑀𝑀−1 𝑀𝑀−1 𝑀𝑀−1
− −1 1− −
= 𝑧𝑧 2 𝑎𝑎0𝑧𝑧 2 + 𝑎𝑎1𝑧𝑧 2 + ⋯ + 𝑎𝑎1𝑧𝑧 2 + 𝑎𝑎0𝑧𝑧 2
 Odd Symmetry
−
2 −1
1− −
= 𝑧𝑧 2 𝑎𝑎0𝑧𝑧 2 + 𝑎𝑎1𝑧𝑧 + ⋯ − 𝑎𝑎1𝑧𝑧 2 − 𝑎𝑎0𝑧𝑧 2
 Even Symmetry
− −1 1− −
𝐺𝐺 𝜔𝜔 = 𝑋𝑋 𝜔𝜔 𝑒𝑒 jΘ 𝜔𝜔
 Odd Symmetry
−
2 −1
1− −
𝐺𝐺 𝜔𝜔 = 𝑋𝑋 𝜔𝜔 𝑒𝑒 jΘ 𝜔𝜔
 Even Symmetry
− −1 1− −
0 for 𝐺𝐺 𝜔𝜔 > 0
𝐺𝐺 𝜔𝜔 = 𝑋𝑋 𝜔𝜔 𝑒𝑒 jΘ 𝜔𝜔 Θ 𝜔𝜔 = �
𝜋𝜋 for 𝐺𝐺 𝜔𝜔 < 0
 Odd Symmetry
−
2 −1
1− −
0 for 𝐺𝐺 𝜔𝜔 > 0
𝐺𝐺 𝜔𝜔 = 𝑋𝑋 𝜔𝜔 𝑒𝑒 jΘ 𝜔𝜔 Θ 𝜔𝜔 = �
𝜋𝜋 for 𝐺𝐺 𝜔𝜔 < 0
 Even Symmetry
− −1 1− −
−𝜔𝜔 𝑀𝑀 − 1 /2 for 𝐺𝐺 𝜔𝜔 > 0

∠𝑋𝑋 𝜔𝜔 = �
−𝜔𝜔 𝑀𝑀 − 1 /2 + 𝜋𝜋 for 𝐺𝐺 𝜔𝜔 < 0
 Odd Symmetry
−
2 −1
1− −
−𝜔𝜔 𝑀𝑀 − 1 /2 for 𝐺𝐺 𝜔𝜔 > 0

∠𝑋𝑋 𝜔𝜔 = �
−𝜔𝜔 𝑀𝑀 − 1 /2 + 𝜋𝜋 for 𝐺𝐺 𝜔𝜔 < 0
 Even Symmetry
− −1 1− −
 Even Symmetry
− −1 1− −
𝑀𝑀/2−1
𝑀𝑀−1 𝑀𝑀−1 𝑀𝑀−1
−
= 𝑧𝑧 −
2 � 𝑎𝑎𝑘𝑘 𝑧𝑧 2 −𝑘𝑘 + 𝑧𝑧 2 − 𝑘𝑘
𝑘𝑘=0
 Even Symmetry
− −1 1− −
𝑀𝑀/2−1
−
= 𝑧𝑧 −
𝑘𝑘=0
 Notice that
𝑋𝑋 𝑧𝑧 = 𝑧𝑧 − 𝑀𝑀−1 𝑋𝑋 𝑧𝑧 −1
 Even Symmetry
− −1 1− −
𝑀𝑀/2−1
−
= 𝑧𝑧 −
𝑘𝑘=0
 Pole-zero plot property?

 Even Symmetry
− −1 1− −
𝑀𝑀/2−1
−
= 𝑧𝑧 −
𝑘𝑘=0
o o
o o
x x(M-1)
 Pole-zero plot property? o o
o o
Causality
 Question: How do we describe causal filter magnitude?
Causality
Often plotted
in dB (decibels)
Causality
Often plotted
in dB (decibels)
Lecture 21: Design of FIR Filters
Outline
• Review Downsampling & Upsampling
• Causality in Filters
Designing with Windows
 Question: How can I design an FIR filter from an ideal filter?
… …
… …
ℱ
Non-causal Ideal filter
Infinite Response
… …
ℱ
Infinite Response
 Answer: Window the response!

