Course 18.327 and 1.

130 Wavelets and Filter Banks

M-band wavelets: DFT filter banks and cosine modulated filter banks. Multiwavelets.

M-channel Filter Banks Used in communication e.g. DSL 1 Scaling function, M-1 wavelets
X()

H0()
H1()


M
M

M
M

F0()
F1()
^ X()

HM-1()

M
[ X( +
n=0 M-1 2n M

FM-1()

(2)(2)X() = [X( )+X( + )] (M)(M)X() =

1

)]

Perfect Reconstruction
M-1

i.e.

k=0 1 M-1

n 1 Fk() M X( + 2M )Hk( + n=0

n X( + 2M ) Fk()Hk( +

M-1

2n M )
2n ) M

= e-iX() = e-iX()

M-1

M n=0

k=0

Matching terms on either side n=0 n0

M-1 k=0 M-1

Fk()Hk() = Me-i

no distortion no aliasing

k=0

n Fk()Hk( + 2M ) = 0

e.g. M = 3 F0()H0() + F1()H1()+F2()H2() = 3e-i F0()H0( + F0()H0( +

2 2 2 3 ) + F1()H1( + 3 ) + F2()H2( + 3 ) 4 4 4 3 ) + F1()H1( + 3 ) + F2()H2( + 3 )

=0 =0

Cast in matrix form [F0() F1() F2()] Hm() = [3e-i 0 H0() H0(+ 3 ) H0(+ 3 ) 2 Hm() = H1() H1(+ 3 ) H1(+ 43 ) H2() H2(+ 23 ) H2(+ 43 )
2
4

0]

Polyphase Representation
X(z)
z-1
z-1

X0(z)
X1(z) X2(z) Hp(z)  XM-1(z) 

Y0(z) Y1(z) Y2(z)


z-1

YM-1(z)

x[Mn] X0(z) = x[0] + z-1x[M] + z-2x[2M] + z-3x[3M] + x[Mn-1] X1(z) = x[1] + z-1x[M-1] + z-2x[2M-1] + x[Mn-2] X2(z) = x[2] + z1x[M-2] + z-2x[2M-2] + 

To recover X(z) from X0(z), X1(z), X2(z), X(z) = zkXk(zM)

k=0 M-1

Much more freedom than 2 channel case e.g. can have orthogonality & symmetry Consider Haar FB (M = 2) 1 1 Then Hp(z) = = F2 (2 pt DFT matrix) 1 -1

M-pt DFT matrix 1 1 1 1 1 w w2 wM-1 1 w2 w4 w2(M-1)   M-1 2(M-1) 1 w w w(M-1)(M-1

FM =

w = e-i M

Suppose Hp(z) = FM
DFT M pts M pts

DFT

Y0(z) = Xn(z)
n=0

M-1

M-1 M-1 = x[-n] + x[M-n] z-1 + n=0 n=0 M-1 M-1

M-1 Y1(z) = wnXn(z) = wnx[n] + wn-Mx[M-n] z-1 + n=0 n=0 n=0  M-1 M-1 M-1 Yk(z) = wknXn(z) = wknx[-n] + wk(n-M)x[M-n] z-1 + n=0 n=0 n=0

Terms in z-k are DFT coefficients of kth block of data. So filter bank performs a block DFT.
8

Modulation followed by filtering

x[n]
Z-1

x
x

p[n]
p[n]

(w0)-n

Z-1 (w)-n x p[n]  2 -n (w ) Z-1 x p[n]

For block DFT, p[n] = [1, 1, 1, , 1] 0 M1 Can generalize by using other prototype filters. p[n] is called the prototype filter.

(wM-1)-n
modulation filtering

If wkn is replaced by ck,n from DCT Cosine-modulated 2 1) ] ck,n = M cos[(k + )(n + M Filter Bank 2 + 2
M

Cosine Modulated Filter Bank (from type IV DCT)

2 hk[n] = p[n] M cos[(k + )(n +

M 2

1 + 2)

] M

center it!

Double-shift p[n] chosen to be symmetric LPF. orthogonality Only p[n] needs to be designed. in M=2 case Let L be the length of p[n]. Symmetry: P[L 1 n] = p[n] L=2M orthogonality: p[n]2 + p[n + M]2 = 1 L=4M orthogonality: p[n]2+p[n+M]2+p[n+2M]2+p[n+3M]2 = 1 p[n]p[n + 2M] + p[n+M]p[n + 3M] = 0

Genus of the prototype filter.

10

Application to DSL
copper wire assign more bits to lower frequency bands orthogonal CMFB can undo the overlaps between channels f Multiwavelets Idea: extend the scalar refinement equation (t) = 2 h0[k](2t k)
k

into a vector refinement equation 1(t) 1(2t k) = 2 2 (t)

2x2
N-1 k=0

H0[k 2 (2t k)
11

e.g. Finite Elements

1 (t) 1

=
0 1 2 0 1

3
2

+
1 2

2(t)

can use to represent piecewise linear function but allows for representing discontinuous function
1

1(t) = 1(2t) + 1(2t 1)+ 2(2t)

1

(t)
0
1

1(t)

2(t) = 2(2t - 1) + 2(2t) + 1(2t-1)

(t)
(t)
12

1 (t) 2 (t)
=

1 0

1 (2t) 2 (2t)

0 1

1 (2t 1) 2 (2t 1)

Finite Element Multiwavelets

f=1 f=0 f=0 f=0

1(t)

2(t)
f=0 f=1 f=0 f=0

satisfy vector refinement equation

can also come up with orthogonal multiwavelets.

13