Chapter 6 the discrete Fourier...

Preview:

Citation preview

Chapter 6 the discrete Fourier transform

6.1 Definition of the discrete Fourier transform6.2 Relationship between DFT & DTFT 6.3 Sampling the Fourier transform6.4 properties of the Fourier transform6.5 linear convolution using the DFT6.6 Fourier analysis of signals using the DFT

6.1 definition of Discrete Fourier Transform(离散傅里叶变换,DFT)

[ ] [ ] [ mod ] [(( )) ]Nr

x n x n rN x n N x n

[ ] [ ] [ ]Nx n x n R n

EXAMPLE

To each sequence x[n] with finite length N, we can always associate a period sequence

is a rectangular sequence.[ ]NR n

后两种只适于无混叠

无混叠时

1 1

0 0

1

0

[ ] [(( )) ]

[ ] ( [ ]) [ ] [ ]

[ ] [ ] [ ] [ ] , 0,1, .... 1

N

N Nkn knN N

n n

Nkn

N Nn

x n x n

X k DFS x n x nW x nW

X k X k R k x nW k N

Let’s consider a sequence with finite-length N and a period sequence , We define a finite-length sequence as discrete Fourier transform (DFT) of x[n] which is a period of DFS .

The procedure can be illustrate by the steps followed:

[ ]x n[ ]x n

[ ]X k[ ]X k

We can reconstruct by get a period of . [ ]x n [ ]x n1 1

0 01

0

1 1[ ] ( [ ]) [ ] [ ] [(( )) ]

1[ ] [ ] [ ] [ ] , 0,1, ... 1

N Nkn kn

N N Nk k

Nkn

N Nk

x n IDFS X k X k W X k W x nN N

x n x n R n X k W n NN

[ ] [ ] [ ]Nx n x n R n

1

0

1

0

[ ] [ ] , 0,1, .... 1

1[ ] [ ] , 0,1, ... 1

NknN

n

Nkn

Nk

X k x nW k N

x n X k W n NN

1

0

1

0

[0] [ ]

1[0] [ ]

N

n

N

k

X x n

x X kN

So, for a sequence with duration N, it’s DFT and reverse transform are defined as:

DFT和IDFT具有固有的周期性

信号的长度,求和项数,频域取样点数,三者一致

{ [ ]}DFS x n

~

5[ ] [ ]X k R k

5[(( )) ]x n

以5为周期延拓

DFT与DFS以及DTFT的关系

40

5 50

1 1[ ] [ ] [0] 1, 0, ..45 5

kn n

kx n X k W X W n

周期1或5

周期信号的频谱只有直流成分

矩形序列的DTFT是连续且周期的,所以DFT不一定能精确反映频域特性。 反变换确实能恢复时域

EXAMPLE

2

1 12 /

0 01

0

2

[ ] [ ] [ ] , 0,1, .... 1

( ) [ ]

[ ] ( ) | ( ) | , 0,1, .... 1j kN

N Nkn j kn NN

n nN

j j n

n

j

k z eN

X k x n W x n e k N

X e x n e

X k X e X z k N

DFT is the sample of Fourier transform at equally spaced frequency 2 / , 0,1, ... 1k k N k N

所以DFT可以与FT独立

1

0

1[ ] [ ]

1[ ] ( )2

Nkn

Nk

j j n

x n X k WN

x n X e e d

二者均能恢复时域信号

6.2 Relationship between DFT & DTFT For sequence x[n] with finite length N

6.1-6.2 小结

1. 利用DFS的正反变换引出DFT的正反变换的定义;

2.DFT是DTFT的取样,取样点数越大,越能逼近DTFT;

3.DFT有自己的正反变换,是独立于DFS和DTFT的一种变换。

1

01

0

[ ] [ ] , 0,1, .... 1

1[ ] [ ] , 0,1, ... 1

NknN

nN

knN

k

X k x nW k N

x n X k W n NN

22[ ] ( ) | ( ) | , 0,1, .... 1j kN

j

k z eN

X k X e X z k N

1

2 / 1 /2

0

2sin1 2

2sin2

Nj k N Nj

k

N k

X e X k eN kN

N

思考题

[ ],[ ] [ 1 ], 0,1, ... 1[ ] [ / 2]

N x nx n x N n n NX k N DFT X N

长度为 (偶数)的因果序列

是它的 点 ,则 。0

6.3 sampling the Fourier transform

6.3.1 sampling points greater than the length of sequence 6.3.2 sampling points less than the length of sequence6.3.3 Sampling theory in frequency domain

提出两个问题:

1.当DFT对FT 取样太稀疏时,无法真实反映FT。能否采用DFT计算FT的任意点数的取样?(点数大于等于或小于信号长度)

2. 当对FT取样稀疏到什么程度时,无法重构时域信号?