… …
ℱ
Infinite Response

… …
ℱ
Infinite Response

 Different Filters
 Windowing the sinc impulse response
Outline
• Designing FIR Filters with Frequency Sampling
Design with Frequency Sampling
 Option 2: Work backwards with constraints
Consider the DFT:

𝑁𝑁−1
2𝜋𝜋
𝑗𝑗 𝑛𝑛𝑛𝑛
ℎ 𝑛𝑛 = � 𝐻𝐻 𝑘𝑘 𝑒𝑒 𝑁𝑁 such that 𝐻𝐻 𝑘𝑘 = 𝐻𝐻 𝑁𝑁 − 𝑘𝑘
𝑘𝑘=0
Consider the DFT:

𝑁𝑁−1
1 2𝜋𝜋
𝑁𝑁
𝑘𝑘=0
𝑁𝑁−1 /2 𝑁𝑁−1
1 2𝜋𝜋
𝑗𝑗 𝑁𝑁 𝑛𝑛𝑛𝑛
2𝜋𝜋
ℎ 𝑛𝑛 = 𝐻𝐻 0 + � 𝐻𝐻 𝑘𝑘 𝑒𝑒 + � 𝐻𝐻 𝑘𝑘 𝑒𝑒
𝑁𝑁
𝑘𝑘=1 𝑘𝑘= 𝑁𝑁+1 /2
Consider the DFT:

𝑁𝑁−1
1 2𝜋𝜋
𝑁𝑁
𝑘𝑘=0
𝑁𝑁−1 /2 𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
𝑗𝑗 𝑁𝑁 𝑛𝑛 𝑁𝑁−𝑘𝑘
𝑁𝑁
𝑘𝑘=1 𝑘𝑘=1
Consider the DFT:

𝑁𝑁−1
1 2𝜋𝜋
𝑁𝑁
𝑘𝑘=0
𝑁𝑁−1 /2 𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
𝑁𝑁
𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
𝑗𝑗 𝑛𝑛 𝑁𝑁−𝑘𝑘
ℎ 𝑛𝑛 = 𝐻𝐻 0 + � 𝐻𝐻 𝑘𝑘 𝑒𝑒 𝑁𝑁 + 𝐻𝐻 𝑘𝑘 𝑒𝑒 𝑁𝑁
𝑁𝑁
𝑘𝑘=1
Consider the DFT:

𝑁𝑁−1
1 2𝜋𝜋
𝑁𝑁
𝑘𝑘=0
𝑁𝑁−1 /2 𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
𝑁𝑁
𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
−𝑗𝑗 𝑛𝑛𝑛𝑛 𝑗𝑗𝑗𝑗𝑗𝑗𝑗
ℎ 𝑛𝑛 = 𝐻𝐻 0 + � 𝐻𝐻 𝑘𝑘 𝑒𝑒 𝑁𝑁 + 𝐻𝐻 𝑘𝑘 𝑒𝑒 𝑁𝑁 𝑒𝑒
𝑁𝑁
𝑘𝑘=1
Consider the DFT:

𝑁𝑁−1
1 2𝜋𝜋
𝑁𝑁
𝑘𝑘=0
𝑁𝑁−1 /2 𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
𝑁𝑁
𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
−𝑗𝑗 𝑁𝑁 𝑛𝑛𝑛𝑛
ℎ 𝑛𝑛 = 𝐻𝐻 0 + � 𝐻𝐻 𝑘𝑘 𝑒𝑒 + 𝐻𝐻 𝑘𝑘 𝑒𝑒
𝑁𝑁
𝑘𝑘=1
Consider the DFT:

𝑁𝑁−1
1 2𝜋𝜋
𝑁𝑁
𝑘𝑘=0
𝑁𝑁−1 /2 𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
𝑁𝑁
𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
−𝑗𝑗 𝑛𝑛𝑛𝑛
ℎ 𝑛𝑛 = 𝐻𝐻 0 + � 𝐻𝐻 𝑘𝑘 𝑒𝑒 𝑁𝑁 + 𝑒𝑒 𝑁𝑁
𝑁𝑁
𝑘𝑘=1
Consider the DFT:

𝑁𝑁−1
1 2𝜋𝜋
𝑁𝑁
𝑘𝑘=0
𝑁𝑁−1 /2 𝑁𝑁−1 /2
1 2𝜋𝜋
2𝜋𝜋
𝑁𝑁
𝑁𝑁−1 /2
1 2𝜋𝜋
ℎ 𝑛𝑛 = 𝐻𝐻 0 + 2 � 𝐻𝐻 𝑘𝑘 cos 𝑛𝑛𝑛𝑛
𝑁𝑁 𝑁𝑁
𝑘𝑘=1
 An inverse DFT that forces time-symmetry
𝑁𝑁−1 /2
1 2𝜋𝜋
𝑁𝑁 𝑁𝑁
𝑘𝑘=1
𝑁𝑁−1 /2
1 2𝜋𝜋
𝑁𝑁 𝑁𝑁
𝑘𝑘=1
 Example: Consider the desired 9-sample frequency response

with the first half defined by [1 1 0 0]
 Compute the frequency sampled filter
𝑁𝑁−1 /2
1 2𝜋𝜋
𝑁𝑁 𝑁𝑁
𝑘𝑘=1

