Course 18.327 and 1.

130 Wavelets and Filter Banks

Refinement Equation: Iterative and Recursive Solution Techniques; Infinite Product Formula; Filter Bank Approach for Computing Scaling Functions and Wavelets

Solution of the Refinement Equation

(t) = 2 h0[k] (2t-k)
k=0
N

First, note that the solution to this equation may not always exist! The existence of the solution will depend on the discrete-time filter h0[k]. If the solution does exist, it is unlikely that (t) will have a closed form solution. The solution is also unlikely to be smooth. We will see, however, that if h0[n] is FIR with h0[n] = 0 outside 0 n N then (t) has compact support: (t) = 0 outside 0 < t < N

Approach 1 Iterate the box function (0)(t) = box function on [0 , 1]

N

(t)

(i + I) (t) = 2 h0[k] (i) (2t k)

k=0

If the iteration converges, the solution will be given by lim (i)(t)

i

This is known as the cascade algorithm.

Example: suppose h0[k] = {, , } (i+1) (t) = (i)(2t) + (i) (2t 1) + (i) (2t 2) Then (0)(t) 1 (0)(t) 1 1 2 0 (2)(t) 1 0 1 2 1 3 2 2 t 1 2 t 0 (3)(t) 1 1 2 1 3 2 2 t

1 2

3 1 2

t
4

Converges to the hat function on [0, 2]

Approach 2 Use recursion First solve for the values of (t) at integer values of t. Then solve for (t) at half integer values, then at quarter integer values and so on. This gives us a set of discrete values of the scaling function at all dyadic points t = n/2i. At integer points: (n) = 2 h0[k] (2n k)
k=0 N

Suppose N = 3 (0) = 2 h0[k] (-k) (1) = 2 h0[k] (2-k) (2) = 2 h0[k] (4-k) (3) = 2 h0[k] (6-k)
k=0 k=0 3 k=0 3 k=0 3 3

Using the fact that (n) = 0 for n < 0 and n > N, we can write this in matrix form as (0) h0[0] (0) h0[2] h0[1] h0[0] (1) (1) = 2 h0[3] h0[2] h0[1] (2) (2) h0[3] (3) (3)
6

Notice that this is an eigenvalue problem = A where the eigenvector is the vector of scaling function
values at integer points and the eigenvalue is = 1.
Note about normalization:
Since (A - I) = 0 has a non-unique solution,
we must choose an appropriate normalization for
The correct normalization is
(n) = 1
n

This comes from the fact that we need to satisfy the partition of unity condition, (x-n) = 1.
n 7

At half integer points: (n/2) = 2 h0[k] (n-k)

k=0 N

So, for N = 3, we have (1/2) (3/2) (5/2) h0[1] h0[0] 2 h0[3] h0[2] h0[1] h0[0] h0[3] h0[2] (0) (1) (2) (3)

Scaling Relation and Wavelet Equation in Frequency Domain

(t) = 2 h0[k] (2t k)
k

(t)e-it dt

= 2 h0[k]
k k

(2t k) e-it dt
()ei( + k)/2 d ( ) e
- i/2 d

= 2 h0[k]
k

= 2 h0[k]e-ik/2

^ ). i.e. () = H0( 2

^( ) 2 M 678

^ = H0( ) . H0 ( ) . ( ) 2 4 4

j=1

^
(0) =

123

H0( 2j )

^ (0)

(t) dt = 1 (Area is normalized to 1)

10

So

^ () = H0 (j ) j=1 2

Infinite Product Formula

Similarly w(t) = 2 h1[k] (2t k)

k

leads to

^ ) = H ( ) ^ w( 1 2

( 2)

Desirable properties for H0(): H(0) = 1, so that ^ (0) = 0 H() should decay to zero as , so that ^() d <
2 11

Computation of the Scaling Function and Wavelet Filter Bank Approach y0[n] 2 x0[n] 2 x1[n] x2[n] Normalize so that h0[n] = 1.
n
[n] H0() y Y () 2
1 1

[n] H0() y Y () 2
2 2

H0()

y3[n] Y3()

H1() 2 H1() 2 H1()

12

i. Suppose y0[n] = [n] and xk[n] = 0. Y0() = 1 Y1() = Y0(2) H0() = H0() Y2() = Y1(2) H0() = H0(2)H0() Y3() = Y2(2) H0() = H0(4) H0(2) H0() After K iterations: YK() = H0(2k)
k=0 K-1

What happens to the sampling period? Sampling period at input = T0 = 1 (say) Sampling period at output = TK = K

13

Treat the output as samples of a continuous time signal, yc (t), with sampling period K:
K

yK[n] =

1 c y (n/2K) 2K K

^ K YK() = Yc (2 ) K

; - c (yK (t) is chosen to be bandlimited)

Replace 2K with :
K-1 K ^ Yc () = YK(/2k ) = H0(/2K-k ) = H0( /2j ) ; K k=0 j=1

-2K 2K So
^c ^ lim YK () = H0( /2j) = () j=1 14

 2K yK[n] converges to the samples of the scaling function, (t), taken at t = n/2K. ii. Suppose y0[n] = 0, x0[n] = [n] and all other xk[n] = 0 YK() = H1(2K-1) H0(2k)
k=0 K-2

Then
K-2
^c K-k)
YK() = YK(/2K) = H1( ) 2 H0(/2 k=0 K-1 j=1

. ) ) H0( 1 = H1( 2 2 2j So
K

^c ^ (/2) = w( ^ ) lim YK () = H1(/2)

2KyK[n] converges to the samples of the wavelet, w(t), taken at t = n/2K.

15

Support of the Scaling Function yk-1[n] 2 v[n] h0[n] yk[n]

length {v[n]} = 2 length {yk-1[n]} - 1 Suppose that h0[n] = 0 for n < 0 and n > N length {yk[n]} = length {v[n]} + length {h0[n]} 1 = 2 length {yK-1[n]} + N 1 Solve the recursion with length {y0[n]} = 1 So length {yk[n]} = (2K 1)N + 1
16

c(t)} = T . length {y [n]} i.e. length {yK K K

(2K 1) N + 1 2K N-1 2K

(t)

= N-

lim K length {(t)} = N 0 N t

So the scaling function is supported on the interval [0, N]

17

18

Matlab Example 6
Generation of orthogonal scaling functions and wavelets

MATLAB M-file

MATLAB M-file

By Inverse DWT

20

10

By Recursion

21

Comparison

22

11

Matlab Example 7
Generation of biorthogonal scaling functions and wavelets.

MATLAB M-file

Primary Daub 9/7 Pair

24

12

Dual Daub 9/7 Pair

25

13