You are on page 1of 47

Chapter 9 Computation of The

Discrete Fourier Transform


Introduction

Efficient algorithms for computing DFT –


Fast Fourier Transform.
Compute only a few points out of all N points
Compute all N points
 What are the efficiency criteria?
 Number of multiplications
 Number of additions
 Chip area in VLSI implementation

2018/9/18 DSP 2
9.1 Efficient Computation of the DFT

2018/9/18 DSP 3
2018/9/18 DSP 4
Fast Fourier Transform

2018/9/18 DSP 5
2018/9/18 DSP 6
 Computational algorithms that exploit both the
symmetry and the periodicity of the sequence
WNkn were known long before the era of high-
speed digital computation.
 Heideman et al. (1984) have traced the origins of
the FFT back to Gauss as early as 1805.
 Runge and later Danielson and Lanczos (1942)
described algorithms for which computation was
roughly proportional to NlogN rather than N2.
 Cooley and Tukey (1965) published an algorithm
for the computation of DFT that is applicable when
N is a composite number.
2018/9/18 DSP 7
9.2 The Goertzel Algorithm

2018/9/18 DSP 8
2018/9/18 DSP 9
2018/9/18 DSP 10
9.3 Decimation-in-Time FFT Algorithm

2018/9/18 DSP 11
2018/9/18 DSP 12
2018/9/18 DSP 13
2018/9/18 DSP 14
2018/9/18 DSP 15
2018/9/18 DSP 16
Butterfly: Basic unit in FFT

 j (2 / N )N / 2  j
W N/2
N e e  1
and
WNr N / 2  WNN / 2WNr  WNr

2018/9/18 DSP 17
2018/9/18 DSP 18
2018/9/18 DSP 19
2018/9/18 DSP 20
Fig.9.10 Flow graph using the butterfly
computation

2018/9/18 DSP 21
In-place computations

2018/9/18 DSP 22
Normal-order sorting

2018/9/18 DSP 23
Bit-Reversed Sorting

2018/9/18 DSP 24
Fig.14 Alternative forms

2018/9/18 DSP 25
9.4 Decimation-in-Frequency Algorithms

2018/9/18 DSP 26
2018/9/18 DSP 27
2018/9/18 DSP 28
2018/9/18 DSP 29
2018/9/18 DSP 30
2018/9/18 DSP 31
Alternative Forms

2018/9/18 DSP 32
2018/9/18 DSP 33
2018/9/18 DSP 34
FFT for Composite N

2018/9/18 DSP 35
2018/9/18 DSP 36
2018/9/18 DSP 37
9.6 Implementation of the DFT using
Convolution

2018/9/18 DSP 38
2018/9/18 DSP 39
2018/9/18 DSP 40
Chirp Signals

The sequence W-n2/2 can be thought of as a


complex exponential sequence with linearly
increasing frequency.
In radar systems, such signals are called
chirp signals; hence the name chirp
transform.

2018/9/18 DSP 41
2018/9/18 DSP 42
2018/9/18 DSP 43
2018/9/18 DSP 44
2018/9/18 DSP 45
2018/9/18 DSP 46
Candidate Questions (50~70%)

Chapter 5: 5.8, 5.12, 5.22, 5.29, 5.31

Chapter6: 6.5, 6.7

Chapter 7: 7.5, 7.16, 7.18, 7.21, 7.26

Chapter 8: 8.6, 8.8, 8.26, 8.39, 8.40

Chapter 9: 9.5, 9.19, 9.26


2018/9/18 DSP 47

You might also like