1
ℎ 𝑛𝑛 = 1 + 2 cos 2𝜋𝜋/9 𝑛𝑛
2
1
2
1
2
 In practice, this should be circularly shifted so that the
maximum is centered.
𝑁𝑁−1 /2
1 2𝜋𝜋 𝑁𝑁 − 1
ℎ 𝑛𝑛 = 𝐻𝐻 0 + 2 � 𝐻𝐻 𝑘𝑘 cos 𝑛𝑛 − 𝑘𝑘
𝑁𝑁 𝑁𝑁 2
𝑘𝑘=1

1
2
1
ℎ 𝑛𝑛 = 1 + 2 cos 2𝜋𝜋/9 𝑛𝑛 − 8/2
2
with the first half defined by [1 1 1 1 0 0 0 0]
1
ℎ 𝑛𝑛 = 1 + 2 cos 2𝜋𝜋/19 𝑛𝑛𝑐𝑐 + 2 cos 4𝜋𝜋/19 𝑛𝑛𝑐𝑐 + 2 cos 6𝜋𝜋/19 𝑛𝑛𝑐𝑐
17
16
𝑛𝑛𝑐𝑐 = 𝑛𝑛 −
2
with the first half defined by [1 1 1 1 0 0 0 0]
1
ℎ 𝑛𝑛 = 1 + 2 cos 2𝜋𝜋/19 𝑛𝑛𝑐𝑐 + 2 cos 4𝜋𝜋/19 𝑛𝑛𝑐𝑐 + 2 cos 6𝜋𝜋/19 𝑛𝑛𝑐𝑐
17
16
𝑛𝑛𝑐𝑐 = 𝑛𝑛 −
2
with the first 10 values defined by 1
with the first 100 values defined by 1
 Note that in practice, this needs to be circularly shifted to the

center
 Note: The definition can be slightly modified
 Our definition:
𝑁𝑁−1 /2
1 2𝜋𝜋 𝑁𝑁 − 1
ℎ 𝑛𝑛 = 𝐻𝐻 0 + 2 � 𝐻𝐻 𝑘𝑘 cos 𝑛𝑛 − 𝑘𝑘
𝑁𝑁 𝑁𝑁 2
𝑘𝑘=1
𝑁𝑁−1 /2
1 2𝜋𝜋 𝑁𝑁 1
= 𝐻𝐻 0 + 2 � 𝐻𝐻 𝑘𝑘 cos 𝑛𝑛 − + 𝑘𝑘
𝑁𝑁 𝑁𝑁 2 2
𝑘𝑘=1
𝑁𝑁−1 /2
1 2𝜋𝜋 1
= 𝐻𝐻 0 + 2 � 𝐻𝐻 𝑘𝑘 cos 𝑛𝑛 + 𝑘𝑘 − 𝜋𝜋𝜋𝜋
𝑁𝑁 𝑁𝑁 2
𝑘𝑘=1
𝑁𝑁−1 /2
1 𝑘𝑘 𝐻𝐻
2𝜋𝜋 1
= 𝐻𝐻 0 + 2 � −1 𝑘𝑘 cos 𝑛𝑛 + 𝑘𝑘
𝑁𝑁 𝑁𝑁 2
𝑘𝑘=1
 Final Definition
𝑁𝑁−1 /2
1 𝑘𝑘 𝐻𝐻
2𝜋𝜋 1
ℎ 𝑛𝑛 = 𝐻𝐻 0 + 2 � −1 𝑘𝑘 cos 𝑛𝑛 + 𝑘𝑘
𝑁𝑁 𝑁𝑁 2
𝑘𝑘=1
Side note: This is very closely related to the

discrete cosine transform
Outline
Design with Equi-ripples
Previously derived:
𝑀𝑀/2−1
− 2 − 2 − 𝑘𝑘
𝑋𝑋 𝑧𝑧 = 𝑧𝑧 � 𝑎𝑎𝑘𝑘 𝑧𝑧 2 −𝑘𝑘 + 𝑧𝑧
𝑘𝑘=0
𝑀𝑀/2−1
−𝑗𝑗𝑗𝑗 2 𝑗𝑗𝑗𝑗 2 −𝑘𝑘 −𝑗𝑗𝑗𝑗 2 − 𝑘𝑘
𝑋𝑋 𝜔𝜔 = 𝑒𝑒 � 𝑎𝑎𝑘𝑘 𝑒𝑒 + 𝑗𝑗𝑗𝑗
𝑘𝑘=0
𝑀𝑀/2−1
−𝑗𝑗𝑗𝑗
𝑀𝑀−1 𝑀𝑀 − 1
= 2𝑒𝑒 2 � 𝑎𝑎𝑘𝑘 cos 𝜔𝜔 − 𝑘𝑘
2
𝑘𝑘=0
 Equi-ripple design
𝑀𝑀/2−1
𝑀𝑀−1
−𝑗𝑗𝑗𝑗 2 𝑀𝑀 − 1
𝑋𝑋 𝜔𝜔 = 2𝑒𝑒 � 𝑎𝑎𝑘𝑘 cos 𝜔𝜔 − 𝑘𝑘
2
𝑘𝑘=0
 Goal: Find the optimal 𝑎𝑎𝑘𝑘 s that satisfies passband / stopband

