Digital Signal Processing

12/17/2015 1
• Reading

• Mitra 2nd edition:

• Chapter 10
– Sections: 10.8.1, 10.8.2, 10.8.3, 10.8.4,

12/17/2015 2
 Two-channel Quadrature-Mirror filterbank (QMF)

• Maximally “critically” sampled filterbank

• Linear Time Varying (LTV) filterbank
• 𝐻0 𝑧 & 𝐺0 (𝑧) are lowpass filters
• 𝐻1 𝑧 & 𝐺1 (𝑧) are highpass filters
• 𝐻1 𝑧 = 𝐻0 (−𝑧) “mirror”
Analysis of the QMF filterbank

Vk ( z )  H k ( z ) X ( z )

Uk ( z) 
1
2
 Vk ( z1 2 )  Vk (  z1 2 )

Vˆk ( z )  U k ( z 2 ) 0  k 1

Vˆk ( z )  Vk ( z )  Vk (  z )
1
2
 H k ( z ) X ( z )  H k (  z ) X (  z )
1
0  k 1
2
The filterbank output
Y ( z )  G0 ( z )Vˆ0 ( z )  G1 ( z )Vˆ1 ( z )
12/17/2015 3
 The reconstructed signal at the filterbank output
Y ( z )  G0 ( z )Vˆ0 ( z )  G1 ( z )Vˆ1 ( z )

Y ( z) 
1
H 0 ( z )G0 ( z )  H1 ( z )G1 ( z )X ( z )
2
 H 0 (  z )G0 ( z )  H1 (  z )G1 ( z )X (  z )
1
2

Y ( z )  T ( z ) X ( z )  A( z ) X (  z )
The distortion transfer function

T ( z) 
1
H 0 ( z )G0 ( z )  H1 ( z )G1 ( z )
2
The aliasing component
A( z )  H 0 (  z )G0 ( z )  H1 (  z )G1 ( z )
1
2

12/17/2015 4
 Y ( z )  T ( z ) X ( z )  A( z ) X (  z )

T ( z) 
1
H 0 ( z )G0 ( z )  H1 ( z )G1 ( z )
2
A( z )  H 0 (  z )G0 ( z )  H1 (  z )G1 ( z )
1
2
• alias-cancellation condition
H 0 (  z )G0 ( z )  H1 (  z )G1 ( z )  0
let G0 ( z )  H1 (  z ) “LPF”
G1 ( z )   H 0 (  z ) “HPF”
 H 0 (  z ) H1 (  z )  H1 (  z ) H 0 (  z )  0

•  Y ( z)  T ( z) X ( z) “ the filterbank becomes LTI”

• perfect reconstruction (PR) condition
y[n ]  x[n  l ]  T ( z )  z l
H 0 ( z )G0 ( z )  H1 ( z )G1 ( z )  2 z l
 H 0 ( z ) H 1 (  z )  H 1 ( z ) H 0 (  z )  2 z l

12/17/2015 5
• example: trivial two-channel PR QMF filterbank

H 0 ( z)  1 G0 ( z )  z 1
H1 ( z )  z 1 G1 ( z )  1
• distortion transfer function
1
T ( z )  {H 0 ( z )G0 ( z )  H1 ( z )G1 ( z )}
2
1
T ( z )  {z 1  z 1}  z 1
2
• aliasing term
1
A( z )  {H 0 (  z )G0 ( z )  H1 (  z )G1 ( z )}
2
1 1 1
A( z )  {z  z }  0
2

12/17/2015 6
 • example

1 1
H 0 ( z)  (1  z 1 ) G0 ( z )  (1  z 1 )  H1 (  z )
2 2
1 1
H1 ( z )  (1  z 1 ) G1 ( z )  ( 1  z 1 )   H 0 (  z )
2 2
• distortion transfer function
1
T ( z )  {H 0 ( z )G0 ( z )  H1 ( z )G1 ( z )}
2
1
T ( z )  {(1  z 1 )(1  z 1 )  (1  z 1 )( 1  z 1 )}  z 1
4
• alias term
1
A( z )  {H 0 (  z )G0 ( z )  H1 (  z )G1 ( z )}
2
1
A( z )  {(1  z 1 )(1  z 1 )  (1  z 1 )( 1  z 1 )}  0
4

analysis synthesis

12/17/2015 7
 Polyphase implementation of the two-channels QMF
• The polyphase decomposition of the • The polyphase decomposition of the
analysis filters 𝐻0 𝑧 &𝐻1 (𝑧) and noting synthesis filters 𝐺0 𝑧 & 𝐺1 (𝑧) and noting
that 𝐻1 𝑧 = 𝐻0 (−𝑧) that 𝐺0 𝑧 = 𝐻0 𝑧 , and
𝐺1 𝑧 = −𝐻1 𝑧 = −𝐻0 (−𝑧)
H 0 ( z )  E0 ( z 2 )  z 1E1 ( z 2 )
1
G0 ( z )  E0 ( z 2 )  z 1E1 ( z 2 )
H1 ( z )  E0 ( z )  z E1 ( z )
2 2

G1 ( z )   E0 ( z 2 )  z 1E1 ( z 2 )
• In matrix form
• In matrix form
 H 0 ( z ) 1 1   E0 ( z )  2

 H ( z )   1  1  1 2 
G0 ( z )   1 1 
G1 ( z )  z 1E1 ( z 2 ) E0 ( z 2 ) 
 1     z E1 ( z ) 
1  1

• distortion transfer function

1
T ( z )  {H 0 ( z )G0 ( z )  H1 ( z )G1 ( z )}
2
T ( z )  2 z 1E0 ( z 2 ) E1 ( z 2 )

12/17/2015 8
• example:

H 0 ( z )  1  z 1
E0 ( z 2 )  1
E1 ( z 2 )  1

 H 0 ( z ) 1 1   1 
 H ( z )   1  1  z 1 
 1    
1 1 
G0 ( z ) 
G1 ( z )  z 1 1  
1  1

12/17/2015 9