6.3.1 sampling points greater than the length of sequence

EXAMPLE

以10为周期延拓

更真实地反映信号频谱

by zero-padding(补零) the N=5 point finite-length sequence to M=10,we can get the M point DFT.

1

01 1

2 /

0 0

2

[ ], 0, ..., 1[ ], 0, ..., 1

'[ ]0, , ..., 1

'[ ] { '[ ]} '[ ]

[ ] [ ]

( ) | , 0,1, .. 1

MknM

nN N

kn j kn MM

n nj

kM

x n n Nx n n N

x nn N M

X k M DFT x n x n W

x n W x n e

X e k M

不是DFT的定义式

M点DFT的定义式

补零后的信号的DFT是原信号DTFT的更密取样

PROVE (M>N)

1[ ]x n

2[ ]x n

两个序列的|X[k]| 相同, |X(ejω)|不一定相同

8点DFT

1| [ ] |X k 2 1| [ ] | | [ ] |X k X k

1| [ ] |X k 2 1| [ ] | | [ ] |X k X k

EXAMPLE

1024点DFT(DTFT的逼近)

Two signal with length 8

长度N的序列,时域补零到M后作M点的DFT,得到对其FT的M点取样;

该频域取样能通过IDFT重构时域信号,该时域信号只有前N个值非零。

6.3.1 小结

2

[ ], 0, ..., 1[ ], 0, ..., 1 '[ ]

0, , ..., 1

'[ ] { '[ ]} ( ) | , 0,1, .. 1j

kM

x n n Nx n n N x n

n N M

X k M DFT x n X e k M

2

[ ], 0, ..., 1[ ], 0, ..., 1 '[ ]

0, , ..., 1

{ ( ) | } '[ ], [ ] '[ ] [ ]jNk

M

x n n Nx n n N x n

n N M

M IDFT X e x n x n x n R n

2( ) | [ ] , 0,1... 1j knMk nM

X e x nW k M

'[ ] ( [ ]) [ ]Mr

x n x n rM R n

Discuss the general situation of sampling the Fourier transform:

If the length of sequence x[n] is N (can be infinite),and the sampling number in frequency domain is M (can be greater than, equal to or less than N),then the reconstructed signals are periodic with period M (may be overlapping). That is , if M points samples of the DTFT are:

Then the IDFT according to these samples is:

时域取样频域周期性延拓频域取样时域周期性延拓

6.3.2 sampling points less than the length of sequence

不是DFT

推导见课堂笔记

当频域取样点数M<序列长度N时,重构的时域信号是原始信号的有混迭的周期性延拓并取主周期。

6.3.2 小结

2

2

[ ], 0,..., 1

( )| [ ] , 0,1... 1( )

{ ( )| } ( [ ]) [ ]

j knMk nM

jMk rM

x n n N

X e x nW k M M N

M IDFT X e x n rM R n

2

'[ ] ( [ ]) [ ]( )

( )| = { '[ ]}, 0,1... 1

Mr

j

kM

x n x n rM R n M N

X e M DFT x n k M

若想用DFT求解FT的M(M<N)点取样,可将时域以M为周期延拓,取主周期,再作M点DFT,但该频域取样无法重构时域信号.