ripple constraints.
 Equi-ripple design
𝑀𝑀/2−1
−𝑗𝑗𝑗𝑗
𝑀𝑀−1 𝑀𝑀 − 1
min 𝑊𝑊 𝜔𝜔 𝐻𝐻𝑑𝑑 𝜔𝜔 − 2𝑒𝑒 2 � 𝑎𝑎𝑘𝑘 cos 𝜔𝜔 − 𝑘𝑘
𝑎𝑎𝑘𝑘 2
𝑘𝑘=0
Desired frequency
response
Equals: 𝛿𝛿2 = stopband ripple

𝛿𝛿2 𝛿𝛿1 = passband ripple
for 𝜔𝜔 in pass band
𝛿𝛿1
1 for 𝜔𝜔 in stop band
Outline
IIR Filter Design from Derivatives
 Designing IIR Filters
 No easy ways to design digital IIR filters
 So let us start from analog filters
 Option 1: Preserve the difference equation!
 Question: What is a derivative in discrete-time?
 In continuous-time
𝑑𝑑𝑑𝑑 𝑡𝑡
→ 𝑠𝑠𝑠𝑠 𝑠𝑠
𝑑𝑑𝑑𝑑
 In discrete-time
𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑 𝑡𝑡 𝑥𝑥 𝑡𝑡 − 𝑥𝑥 𝑡𝑡 − Δ𝑇𝑇 T=1
= lim
𝑑𝑑𝑑𝑑 Δ𝑇𝑇→0 Δ𝑇𝑇
𝑑𝑑𝑑𝑑 𝑡𝑡 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑇𝑇
� = = 𝑥𝑥 𝑛𝑛 − 𝑥𝑥 𝑛𝑛 − 1
𝑑𝑑𝑑𝑑 𝑡𝑡=𝑛𝑛𝑛𝑛 𝑇𝑇
𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑 𝑡𝑡 𝑥𝑥 𝑡𝑡 − 𝑥𝑥 𝑡𝑡 − Δ𝑇𝑇
= lim
𝑑𝑑𝑑𝑑 𝑡𝑡 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑇𝑇 1
� = = 𝑥𝑥 𝑛𝑛 − 𝑥𝑥 𝑛𝑛 − 1
𝑑𝑑𝑑𝑑 𝑡𝑡=𝑛𝑛𝑛𝑛 𝑇𝑇 𝑇𝑇
𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑 𝑡𝑡 𝑥𝑥 𝑡𝑡 − 𝑥𝑥 𝑡𝑡 − Δ𝑇𝑇
= lim
𝑑𝑑𝑑𝑑 𝑡𝑡 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑇𝑇 1
� = = 𝑥𝑥 𝑛𝑛 − 𝑥𝑥 𝑛𝑛 − 1
𝑑𝑑𝑑𝑑 𝑡𝑡 1
� → 1 − 𝑧𝑧 −1 𝑋𝑋 𝑧𝑧
 Question: What is a second-derivative in discrete-time?
𝑑𝑑2𝑥𝑥 𝑡𝑡 2
2
→ 𝑠𝑠 𝑋𝑋 𝑠𝑠
𝑑𝑑𝑡𝑡
𝑑𝑑2𝑥𝑥 𝑡𝑡 𝑑𝑑𝑑𝑑 𝑡𝑡 𝑑𝑑𝑑𝑑 𝑡𝑡
2
=
𝑑𝑑𝑡𝑡 𝑑𝑑𝑑𝑑 𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑 𝑡𝑡 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑇𝑇
� =
2
𝑑𝑑𝑡𝑡
2
=
