Multi-rate Digital Signal processing

1
 Multi-rate Systems
Discrete time systems with unequal sampling rates at
various parts of the system are called multi-rate systems.

3 sampling rates in digital audio:-

32 KHz in broadcasting
44.1 KHz in digital CD
48 KHz in digital audio tape.

Sampling rate alteration devices:

Down sampler (Decimator)
Up sampler (Interpolation)
Process of reducing the sampling rate. - Decimation
Process of increasing the sampling rate - Interpolation.
2
 Time domain characterization
• M – Fold decimation

M yD (n)
x(n)

yD n  xMn
: M is an Integer.

3
Only those samples of x(n) which occur at multiples of M are
retained

x(n)

-2
n
-1 0 1 2 3
M=2 yD(n)

-2 -1
n
0 1

. 1
Sampling rate of y D ( n)  sampling rate of x(n)
M
4
Decimator– Sub sampler – down sampler –
sampling rate compressor- Compressor

M=3

n
0 1 2 3 4 5 6 7 8 9 10 11 12

0 1 2 3 4
n

5
 The L – fold Interpolator
x(n) L yU(n)

yU (n) x(n ) ; if n = an integer multiple of L;

=0 L ; Otherwise

L=2
x(n)

0 1 2 3 4 5 6 7 8 9

yU (n)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

x (n) can be recovered from yU (n) by L fold decimation

6
 Application : -

• Up sampler and down sampler used together in

multirate signal processing.

x (n) V(n) Vu (n)

2 2

z -1 z -1

W (n)
Wu(n)
2 2 +
y (n)

7
 x (n) V(n) Vu (n)
2 2

Z-1 Z-1
x(n-1) 2 2 W (n) +
W (n) u y (n)

n: 0 1 2 3 4 5 6 7 8

x (n): x (0) x (1) x (2) x (3) x (4) x (5) x (6) x (7) x (8)

V(n): x (0) x(2) x (4) x (6) x (8) x (10) x (12) x (14) x (16)

Vu(n): x (0) 0 x (2) 0 x (4) 0 x (6) 0 x (8)

Vu(n-1): 0 x (0) 0 x (2) 0 x (4) 0 x (6) 0

x(n-1) x(-1) x (0) x (1) x (2) x (3) x (4) x (5) x (6) x(7)

W(n): x (-1) x(1) x (3) x (5) x (7) x (9) x (11) x (13) x (15)

Wu (n): x(-1) 0 x (1) 0 x (3) 0 x (5) 0 x (7)

y(n) x (-1) x (0) x (1) x (2) x (3) x (4) x (5) x (6) x (7)

y(n)= Vu (n-1) + Wu (n) = x (n-1)

8
Frequency domain characterization
Relation between the spectrum of the input and the output
of a factor of 2 up sampler.

xu (n) = x[n/2] , n=0 ,  2,  4,…

=0 otherwise

X ( z)   u
x
n  
( n ) z n



x ( n / 2) z n
n  
n even

 x[m] z
m
2 m = X(z2)

For factor of L up sampler, Xu (z) = X(zL)

9
Take the DTFT
X ( e j )
X ( e j )

 
 2 2

X u (e j )
Image

L=2

 
 /2 2

L=2 leads to a 2 fold repetition of X (e j )

FT is compressed by a factor of 2
Imaging L -1 additional images of the input spectrum.
10
 Down Sampler

y(n) = x [nM] y ( z)   x[nM ] z n
D n
 x(n) n  0,  M , 2M ,
Let x [n] 
int 0 otherwise
x (n)  c(n) x(n)
int

1 n  0, M , 2M
c(n)  
0
 otherwise

M 1
x
int
(n)  c(n) x(n)  1
 W  kn x(n)
M k 0 M
 j 2
W e M
M
11
 M 1
x (n)  c(n) x(n)  1
 W  kn x(n)
int M k 0 M

X (z)  
 c(n) x(n) z  n
int
n  
 

1  M 1 W  kn  n
    x(n) z
M n   k  0 M 

 

1 M 1   W  kn x(n) z  n 
   M 
M k 0  n  
 

 n
 x(n) zW
 
 M 1
1  k 
   
M k 0  M 
 n  
   


 M 1 X ( zWM k )
1

M k 0      

12
n: 0 1 2 3 4 5 6 7 8

x (n): x (0) x (1) x (2) x (3) x (4) x (5) x (6) x (7) x (8)

x(2n): x (0) x(2) x (4) x (6) x (8) x (10) x (12) x (14) x (16)

xint(n): x (0) 0 x (2) 0 x (4) 0 x (6) 0 x (8)

xint(2n): x (0) x (2) x (4) x (6) x (8) x (10) x (12) x(14) x (16)

xint(2n)=x(2n)

xint(Mn)=x(Mn)