1 1[ ] ( )FT

jx n X e

2 2[ ] [ ]M IDFT

X k x n点

3 3[ ] [ ]M IDFS

X k x n 点

M点取样

M点取样

时域周期性延拓频域取样

取长为M的主周期

1[ ] [ ]Mr

x n rM R n

1[ ]r

x n rM

一长度为M的x2[n]的M点DFT、一周期为M的序列x3[n]的DFS,

均为另一序列x1[n]的DTFT的M点取样,则三个序列的时域关系:

If sampling points in frequency domain M is greater than or equal to the sequence length N, then we can reconstruct the signal in time domain; besides, we can express the sampling spectrum as described below (M=N):

1

0

1 [ ]( )1

j N Nj

k jk N

e X kX eN W e

1 1 1

0 0 01 1

0 0

1

0

:1( ) [ ] ( [ ] )

1 [ ]( )

1 [ ]1

N N Nj j n kn j n

Nn n k

N Nkn j n

Nk n

j N N

k jk N

prove

X e x n e X k W eN

X k W eN

e X kN W e

6.3.3 Sampling theory in frequency domain(频域取样定理)

1.DFT是DTFT的取样,频域取样点数越多,越能反映真实

的DTFT(通常DFT点数M远大于信号长度N)。

N点x[n]的DTFT的M点取样,通过M点IDFT重构的时域信号:

(1)M=N, x[n](2)M>N,x[n]的补零

(3)M<N,x[n]的混叠

所以用M点DFT求N点序列x[n]的DTFT的M点取样的方法:

(1)M=N,直接作x[n]的M点DFT(2)M>N,x[n]补零到M点,再作M点DFT(3)M<N,x[n]的混叠

2. 频域采样定理:频域取样点数大于等于信号长度时可以重构时

域,反之则不能。

3. 两个信号的DTFT相同,则DTFT的取样也相同;反之不成立。

DTFT广义线性相位,则DTFT的取样也广义线性相位;反之不成立。

6.3小结

思考题

(1)以下关于频域取样错误的是 ( )

(A)长度为N的有限长信号的N点DFT是其傅立叶变换的N点取样。(B)如果频域取样点数小于序列长度则无法通过频域取样重构

时域信号。(C)当频域取样点数小于序列长度时,也可以通过DFT计算频域

取样。(D)序列的频域取样如果是实序列,则傅立叶变换一定是实函数。

D

1 1 2 2[ ] [ ], [ ] [ ], [ ] [ ]DFT DFT DFT

x n X k x n X k x n X k

6.4 properties of the Discrete Fourier transform

1. linearity

2. circular shift (循环或圆周移位)

1 2 1 2[ ] [ ] [ ] [ ]DFT

ax n bx n aX k bX k

[(( )) ] [ ] [ ]DFT

kmN N Nx n m R n W X k

[ ] [(( )) ] [ ]DFT

nlN N NW x n X k l R k

Assume:

图示循环移位

1[ ] [(( )) ] [ ] [(( ( ))) ] [ ]N N N Nx n x n m R n x n N m R n Definition of circular shift of a sequence

EXAMPLE

][1 nh

][2 nh

8points DFT1| [ ] |H k 2| [ ] |H k

1024 points DFT1| ( ) |jH e 2| ( ) |jH e

h1[n]和h2[n]当成8点信号是循环移位关系,当成1024点则不是。

EXAMPLE

[ ] [0]

[ ] [(( )) ] [ ] [(( )) ] [ ][ ], 1, .., 1

[ ], 0, .., 1[0], 0

DFT

N N N N

x N x

X n Nx k R k Nx N k R kNx N k k N

Nx N k k NNx k

定义

3. Duality(对偶性)

EXAMPLE

[ ] [0]

[(( )) ] [ ] [(( )) ] [ ] [ ]x N x

N N N Nx k R k x N k R k x N k

定义

[ ]x k利用固有的周期性,定义:x[N]=x[0]

原点处不动,其余点相对于N/2左右翻折

[(( )) ]Nx k

[(( )) ]Nx k

[ ] [0]

[(( )) ] [ ] [ ]x N x

N Nx k R k x N k定义

以N为模反转的近似写法

[ ] [0]* * * *[ ] [(( )) ] [ ] [(( )) ] [ ] [ ]