𝑑𝑑2𝑥𝑥 𝑡𝑡 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑇𝑇 /𝑇𝑇 − 𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑇𝑇 − 𝑥𝑥 𝑛𝑛𝑛𝑛 − 2𝑇𝑇 /𝑇𝑇
2
� =
𝑑𝑑𝑡𝑡 𝑇𝑇
𝑡𝑡=𝑛𝑛𝑛𝑛
2
𝑑𝑑𝑡𝑡
2
=
𝑑𝑑2𝑥𝑥 𝑡𝑡 𝑥𝑥 𝑛𝑛𝑇𝑇 − 2𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑇𝑇 + 𝑥𝑥 𝑛𝑛𝑛𝑛 − 2𝑇𝑇 𝑥𝑥 𝑛𝑛 − 2𝑥𝑥 𝑛𝑛 − 1 + 𝑥𝑥 𝑛𝑛 − 2
� = →
𝑑𝑑𝑡𝑡 2 𝑇𝑇2 𝑇𝑇2
2
𝑑𝑑𝑡𝑡
2
=
𝑑𝑑2𝑥𝑥 𝑡𝑡 𝑥𝑥 𝑛𝑛𝑇𝑇 − 2𝑥𝑥 𝑛𝑛𝑛𝑛 − 𝑇𝑇 + 𝑥𝑥 𝑛𝑛𝑛𝑛 − 2𝑇𝑇 𝑥𝑥 𝑛𝑛 − 2𝑥𝑥 𝑛𝑛 − 1 + 𝑥𝑥 𝑛𝑛 − 2
� = →
𝑑𝑑𝑡𝑡 2 𝑇𝑇2 𝑇𝑇2
𝑑𝑑𝑑𝑑 𝑡𝑡 1 1
� → 2 1 − 2𝑧𝑧 + 𝑧𝑧 𝑋𝑋 𝑧𝑧 = 2 1 − 𝑧𝑧−1 2𝑋𝑋 𝑧𝑧
−1 −2
 Translate continuous-time to discrete-time
𝑑𝑑𝑘𝑘 𝑥𝑥 𝑡𝑡 𝑘𝑘 𝑋𝑋 𝑠𝑠
→ 𝑠𝑠
𝑑𝑑𝑡𝑡𝑘𝑘
𝑑𝑑𝑘𝑘 𝑥𝑥 𝑡𝑡 1 −1 𝑘𝑘 𝑋𝑋 𝑧𝑧
� → 1 − 𝑧𝑧
𝑑𝑑𝑡𝑡𝑘𝑘 𝑇𝑇
𝑑𝑑𝑘𝑘 𝑥𝑥 𝑡𝑡 𝑘𝑘 𝑋𝑋 𝑠𝑠
→ 𝑠𝑠
𝑑𝑑𝑡𝑡𝑘𝑘
𝑑𝑑𝑘𝑘 𝑥𝑥 𝑡𝑡 1 −1 𝑘𝑘 𝑋𝑋 𝑧𝑧
� → 1 − 𝑧𝑧
𝑑𝑑𝑡𝑡𝑘𝑘 𝑇𝑇
1
𝑠𝑠 → 1 − 𝑧𝑧−1
𝑇𝑇
1
 Example: 𝑠𝑠 → 1 − 𝑧𝑧 −1
𝑇𝑇
 Use the derivative conversion to transform the following biquad
filter into the discrete-time domain.
1
𝐻𝐻 𝑠𝑠 =
𝑠𝑠 + 0.1 2 + 9
1
𝑇𝑇
 Use the derivative conversion to transform the following bi-quad
1
𝐻𝐻 𝑠𝑠 =
𝑠𝑠 + 0.1 2 + 9
1 𝑇𝑇2
𝐻𝐻 𝑧𝑧 = 2 = 2
1 1
1 − 𝑧𝑧−1 + 0.1 +9 𝑇𝑇2 1 − 𝑧𝑧−1 + 0.1 +9
𝑇𝑇 𝑇𝑇
𝑇𝑇2 𝑇𝑇2
= 2 = 2
1 − 𝑧𝑧−1 + 0.1𝑇𝑇 + 9𝑇𝑇2 1 + 0.1𝑇𝑇 − 𝑧𝑧 −1 + 9𝑇𝑇2
1
𝑇𝑇
1
𝐻𝐻 𝑠𝑠 =
𝑠𝑠 + 0.1 2 + 9
𝑇𝑇2 Finding poles
𝐻𝐻 𝑧𝑧 = 2
1 + 0.1𝑇𝑇 − 𝑧𝑧−1 + 9𝑇𝑇2
−1 2
1 + 0.1𝑇𝑇 − 𝑧𝑧 + 9𝑇𝑇2 = 0
−1 2
1 + 0.1𝑇𝑇 − 𝑧𝑧 = −9𝑇𝑇2
1 + 0.1𝑇𝑇 − 𝑧𝑧 −1 = ±3𝑇𝑇𝑇𝑇
1
𝑧𝑧−1 = 1 + 0.1𝑇𝑇 ∓ 3𝑇𝑇𝑇𝑇 𝑧𝑧 =
1 + 0.1 ∓ 3𝑗𝑗 𝑇𝑇
1
𝑇𝑇
1
𝑧𝑧 = Poles
1 + 0.1 + 3𝑗𝑗 𝑇𝑇
1
𝑇𝑇
1
𝑧𝑧 =
1 + 0.1 + 3𝑗𝑗 𝑇𝑇
1
𝑠𝑠 → 1 − 𝑧𝑧−1
𝑇𝑇
Stable IIR
analog filters
 Pros: map here
 Relatively simple
 Cons:
 Very limiting 1
 Stable continuous-time poles can

