Lattice-Structure for

FIR filters

Spring 2009

© Ammar Abu-Hudrouss - ١
Islamic University Gaza

Lattice Structures

Lattice filter is used in extensively in digital speech

processing and in implementing adaptive filters
H m ( )  Am ( z ) m  0,1,2,......, M  1
Where by definition
m
Am ( z )  1    m ( k ) z  k m 1
k 1

Which leads to
1 n0
h ( n)  

 m ( n ) n  1,2,...., m

Digital Signal Processing

Slide ٢

١
 Lattice Structures

So the output can be expressed as

m
y ( n )  x ( n )    m ( k ) x( n  k )
k 1
This can be expressed by the direct form as

x(n) z -1 z -1 z -1 z -1

1 αm(1) αm(2) αm(3) αm(m -1) αm(m )

+ + + + + y(n)

Digital Signal Processing

Slide ٣

Lattice Structures

The lattice filter is generally described by the

following set of equations

f 0 (n)  g 0 (n)  x(n )

f m (n)  f m 1 (n)  K m g m 1 (n  1) m  1,2,......, M  1
g m (n)  K m f m1 (n)  g m 1 (n  1) m  1,2,......, M  1

The output of the (M-1) stage filter

y (n)  f M 1 (n)

Digital Signal Processing

Slide ٤

٢
 Lattice Structures

f 0(n) f 1(n) f 2(n) f M-2(n) y(n)=f M-1(n)

First Second Second
Stage Stage Stage
gM-1(n)
g0(n) g1(n) g2(n) gM-2(n)

+ fm(n)
fm-1(n)
Km

Km

z -1 + gm(n)
gm-1(n)

Digital Signal Processing

Slide ٥

Lattice Structures

Suppose we have a filter of order one (m = 1). The

output of this filter can be expressed as

y ()n )  x ( n )   1 (1) x ( n  1)
f0(n
+ f1(n) =y(n)
K1

x(n)
K1

z -1 + g1(n)
g0(n)

f1 (n)  x (n)  K1 x (n  1) K 1   1 (1)

g1 (n)  K1 x(n)  x(n  1)

Digital Signal Processing
Slide ٦

٣
 Lattice Structures

Suppose we have a filter of order one m = 2. The

output of this filter can be expressed as

y ( n )  x ( n )   2 (1) x ( n  1)   2 ( 2 ) x ( n  2 )
+ + f2(n) =y(n)
f0(n) f1(n)
K1 K2
x(n)
K1 K2

z -1 + z -1 + g2(n)
g0(n) g1(n)

Digital Signal Processing

Slide ٧

Lattice Structures

The output from the lattice implementation is

f 2 ( n )  f1 ( n )  k 2 g 1 ( n  1)
Substituting for g1(n - 1) and f1(n)
f 2 (n)  x (n)  K1 x(n  1)  K 2 K1 x (n  1)  K 2 x (n  2)
f 2 (n)  x(n)  ( K1  K 2 K1 ) x(n  1)  K 2 x(n  2)
Comparing that with
y ( n)  x( n)   2 (1) x( n  1)   2 ( 2) x( n  2)

 2 (1)  K1 (1  K 2 )  2 (2)  K 2
 2 (1)
K1  K1   2 (2)
1   2 ( 2)
Digital Signal Processing
Slide ٨

٤
 Lattice Structures

The output from the lattice implementation is

g 2 ( n )  K 2 f1 ( n )  g 1 ( n  1)
Substituting for g1(n - 1) and f1(n)
g 2 (n)  K 2 x(n)  K 2 K1 x (n  1)  K1 x(n  1)  x (n  2)
g 2 (n)  K 2 x(n)  K1 (1  K 2 ) x (n  1)  x(n  2)
g 2 (n)   2 (2) x(n)   2 (1) x (n  1)  x (n  2)
m
g m (n)    m (k ) x (n  k )
k 0

 m ( k )   m (m  k ) k  0,1,......, m

Digital Signal Processing

Slide ٩

Lattice Structures
m
g m (n)    m (k ) x (n  k )   m (n) * x (n)
k 0
m
f m ( n)    m ( k ) x( n  k )   m ( k ) * x ( k )
k 0