X N XDFT

N N N Nx n X k R k X N k R k X N k

定义

0[ ] [ ]* * *[(( )) ] [ ] [ ] [ ]

x N x DFT

N Nx n R n x N n X k

定义

* *1 1[ ] ( [ ] [ ]) ( [ ] [ ]) Re{ [ ]}2 2

DFT

epx n x n x N n X k X k X k

* *1 1Re{ [ ]} ( [ ] [ ]) ( [ ] [ ]) [ ]2 2

DFT

epx n x n x n X k X N k X k

* *1 1Im{ [ ]} ( [ ] [ ]) ( [ ] [ ]) [ ]2 2

DFT

opj x n x n x n X k X N k X k

* *1 1[ ] ( [ ] [ ]) ( [ ] [ ]) Im{ [ ]}2 2

DFT

opx n x n x N n X k X k j X k

4. Symmetry properties

Here, we define:: the periodic conjugate-symmetric components

(圆周(周期)共轭对称分量)

: the periodic conjugate-antisymmetric components(圆周(周期)共轭反对称分量)

[ ]epX k

[ ]opX k

[ ] [0]* *

[ ] [0]* *

[ ] [ ] [ ]

,1[ ] ( [ ] [ ]) [ ]21[ ] ( [ ] [ ]) [ ]2

ep op

X N X

ep ep

X N X

op op

X k X k X k

where

X k X k X N k X N k

X k X k X N k X N k

定义

定义

Any finite-length sequence can be decomposed as:

The length of the three sequences are all N.

原点不算,其余点相对于N/2共轭对称

*

[ ] [0]*

[ ] [(( )) ] [ ]

[ ], 0, ..., 1

N NX N X

X k X k R k

X N k k N

定义

| [ ] | | [ ] |[ ] [ ]

X k X N kX k X N k

Re{ [ ]} Re{ [ ]}Im{ [ ]} Im{ [ ]}

X k X N kX k X N k

*[ ] [ ]x n x n

5. for a real sequence

圆周(周期)共轭对称:周期延拓后共轭对称

N=10[ ] 0.5 cos(0.5 ), 0...9nx n n n Real{X[k]} Imag{X[k]}

|X[k]| arg{X[k]}

实序列的DFT

EXAMPLE

N=9(奇数点)

|X[k]|

Arg{X[k]}

实序列的DFT

(1) The first five points of the 8-point DFT of a real sequence are {0.25, 0.125-j0.3, 0 , 0.125-j0.06, 0.5}, and we can know that the last three points are

思考题

(2)满足 的实序列,其N点DFT一定是 ( )

(A)实序列且周期性偶对称(B)复数序列且周期性共轭对称(C)实序列且周期性奇对称(D)复数序列且周期性共轭反对称

[ ] [ ] 1, ... -1x n x N n n N ,

A

{0.125+j0.06,0, 0.125+j0.3}

(A)

(B)

(C) 是实序列

(D)

(3)序列 ,傅立叶变换和5点DFT分别为和 ,则错误的是 ( )

5[ ] [ ]x n R n ( )jX e

[ ]X k2.5( ) ( ), ( )j j j jX e e A e A e 其中 是实函数

4 / 5[ ] [ ] [ ]j kX k e A k A k ,其中 是实序列

[ ]X k

[ ] [(( )) ] [ ]N NX k X N k R k

A

不存在FT和DFT均为实的序列。

FT和DFT的对称性质比较

时域 FT DFT(FT的取样)相对原点共轭对称

因果相对a共轭对称

周期共轭对称

因果矩形序列

纯虚

实函数(广义线性相位)广义线性相位函数 广义线性相位序列

非线性相位复数函数 实序列(广义线性相位)

广义线性相位函数 实序列(广义线性相位)

共轭对称 周期性共轭对称

共轭反对称 周期性共轭反对称

( )jA e

( )j j aA e e 2

[ ]j akNA k e

[ ]A k

( )j j aA e e [ ]A k

1 1* *

0 0

1[ ] [ ] [ ] [ ]N N

n kx n y n X k Y k

N

1 12 2

0 0

1| [ ] | | [ ] |N N

n kx n X k

N

6.paswal’s theory

prove

1 1 1* *

0 0 01 1 1

* *

0 0 0

1[ ] [ ] [ ]( [ ] )

1 1[ ] [ ] [ ] [ ]

N N Nkn

Nn n k

N N NknN

k n k

x n y n x n Y k WN

Y k x nW Y k X kN N

1

1 20

( [(( )) ] [(( )) ]) [ ]N

N N Nm

x m x n m R n

Definition of circular convolution

Length of x1[n],x2[n] and y[n] are all N.