only be mapped to low frequencies
Outline
IIR Filter Design by Impulse Invariance
 Option 2: Preserve the impulse response!
 Question: How else can I represent my transfer function?
𝐾𝐾
1
𝐻𝐻 𝑠𝑠 = �
𝑠𝑠 − 𝑝𝑝𝑘𝑘
𝑘𝑘=1
𝐾𝐾 𝐾𝐾
1
𝐻𝐻 𝑠𝑠 = � = � 𝑐𝑐𝑘𝑘 𝑒𝑒𝑝𝑝𝑘𝑘𝑡𝑡
𝐾𝐾 𝐾𝐾
1
𝐾𝐾
ℎ 𝑡𝑡 = � 𝑐𝑐𝑘𝑘 𝑒𝑒 𝑝𝑝𝑘𝑘𝑡𝑡
𝑘𝑘=1
𝐾𝐾 𝐾𝐾
1
𝐾𝐾
𝑘𝑘=1
𝐾𝐾 𝐾𝐾
ℎ 𝑛𝑛𝑛𝑛 = ℎ 𝑛𝑛 = � 𝑐𝑐𝑘𝑘 𝑒𝑒 𝑝𝑝𝑘𝑘𝑛𝑛𝑛𝑛 = � 𝑐𝑐𝑘𝑘 𝑒𝑒 𝑝𝑝𝑘𝑘𝑇𝑇 𝑛𝑛
𝐾𝐾 𝐾𝐾
1
𝐾𝐾
𝑘𝑘=1
𝐾𝐾 𝐾𝐾
ℎ 𝑛𝑛𝑛𝑛 = ℎ 𝑛𝑛 = � 𝑐𝑐𝑘𝑘 𝑒𝑒 𝑝𝑝𝑘𝑘𝑛𝑛𝑛𝑛 = � 𝑐𝑐𝑘𝑘 𝑒𝑒 𝑝𝑝𝑘𝑘𝑇𝑇 𝑛𝑛
𝐾𝐾
𝑐𝑐𝑘𝑘
𝐻𝐻 𝑧𝑧 = �
1 − 𝑒𝑒 𝑝𝑝𝑘𝑘𝑇𝑇 𝑧𝑧 −1
𝑘𝑘=1
𝐾𝐾 𝑐𝑐𝑘𝑘
 Example: 𝐻𝐻 𝑧𝑧 = ∑𝑘𝑘=1
1−𝑒𝑒 𝑝𝑝𝑘𝑘𝑇𝑇𝑧𝑧−1
 Use impulse invariance to transform the following biquad filter into
the discrete-time domain.
1
𝐻𝐻 𝑠𝑠 =
𝑠𝑠 + 0.1 2 + 9
1
𝐻𝐻 𝑠𝑠 =
𝑠𝑠 + 0.1 2 + 9
Poles:
𝑠𝑠 + 0.1 2 +9=0
𝑠𝑠 = ±3𝑗𝑗 − 0.1
1 1/2 1/2
𝐻𝐻 𝑠𝑠 = 2
= +
𝑠𝑠 + 0.1 + 9 𝑠𝑠 + 3𝑗𝑗 + 0.1 𝑠𝑠 − 3𝑗𝑗 + 0.1
Poles:
2
𝑠𝑠 + 0.1 +9=0
𝑠𝑠 = ±3𝑗𝑗 − 0.1
1 1/2 1/2
𝐻𝐻 𝑠𝑠 = 2
= +
𝑠𝑠 + 0.1 + 9 𝑠𝑠 + 3𝑗𝑗 + 0.1 𝑠𝑠 − 3𝑗𝑗 + 0.1
1/2 1/2
𝐻𝐻 𝑧𝑧 = −3𝑗𝑗−0.1 𝑇𝑇 −1
+
1 − 𝑒𝑒 𝑧𝑧 1 − 𝑒𝑒 3𝑗𝑗−0.1 𝑇𝑇 𝑧𝑧 −1
1/2
𝐻𝐻 𝑧𝑧 = −3𝑗𝑗−0.1 𝑇𝑇 𝑧𝑧 −1
1 − 𝑒𝑒
1/2
+ 3𝑗𝑗−0.1 𝑇𝑇 𝑧𝑧 −1
1 − 𝑒𝑒
Outline
IIR Filter Design by Bilinear Transform
 Option 3: Preserve the definition of z!
 Question: How are z and s related?
 From continuous-time to discrete-time
𝑠𝑠 = 𝑗𝑗Ω Taylor Series
𝑒𝑒 𝑠𝑠𝑠𝑠 = 𝑒𝑒 𝑠𝑠𝑛𝑛𝑛𝑛 = 𝑧𝑧𝑛𝑛 Expansion / Approximation
𝑧𝑧 = 𝑒𝑒 𝑠𝑠𝑠𝑠
 Building an approximation (𝑒𝑒 𝑥𝑥 ≈ 1 + 𝑥𝑥)
𝑠𝑠𝑇𝑇
𝑒𝑒 2 1 + 𝑠𝑠𝑠𝑠/2
𝑧𝑧 = 𝑠𝑠𝑇𝑇 ≈
−2 1 − 𝑠𝑠𝑠𝑠/2
𝑒𝑒
 Question: How are z and s related?
 From continuous-time to discrete-time
𝑠𝑠 = 𝑗𝑗Ω
𝑒𝑒 𝑠𝑠𝑠𝑠 = 𝑒𝑒 𝑠𝑠𝑛𝑛𝑛𝑛 = 𝑧𝑧𝑛𝑛 Bilinear
1 Expansion / Approximation
𝑧𝑧 = 𝑒𝑒 𝑠𝑠𝑠𝑠 𝑠𝑠 = ln 𝑧𝑧
𝑇𝑇
 Building an approximation
2 𝑧𝑧 − 1
𝑠𝑠 ≈
𝑇𝑇 𝑧𝑧 + 1
 The Bilinear Transform
Continuous-time to discrete-time
2 1 − 𝑧𝑧 −1
𝑠𝑠 →
𝑇𝑇 1 + 𝑧𝑧 −1
Discrete-time to continuous-time
1 + 𝑠𝑠𝑠𝑠/2
𝑧𝑧 →
1 − 𝑠𝑠𝑠𝑠/2
2 1−𝑧𝑧−1 1+𝑠𝑠𝑠𝑠/2
 Example: 𝑠𝑠 → , 𝑧𝑧 →
𝑇𝑇 1+𝑧𝑧−1 1−𝑠𝑠𝑠𝑠/2
 Use the bilinear transform to transform the following biquad filter
into the discrete-time domain.
1
𝐻𝐻 𝑠𝑠 =
𝑠𝑠 + 0.1 2 + 9
2 1−𝑧𝑧−1 1+𝑠𝑠𝑠𝑠/2
 Use the bilinear transform to transform the following biquad filter
into the discrete-time domain.
1
𝐻𝐻 𝑠𝑠 =
𝑠𝑠 + 0.1 2 + 9
1
2 1 − 𝑧𝑧−1
+ 0.1 +9
𝑇𝑇 1 + 𝑧𝑧−1
1 + 𝑧𝑧−1 2
= 2
2
1 − 𝑧𝑧−1 + 0.1 + 9 1 + 𝑧𝑧−1 2
𝑇𝑇
2 1−𝑧𝑧−1 1+𝑠𝑠𝑠𝑠/2
1 + 𝑧𝑧−1 2
2
1 − 𝑧𝑧−1 + 0.1 + 9 1 + 𝑧𝑧−1 2
𝑇𝑇
Outline
Multi-pole Filters
 Butterworth: Maximally flat passband
 Chebyshev: Faster cutoff with passband ripple
 Elliptic: Fastest cutoff with passband and stopband ripple