Convert to z-transform

Gm ( z )  Am ( z ) X ( z ) Fm ( z )  Bm ( z ) X ( z )
Then if we convert the recursive lattice equation to z domain

F0 ( z )  G0 ( z )  X ( z )
Fm ( z )  Fm1 ( z )  K m z 1Gm 1 ( z )
Gm ( z )  K m Fm 1 ( z )  z 1Gm 1 ( z )

Digital Signal Processing

Slide ١٠

٥
 Lattice Structures

Divide the previous equation by X (z)

A0 ( z )  B0 ( z )  1
Am ( z )  Am 1 ( z )  K m z 1 Bm 1 ( z ) m  1,2,3,....., M  1
Bm ( z )  K m Am1 ( z )  z 1  m 1 ( z ) m  1,2,3,....., M  1
Divide the previous equation by X(z)

 Am ( z )  1 K m   Am 1 ( z ) 
 B ( z )   K  
1   z 1Bm 1 ( z )
 m   m

Digital Signal Processing

Slide ١١

Lattice to Direct Form

To get the direct form coefficients from the lattice constants

we have
A0 ( z )  B0 ( z )  1
Am ( z )  Am 1 ( z )  K m z 1Bm1 ( z )
m m
Bm ( z )    m ( k ) z  k    m (l ) z l  m
k 0 l 0
m m
Bm ( z )    m (l ) z l m  z m   m (l ) z l
l 0 l 0
m 1
Bm ( z )  z Am ( z )
Solve the previous equation recursively to get Am(z)

Digital Signal Processing

Slide ١٢

٦
 Lattice to Direct Form

Example: Given a three stage lattice filter with coefficients K1 =

0.25, K2 = 0.5 and K3 = 1/3, determine the FIR filter coefficients
for the direct-form structure.

A1 ( z )  A0 ( z )  K 1 z 1 B0 ( z )
 1  (1 / 4) z 1
By reversing the order of A1(z), we get
B1 ( z )  1 / 4  z 1
2nd stage
A2 ( z )  A1 ( z )  K 2 z 1 B1 ( z )
 1  (3 / 8) z 1  (1 / 2) z 2
B2 ( z )  (1 / 2)  (3 / 8) z 1  z 2
Digital Signal Processing
Slide ١٣

Lattice to Direct Form

The third stage

A3 ( z )  A2 ( z )  K 3 z 1 B2 ( z )
 1  (13 / 24) z 1  5 / 8z  2  (1 / 3) z  2

By performing the inverse z transform

 3 ( 0)  1
 3 (1)  13 / 24
 3 ( 2)  5 / 8
 3 (3)  1 / 3

Digital Signal Processing

Slide ١٤

٧
 Direct Form to Lattice

To get the lattice constants from the direct form coefficients

Am ( z )  Am 1 ( z )  K m z 1 Bm 1 ( z )

Am ( z )  Am 1 ( z )  K m Bm ( z )  K m Am1 ( z )

Solve the previous equation to get Am-1(z)

Am ( z )  K m Bm ( z )
Am 1 ( z )  m  M  1, M  2,....,1
1  K m2
This is a step-down recursion

Digital Signal Processing

Slide ١٥

Direct Form to Lattice

Example Determine the lattice coefficients corresponding

Example:
to the FIR filter
H ( z )  A3 ( z )  1  (13 / 24) z 1  (5 / 8) z 2  (1 / 3) z 3
K3  1 / 3
B3 ( z )  (1 / 3)  (5 / 8) z 1  (13 / 24) z 2  z 3
The step down relationship with m = 3
A3 ( z )  K 3 B3 ( z )
A2 ( z ) 
1  K 32
A2 ( z )  1  (3 / 8) z 1  (1 / 2) z 2

K2  1/ 2
Digital Signal Processing
Slide ١٦

٨
 Direct Form to Lattice

B2 ( z )  (1 / 2)  (3 / 8) z 1  z 2

The step down relationship with m = 2

A2 ( z )  K 2 B2 ( z )
A1 ( z ) 
1  K 22
A1 ( z )  1  (1 / 4) z 1

K1  1 / 4

Digital Signal Processing

Slide ١٧