Y (Z )   x[nM ]z n
D n

  x [nM ]z n
n int

  x [k ]z k M  X [ z1 M ]
k  int int
  1 
 M 
Y ( z)  X z 
D int  
 
1
 M 1 X ( z M WM
1
 k )
M k 0
 2 

 j j k 

 M e M 
 M 1 X
1
 e



M k 0  
 
 
 

j (  2k )
 
1 M 1 
M

  X  e 

M k 0  
 
M 1
YD z    X e 
1 j (   2k ) M

M k 0 14
 Graphical interpretation
1. Stretch X e 
j
by a factor M to obtain 
X e j / M 
2. Create (M-1) copies of the stretched version, by shifting it uniformly in
successive amounts of 2 
3. Add all these shifted, stretched versions to the un-shifted stretched
version and divide by M.


X e j / M  does not have period 2 
But the result is periodic with period 2 

 
X e j

-2 - 0  2
2X e j / 2
  j   22  
2X e  
 
 

 2 X  e j / 2 
-2 - 0  2 15
    2 
j 
  X Cos 
    
X (e  2 
    j Sin    
2  2 
    Sin  Cos 
  
 X Cos   Cos  Sin
   Sin  j 
 Cos Sin 
 
 2 2  2 2 
   
 X  Cos  j Sin 
 2 2 
    
 X   Cos  j Sin 

 2 2 
 

 j 
 X e 2 
 
 

16
Graphical interpretation

X ( e j )
Fourier transform
Input signal
YU (e j ) Interpolator
(L=5)

Interpolated signal

j / 2 X (  e j / 2 ) Decimator(M=2
X (e )
Decimated signal

17
Aliasing created by decimation
The stretched version can overlap its
shifted replicas.
We cannot recover x(n) from the
decimated version .

This overlap effect is called aliasing

 
X e j

-2 - 0  2

X e j / 2
  j   22  
X e  
 
 

 X  e j / 2 
-2 - 0  2

18
 Avoiding Aliasing.

Alaising can be avoided if X (e j )  0 for    / M

i.e. x(n) is a low pass signal band limited to .    / M

X(ej)

-2 - -/2 0 +/2  2 

2Y(ej)
M=2
X(ej/2) X(-ej/2)

-2 - 0  2 

19
Demonstrating the frequency domain effects of decimating
j
with M=3
X (e )

 /3  /3

(a) X ( e j / 3 )

 

j   2
X (e 3
)
(b)

j   4
© X (e 3
)

(adding a, b, c) YD (e j )

20
-6  -4  -2 0 2 4 6
 Decimation Filters
To make sure that the signal is band limited, a filter
precedes the decimator
Decimation Filter
x(n) H(z) M y(n)

Decimation Decimation
filter

H ( e j )
1

Typical response of the decimation filter


 
 p M s

21
 Interpolation Filter.
This follows the interpolator.
This suppresses images.
yu(n)
x(n) L H(z) y(n)

Expander Interpolation filter

H ( e j )
1



 p  s
L

y (n)  yu (n)  h(n)

The zero valued samples introduced by the expander are filled with
interpolated values. 22
x(n) yu (n)
L H(z) y(n)

3
x(n)
-1 0 1 2 4
n

yu(n) , L=2
-1 0 1 2 3 4 n

Filter, h(n)
-1 0 1 2 3 4

y(n)

0 1 2
n

23
 yD n V(n) H(z)
x(n) H(z) 3 3 Y(n)

X(ej)
1


 2  /3  /3 2

1/3
YD e j  
0


 2
  2

1/3
 
V e j


 /3
 2 2
H(ej)


 /3 2

Y(ej)

 
 2 2 24
3
Interconnection of building blocks
Some common blocks in multirate systems

a a
M M

x1(n)
x1(n) M
M

x2(n) M
x2(n)

x(n) M
X
x(n) M
X
d(n) M
d(n)

• Input- output relations are invariant

• All the above identities hold if the decimators are replaced
with interpolators
25
 Noble Identities for multirate systems

y1(n) x(n) G(zM) M Y2(n)

M G(Z)
x(n)

Identity 1

x(n) Y3(n)
L L G(zL)
G(Z)
y4 (n)

Identity 2

y1(n) x(n) V2(n)

V1(n) M L Y2(n)
L M
x(n)

This interchange is possible only if L and M are relatively prime.