1.5
Magnitude Response
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
10
0
Phase Response
-10
-20
-30
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Butterworth Filter
 Butterworth Filter of order N
1 𝑗𝑗 2𝑘𝑘+𝑁𝑁−1 𝜋𝜋
𝐻𝐻 𝑗𝑗𝑗𝑗 =
∑𝑁𝑁
𝑘𝑘=1 𝑠𝑠 − 𝑠𝑠𝑘𝑘 𝑠𝑠𝑘𝑘 = 𝑒𝑒 2𝑁𝑁
Butterworth Filter
 Butterworth Filter of order N
 N equally spaced poles on a circle on the left-hand-side of the s-
plane
Butterworth Filter
 Properties of the Butterworth Filter
 It is maximally flat at 𝜔𝜔 = 0
1
 It has a cutoff frequency 𝐻𝐻 𝜔𝜔 = at 𝜔𝜔 = 𝜔𝜔𝑐𝑐
2
 For large 𝑛𝑛, it becomes an ideal filter
Chebyshev Filter
 Chebyshev Filter of order N
1
1 + 𝜖𝜖 2 𝐶𝐶𝑛𝑛2 𝜔𝜔
𝐶𝐶𝑛𝑛2 𝜔𝜔 is an nth-order Chebyshev Polynomial 𝜖𝜖 2 controls ripple
Chebyshev Filter
 Chebyshev Filter of order N
Chebyshev Filter
 Properties of the Chebyshev Filter
 It has ripples in the passband and is smooth in the stopband.
 The ratio between the maximum and minimum ripples in the
passband is
(1 + 𝜖𝜖 2) − 1/2
 If ϵ is reduced (i.e., the ripple size is reduced), then the stopband