1 2[ ] [ ] [ ]y n x n x n

7. circular convolution(循环卷积)

1

1 20

[ ] [(( )) ] [ ]N

N Nm

x m x n m R n

1 2( [(( )) ] [(( )) ]) [ ]N N Nx n x n R n

1

0( ) (( ))

0 0 1 2 1 01 1 0 1 2 12 2 1 0 3 2

1 1 2 3 0 1

N

Nm

y n g n N h n g m h n m

y h h N h N h gy h h h N h gy h h h h g

y N h N h N h N h g N

circulant matrix

(循环矩阵, )

矩阵表示

3 2[ ] [ ]( ) [ 1]x n x n N n

[(( 1)) ] [ ]N Nx n R n

求x3[0]

求x3[1]

x2[N-m]=

1

3 1 20

[ ] [ ] [(( )) ] [ ]N

N Nm

x n x m x n m R n

图解法

EXAMPLE

(1)N (convolution point)=L (signal length)

循环卷积结果与N有关

EXAMPLE

(2)N(convolution point) =2L (signal length)

循环卷积要指定点数

(一般大于等于两个

序列长度的最大值)

The properties of circular convolution are:

证明见课堂笔记

1 2 1 2(1) [ ]( ) [ ] [ ] [ ]DFT

x n N x n X k X k

1 2 1 21(2) [ ] [ ] [ ]( ) [ ]

DFTx n x n X k N X k

N

DFT法求循环卷积

6.5 linear convolution using the DFT

If the length of the two sequence are Nx and Nh separately,and the point of circular convolution N≥ Nx+Nh-1, then

if: [ ] [ ]* [ ]

then : [ ] [ ] [ ] [ ]Nr

y n x n h n

x n N h n y n rN R n

( )

[ ]* [ ] [ ] [ ]x n h n x n N h n ( )

Relationship of linear convolution and circular convolution

[ ] [ ] [ ]x n h n y n

( ) ( )j jX e H e

[ ] [ ]X k H kN points D FT

N point sample

N point IDFT

[ ] [ ]Nr

y n rN R n

2 /( ) |jk NY e

( )jY e

根据循环卷积性质 根据频

域取样

FT

频域取样关系

[ ]( ) [ ]x n N h n

If N>=Nx+Ny-1,then x[n]*h[n]=x[n] (N) h[n]

利用频域取样的概念证明线性卷积与循环卷积的关系:

循环卷积与线性卷积在频域是取样关系,所以时域是周期性延拓取主周期的关系。

PROVE

N>=max{Nx,Nh}

线性卷积右移

线性卷积左移

6点循环卷积=线性卷积的混迭

12点循环卷积=线性卷积

EXAMPLE

线性卷积

2

(2) calculate linear convolution by circular convolution

1 2( )zero-padding [ ] and [ ] to length of 1a x n h n N L L

1 2( )zero-padding [ ] and [ ] to length of 1a x n h n N L L

(1)calculate N point circular convolution by linear convolution

(3) calculate linear convolution by DFT

( )N points DFT of [ ] and [ ]b x n h n

( ) [ ] [ ]* [ ]a y n x n h n

( ) [ ]( ) [ ] [ ] [ ]Nr

b x n N h n y n rN R n

( ) [ ] [ ] [ ]( ) [ ]b x n h n x n N h n

( ) [ ]( ) [ ] { [ ] [ ]}[ ]* [ ] [ ]( ) [ ]

c x n N h n IDFT X k H kx n h n x n N h n

Conclusion

思考题

已知FIR的单位脉冲响应

输入信号

采用DFT求 在区间的值,则DFT的最小点数是 ( )

0 1h n n M 在 之外为零

0 1 ( )x n n L L M在 之外为零

[ ] [ ] [ ]y n x n h n 1 1M n L

(A)M (B)M-1(C)L (D)L+M-1

C

6.4-6.5 小结

1 2 1 2[ ] [ ] [ ] [ ]DFT

ax n bx n aX k bX k

[(( )) ] [ ] [ ]DFT

kmN N Nx n m R n W X k [ ] [(( )) ] [ ]