26
 Noble Identities Proof

Identity 1

y1(n) x(n)
x2(n) M Y2(n)
M G(Z)
x(n) G(zM)

Identity 1

If the function G(z) is a ratio of polynomials in z or z-1, then Noble identity 1 holds

27
 y1(n x(n) M Y2(n)
M G(Z) )
x(n) G(zM)
M 1  1 M 

Y ( z)  1  Xz W k G( z)
Identity 1
1 M k o  M 


X ( z)  X ( z) G( z M )
2
1 M 1  1 M 



1
M k M 

Y ( z)  k
M k
Xz W  G (z W ) 
2 0  M   M 
   

1 M 1  1 M 
  kM 
  Xz W k  G  z W 

M k 0  M  
M 
 

kM  j 2kM
W e M 1
M

1 M 1  1 M 

Y ( z )  k
M k
X  z W  G( z)
2 o  M 

 Y ( z) 28
1
Identity 2

y3(n) x(n)
Y4(n)
G(z) L
x(n) L G(zL)

Identity

Y ( z)  X ( z L ) G( z L )
3
Y ( z )  X ( z L ) G( z L )
4
Y ( z)
3
29
 V1(n) V2(n)
Y1(n)
L M x(n) M L Y2(n)
x(n)

This interchange is possible only if L and M are relatively prime.

V ( z)  X ( z L )
1
    L
1 M 1  1 M 
1 M 1  1
M 
Y ( z)   V z W k   Xz W k 
1 M k o 1 
M  M k o   M 

   
 L 
1 M 1 
M kL 
M k
 Xz W 
 M 
o  

 
M 1  1 
V ( z)  1 M k
2  
X z
M k o 
W 
M 
 
 L 
1 M 1 
M 
Y ( z)   
X z W k 
2 M k o  M 
 
30
  j 2 k
W k e M ; k  0,1,2,....M 1
M
 j 2 kL
W kL  e M
M
 j 2 L
W Le M
M
The set of M terms in Y1(z) = set of M terms in Y2(z) ; for k=0,1,2,M-
1 [only when L and M are relatively prime.]
 Y1(z)= Y2(z); when L and M are relatively prime.

31
Find Y(z) in terms of X(z) x (n) V(n) Vu (n)
2 2

z -1 z -1

W (n)
Wu(n)
2 2 +
y (n)

1 1 1 1
V ( z )  X ( z )  X ( z 2 )
2
2 2
 1 12 1 1 12 1 
W ( z )   z X ( z )  z X ( z 2 )
2

2 2 
1 1
Vu ( z )  X ( z )  X ( z )
2 2
 1 1 1 1 
Wu ( z )   z X ( z )  z X ( z )
2 2 
Y ( z )  z 1Vu ( z )  Wu ( z )  z 1 X ( z )  x(n  1)
 Fractional Sampling Rate Alteration

The sampling rate of a signal can be altered by an integer

factor L or M
In some applications, it may be necessary to change the rate
by a rational fraction such as L/M
2
Eg. For a signal band-limited to  
3

 
X e j

 2 0 2
3 3

33
 Fractional Sampling Rate Alteration

The sampling rate of a signal can be altered by an integer

factor L or M
In some applications, it may be necessary to change the rate
by a rational fraction such as L/M
2
Eg. For a signal band-limited to  
3

 
X e j

 2 0 2
3 3

34
 We cannot decimate the signal by 2 because that
creates aliasing error.

 We have to decimate by 1.5 ( 3 2) so that the FT is as

shown below

 
Y e j

  2 0 2 
3 3

35
 x1(n) x2(n)
H(z)
L M
x(n) y(n)

Take L=2, M=3, The overall reduction of sampling rate is by the factor M  3
L 2
X (e j )
Y (e j )
 2 0 2


X (e j )  X (e2 j )
3 3

1


0 2 
3
3


3
1

H (e j )

 2
3 3

 36
3
If H(z) is a lpf with response as shown, then X (e j ) is the shaded
1
portion
Decimation by 3 results finally in Y (e j )
3
The time domain meaning of decimation by a factor of is shown
below. 2

37
 x1(n) x2(n)
H(z)
L M
x(n) y(n)

x(n)

x(n)

-1 0 1 2 3 4 5 n

x1(n)
L=2
-2 -1 0 1 2 3 4 n

x2(n)
L=2

-2 -1 0 1 2 3 4 5 6 7
n

y(n) M=3

-1 0 1 2 3 4
n