attenuation is reduced.
 It has a sharper cut-off than a Butterworth filter, but at the
expense of passband rippling
Elliptic Filter
 Elliptic Filter of order N
1
1 + 𝜖𝜖 2 𝑅𝑅𝑛𝑛2 𝜔𝜔
𝑅𝑅𝑛𝑛2 𝜔𝜔 is an nth-order elliptic function 𝜖𝜖 2 controls ripple
Elliptic Filter
 Elliptic Filter of order N
Elliptic Filter
 Properties of the Elliptic Filter
 It has ripples in the passband and the stopband
 The ratio between the maximum and minimum ripples is larger
than the Chebyshev filter, but it has an even quicker transition
from passband to stopband
 It has poles and zeros, but they are much more difficult to compute
compared with the Butterworth and Chebyshev filters

Ece Filter Design

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ece Filter Design

Uploaded by

Copyright:

Available Formats

Lecture 22: Design of FIR / IIR Filters

Foundations of Digital Signal Processing

• Designing FIR Filters with Frequency Selection

• Designing FIR Filters with Equi-ripples

• Designing IIR Filters with Discrete Differentiation

• Designing IIR Filters with Impulse Invariance

• Designing IIR Filters with the Bilinear Transform

• Related Analog Filters

• Designing FIR Filters with Frequency Selection

• Designing FIR Filters with Equi-ripples

• Designing IIR Filters with Discrete Differentiation

• Designing IIR Filters with Impulse Invariance

• Designing IIR Filters with the Bilinear Transform

• Related Analog Filters

 𝑥𝑥 𝑛𝑛 = ±𝑥𝑥 −𝑛𝑛 + 𝑁𝑁 − 1 = ±𝑥𝑥 𝑁𝑁 − 1 − 𝑛𝑛

 Negative: Odd symmetry

−𝜔𝜔 𝑀𝑀 − 1 /2 for 𝐺𝐺 𝜔𝜔 > 0

−𝜔𝜔 𝑀𝑀 − 1 /2 for 𝐺𝐺 𝜔𝜔 > 0

 Pole-zero plot property?

• Designing FIR Filters with Windows

• Designing FIR Filters with Frequency Selection

• Designing FIR Filters with Equi-ripples

 Answer: Window the response!

 Answer: Window the response!

 Answer: Window the response!

• Designing FIR Filters with Frequency Sampling

• Designing FIR Filters with Equi-ripples

• Designing IIR Filters with Discrete Differentiation

• Designing IIR Filters with Impulse Invariance

• Designing IIR Filters with the Bilinear Transform

• Related Analog Filters

Consider the DFT:

Consider the DFT:

Consider the DFT:

Consider the DFT:

Consider the DFT:

Consider the DFT:

Consider the DFT:

Consider the DFT:

 Example: Consider the desired 9-sample frequency response

 Compute the frequency sampled filter

 Example: Consider the desired 9-sample frequency response

 Compute the frequency sampled filter

 Compute the frequency sampled filter

 Compute the frequency sampled filter

 Example: Consider the desired 9-sample frequency response

 Compute the frequency sampled filter

 Compute the frequency sampled filter

 Compute the frequency sampled filter

 Compute the frequency sampled filter

 Compute the frequency sampled filter

 Compute the frequency sampled filter

 Note that in practice, this needs to be circularly shifted to the

Side note: This is very closely related to the

• Designing FIR Filters with Frequency Sampling

• Designing FIR Filters with Equi-ripples

• Designing IIR Filters with Discrete Differentiation

• Designing IIR Filters with Impulse Invariance

• Designing IIR Filters with the Bilinear Transform

• Related Analog Filters

 Goal: Find the optimal 𝑎𝑎𝑘𝑘 s that satisfies passband / stopband

Equals: 𝛿𝛿2 = stopband ripple

• Designing FIR Filters with Frequency Sampling