DFTnl

N N NW x n X k l R k

[ ] [ ], 1, .., 1DFT

X n Nx N k k N

[ ] Re{ [ ]}DFT

epx n X k

Re{ [ ]} [ ]DFT

epx n X k Im{ [ ]} [ ]DFT

opj x n X k

[ ] Im{ [ ]}DFT

opx n j X k

*[ ] is real, [ ]= [ ], 1, ..., 1x n X k X N k k N

1 1* *

0 0

1[ ] [ ] [ ] [ ]N N

n kx n y n X k Y k

N

1 2 1 2[ ]( ) [ ] [ ] [ ]DFT

x n N x n X k X k 1 2 1 21[ ] [ ] [ ]( ) [ ]

DFTx n x n X k N X k

N

if: [ ] [ ]* [ ]

then : [ ] [ ] [ ] [ ]Nr

y n x n h n

x n N h n y n rN R n

( )

6.6 Fourier analysis of signals using the DFT

6.6.1 Definition of the time-dependent Fourier transform 6.6.2 The system of the time-dependent Fourier analysis

using the DFT6.6.3 DFT analysis of sinusoidal signals

For finite-length signals, the DFT provides frequency-domain samples of the Discrete-time Fourier transform. In many cases, the signals do not inherently have finite length. The inconsistency between the finite-length requirement of the DFT and the reality of indefinitely long signals can be accommodated exactly or approximately through the concepts of windowing, block processing, and the time-dependent Fourier transform(短时傅立叶变换).

6.6.1 Definition of the time-dependent Fourier transform (短时/依时傅里叶变换)

commonly used windows

(长度M+1)

[ , ) [ ] [ ]10

Mj m

Mm

X n x n m w m e

[ ] [ ] [ , ) j mx n m w n X n e d

2

0

12

布莱克曼窗族

(三角形窗除外)

1 0(1)rectangular : [ ]

0n M

w nother

2 02

2(2)bartlett(triangular) : [ ] 22

0

n MnM

n Mw n n MM

other

20.5 0.5cos 0(3)hanning : [ ]

0

nn M

w n Mother

20.54 0.46cos 0(4)hamming : [ ]

0

nn M

w n Mother

2 40.42 0.5cos 0.08cos 0(5)blackman : [ ]

0

n n n Mw n M M

other

Blackman windows family

一阶升余弦窗

不用背

x n n 20[ ] cos( )

EXAMPLE

谱图

采用长度为400的汉明窗

6.6.2 The system of the time-dependent Fourier analysis using the DFT

用V[k]来估计sc(t)的傅里叶变换Sc(jΩ)

One of the major applications of the DFT is in analyzing the frequency content of continuous-time signals.

图6.5-4

真实频谱

由滤波器非理想引入误差

由量化和混迭引入误差

时域加窗和频域取样引入误差

抗混迭滤波器频响

窗序列频谱

EXAMPLE系统中各信号的频谱

Sc(jΩ)

Haa(jΩ)

Xc(jΩ)

X(ejω)

W(ejω)

V(ejω),V[k]

用V[k]推断Sc(jΩ)的步骤:(1)线性内插,(2)kω=2πk/N,(3)ωΩ= ω/T =2πk/(NT)

假设理想

第k(k=0,…,N/2-1)条谱线对应的频率:

V[k]的相临谱线所对应的频率间距与DFT点数N的关系:

2 // 2 // 2 /

s s

s

NT f f N

f f N

2 //

2 //

k

k k k s

k s

k s

k NT ff k N

f f k N

k,ω和Ω之间的关系

kωωΩkΩkf

实信号sc(t)的最高频率为5kHz,采样率为10000Hz,(1)要求DFT的相临样本间隔为10Hz,则DFT点数N为多少?(窗长<=N)(2)若作1000点DFT,已知V[11]=2000(1+2j),可以推断V[k]和Sc(jΩ)或Xc(jΩ)的其他值是多少?

*

11

*

(2) [1000 -11] [11] 200(1- 2 )

11, 2 / 2 110( 2 110) [11] 0.2(1 2 )

( 2 110) ( 2 110) 0.2(1- 2 )

s

c

c c

V V jk f k NS j V T j

S j S j j

根据实信号的DFT的共轭对称性:

解:

EXAMPLE

(1)10 / 10000 /1000

sHz f N NN

6.6.3 DFT analysis of sinusoidal signals

1. the effect of windowing2. the effect of spectral sampling

We choose sinusoidal signals as the specific class of examples to discuss, but most of the issues raised apply more generally.

0 0 0 1 1 1[ ] cos( ) cos( )x n A n A n

1. the effect of windowing

Before windowing

[ ] [ ] [ ], ( ) ( )j jx n w n v n X e V e

0 0

1 1

0 00 0

1 11 1

( ) 2 [ ( ) ( )2 2

( ) ( )]2 2

j jj

j j

A AX e e e

A Ae e

0 0 0 0 1 1 1 10 0 1 1

2 2 2 2j j n j j n j j n j j nA A A Ae e e e e e e e

n

[ ] [ ] [ ]v n x n w n

0 0 0 0

1 1 1 1

( ) ( )0 0

( ) ( )1 1

( ) ( ) ( )2 2

( ) ( )2 2

j j n j j nj

j j n j j n

A AV e e W e e W e

A Ae W e e W e

0 0 0 0

1 1 1 1

0 0

1 1

[ ] [ ]2 2

[ ] [ ]2 2

j j n j j n

j j n j j n

A Aw n e e w n e e

A Aw n e e w n e e

After windowing

(1)谱线展宽成窗

频谱的主瓣宽

(2)产生旁瓣,衰

减等于窗频谱

的旁瓣衰减

| ( ) |jX e

| ( ) |jV e

EXAMPLE

(3)谱泄露

谱线展宽和谱泄露导致:难以确定频率的位置和幅度;降低频率分辨率。

产生旁瓣导致:

产生假信号;

淹没小信号。

EXAMPLE

We can find that windowing smears or broadens the impulse in theoretical Fourier representation, and thus reduces the ability to resolve sinusoidal signals that are closely spaced in frequency . The amplitude of one spectrum is affected by the amplitude of another and vice versa when two components are closely spaced in frequency. This interaction is called leakage(泄露). The component at one frequency leaks into the vicinity of anothercomponent due to the spectral smearing introduced by the window.So reduced resolution and leakage are the two primary effects on the spectrum as a result of applying a window to the signal. The resolution is influenced primarily by the width of the main lobe of ,while the degree of leakage depends on the relative amplitude of the main lobe and the side lobes of . We define the frequency resolution(频率分辨率,能够分辨的最小频率间隔) is equal to the width of the main lobe of .注意:频率分辨率=窗频谱的主瓣宽,与DFT谱线间距无关

( )jW e

( )jW e

( )jW e

42M

64M

32M窗长 ( )jV e

54M

Conclusion: increase M can increase frequency resolution,but decrease time resolution.

ml能分辨的最小频率间隔

EXAMPLE

2[ ] (cos( )14

40.75cos( ))15

[ ] (32 ~ 64, 5.48)

[ ] [ ] [ ]

x n n

n

w n kaiserMv n x n w n

凯泽窗

2[ ] (3.5*cos( )14

23.5*0.75cos( ))25

[ ]:[ ] :

32

R

hanning

v n n

n

w n redw n blue

M

Conclusion: shape of window has effect on frequency resolution

EXAMPLE

(最大幅度归一化成0dB):

(a)矩形(b)三角(c)汉明(d)汉宁(e)布来克曼

(a)-(e) 旁瓣衰减增加,主瓣宽度增加

Blackman窗族(M=50)FT的对数幅度

频率分辨率Δωmin或窗主瓣宽Δml与窗形状和窗长的定量关系

窗频谱的旁瓣相对幅度(dB)

窗频谱的主瓣宽

min

min min

min min

/2

/ 22

ml

s

s

DM

DTMDf fM

ml

只与窗形状有关(定性记住)

与窗形状和窗长都有关(要背)

4 / M大致估计为

[ ] cos(0.5 ) cos(0.2 )x n n n

EXAMPLE

采用汉宁窗分析其频谱,问最小窗长?

8 0.3

27

ml MM

解:

2. the effect of spectral sampling

( ) [ ]jV e V k

The DFT of the windowed sequence provides samples of .Spectral sampling can sometimes produce misleading results.

( )jV e

2 4[ ] cos( ) 0.75cos( )14 15

[ ]' 64[ ] [ ] [ ]R

x n n n

w n s length Mv n x n w n

N=128

N=64

After sampling

Before sampling

DFTpointN=64

V[k]

峰值未取到

增加N峰值取到

EXAMPLE

N=64

[ ] ( ) ( )j jV k X e V e 只取到峰值和零值

N=128Conclusion : increase N can

fine the sampling of the spectrum.

增加N其他值也取到

EXAMPLE

2 2[ ] cos( ) 0.75cos( )16 8

[ ]' 64[ ] [ ] [ ]R

x n n n

w n s length Mv n x n w n

2 4[ ] cos( ) 0.75cos( )14 15

[ ]( 32, 5.48)[ ] [ ] [ ]

point 32 ~ 1024

kaiser

x n n n

w n length Mv n x n w nDFT N

Conclusion: increase N (point of DFT) can’t increase frequency resolution.

DFT点数N=32

N=64

N=128

N=1024

EXAMPLE

6.6小结

以具有两个频率的正弦信号为例研究加窗和频域取样的影响(1)加窗:

增大窗长或改变窗形状(例如选矩形窗)可提高频率分辨率;改变窗形状可改变旁瓣相对幅度。通常用增加窗长的方式增大频率分辨率,而不用改变窗形状,因为在增加频率分辨率的同时会增大旁瓣幅度。但窗长M太大,时间分辨率降低。

(2)频域取样:取样点数N >=窗长M(满足频域取样定理)才能重构时域;增大点数N不能提高分辨率,但可取到每个细节,以至于线性内插就可得到较精细的连续频谱。但点数大运算量也大。

所以N和M的取值需折中。

每条谱线对应的频率:

采样率与抗混迭滤波器的截止频率的关系:2s cf f

相临谱线间的频率间距与DFT点数的关系:

2 // 2 // 2 /

s

s

NT f N

f f N

2 /2 /

/

k

k s

k s

k Nf k N

f f k N

频率分辨率(可分辨的最小频率间隔=窗频谱的主瓣宽)与窗形状和窗长M的关系(注意单位是弧度):

矩形窗:

三角/汉宁/明窗:

布莱克曼窗:

min

min

min

4 /8 /12 /

ml

ml

ml

MMM

第6章总结

1、DFT的引出采用FT或DFS,但是与它们独立成一体系。2、DFT点数大则取样密,能重构时域;点数少(时域混迭后变换),则取样稀,不能重构时域。3、DFT的性质4、循环卷积与线性卷积(前者的DFT是后者FT的取样)的关系类似重构信号与原信号的关系,无混迭或有混迭。DFT可用于计算线性卷积(FIR系统的实现)。5、DFT可用于分析有限长信号的频谱。

DFT还可用于分析FIR和IIR的频响。DFT用于分析无限长或不定长信号频谱时:需要分段,有

加窗效应;若DFT点数足够,频域取样引起的误差可忽略。

重点和难点:频域取样及DFT性质

第6章作业

6-12(DFT定义)

6-28(时域内插0)6-30(频域取样)

6-31(DFT性质)

6-32(c)-(e)(循环移位、时域奇数点清0)6-34(频域抽选、内插0等)

6-37(DFT性质)

MATLAB练习题

(1)验证DFS与DFT的关系:(A)周期序列

画出其DFS和傅立叶变换的幅度特性。(B)有限长序列

画出其50点DFT和傅立叶变换的幅度特性。提示:可以调用的函数有fft()等。

(2)画出以下序列的10点DFT的实部和虚部:(A)(B)

比较二者的实部和虚部。提示:可以调用的函数有fft()、real()和imag()等

[ ] cos 0.1 0.5cos 0.04x n n n

[ ] cos 0.1 0.5cos 0.04 ,0 49x n n n n

1[ ] 0.2 ,0 9nx n n n 1 1 10

2[ ] [((10 )) ][ ] ,0 9

2x n x nx n n

Recommended