Wavelets and Filter Banks

Inheung Chon

Department of Mathematics
Seoul Woman’s University
Seoul 139-774, Korea

Abstract

We show that if an even length filter has the same length complementary
filter in a generalized linear phase case, the complementary filter is unique,
we find sufficient conditions for a unique existence of even length N com-
plementary filter in a quadrature mirror filter bank, and we find all higher
degree symmetric filters of length N + 4m which are complementary to a
given symmetric filter of even length N.

1 Introduction
Wavelet theory has attracted much attention from many researchers in mathemat-
ics and electrical engineering fields. The integral wavelet transform and wavelet
series, which are main topics in wavelet theory, are performed using wavelets.
Thus constructing proper wavelets for applications is considered to be very im-
portant. Filter banks are efficient convolution structures that have been used in
subband coders for speech. In a filter bank, a data sequence x(n) is decomposed
into M channels by convolution with sequences h i (n), called the analysis filters,
down-sampled by M on each channel, upsampled by M on each channel, and then
convolved with the sequences gi (n), called the synthesis filters, and recombined to
give x̂(n). An important problem in a filter bank theory is how to design perfect
reconstruction filter banks.
1991 Mathematics Subject Classification. 94A12.
Key words and phrases. quadrature mirror filter bank, perfect reconstruction, valid polynomial,
complementary filter.
This paper was supported (in part) by the research fund of Seoul Women’s Unuversity.

55
56 I NHEUNG C HON

Daubechies ([4]) constructed compactly supported orthonormal wavelets and

orthonormal scaling functions based on iterations of discrete filters and Vetterli
([7]) showed that biorthogonal perfect reconstruction filter banks generate biorthog-
onal wavelets and biorthogonal scaling functions. In order to generate wavelets
using perfect reconstruction filter banks, it is required to find high pass comple-
mentary filter with respect to a given lowpass filter.

In section 2, we review some basics of scaling function, wavelets, and filter

banks and we prove that a Laurent series under some assumptions generates or-
thonormal scaling function φ and orthonormal wavelet ψ and generated φ and ψ
lead to perfect reconstruction filter bank, which is a variant of Vetterli’s work ([7]).
In section 3, we show that if an even length filter has the same length complemen-
tary filter in a generalized linear phase case, the complementary filter is unique, we
find sufficient conditions for a unique existence of even length N complementary
filter in a quadrature mirror filter bank, and we find all higher degree symmetric
filters of length N + 4m which are complementary to a given symmetric filter of
even length N .

2 Wavelets and Filter Banks

R
Definition 2.1. A function ψ ∈ L 2 (R) is called an − f unction if {ψ j,k :=
22/j ψ(2 j x − k)} is a Riesz basis of L 2 (R) in the sense that the linear span of ψ j,k
is dense in L 2 (R) and positive constants A and B exist, with 0 < A ≤ B < ∞,
such that
∞
X ∞
X
A k {c j,k } kl22 ≤k c j,k ψ j,k k22 ≤ B k {c j,k } kl22
j =−∞ k=−∞

for all doubly bi-infinite square-summable sequences {c j,k } ; that is,

∞
X ∞
X
k {c j,k } kl22 := |c j,k | < ∞.
j =−∞ k=−∞

Definition 2.2. A function φ ∈ L 2 (R) is called a scaling fuction if the subspaces

V j of L 2 (R) defined by

V j := hφ j,k : k ∈ Zi, j ∈ Z,

satisfy
Wavelets and Filter Banks 57

(1) · · · ⊂ V−1 ⊂ V0 ⊂ V1 ⊂ . . .
S
j ∈Z = L (R)
2
(2)

(3) f (x) ∈ V j if and only if f (2x) ∈ V j +1 , j ∈ Z

(4) f (x) ∈ V j if and only if f (x + 1

2j
) ∈ Vj , j ∈ Z

and if {φ(· − k) : k ∈ Z} is a Riesz basis of V0 . We say that the scaling function

φ generates a multiresolution analysis {V j } of L 2 (R). Hereafter multiresolution
analysis is abbreviated by MRA.

It is well known (see [3]) that if φ ∈ L 2 (R) is a scaling function that generates
an MRA {V j } of L 2 (R), then ∩ j ∈Z V j = {0}.

Definition 2.3. If a scaling function Pφ generates an MRA, then there exists a

unique sequence { pk } such that φ(x) = ∞k=−∞ pk φ(2x −k). We call the sequence
{ pk } two scale sequence of φ. Corresponding to this l 2 sequence, we define two
scale symbol of φ by
∞
1 X
P(z) = Pφ (z) := pk z k .
2 k=−∞

Definition 2.4. A function ψ ∈ L 2 (R) is called an orthogonal wavelet if the

family
{ψ j,k (x) := 2 j/2 ψ(2 j x − k) : j, k ∈ Z}
is an orthonormal basis of L 2 (R); that is,

hψ j,k , ψl,m i = δ j,l δk,m , j, k, l, m ∈ Z,

and every f ∈ L 2 (R) can be written as

∞
X
f (x) = c j,k ψ j,k (x),
j,k=−∞

where the convergence of the series is in L 2 (R).

Definition 2.5. A collection of filters with a common input or a common output

is called a filter bank. The filter bank which splits a signal x(n) into M signals
x0 (n), x1 (n), . . . , x M−1 (n) is called an analysis bank and the filter bank which col-
lects M signals y0 (n), y1 (n), . . . , y M−1 (n) into a signal x̂(n) is called a synthesis
58 I NHEUNG C HON

bank. The splitted signals x0 (n), x1 (n), . . . , x M−1 (n) are called subband signals.

Definition 2.6. The device which takes an input sequene x(n) and produces the
output sequence y D (n) = x(Mn) with a positive integer M is called an M-fold
decimator, a downsampler, or a subsampler. The device which takes an input x(n)
and produces an output sequence
(
x( Ln ), n is integer multiple of positive integer L
y E (n) =
0, otherwise.

is called an L-fold expander, an upsampler, or a intepolator.

We review the following basic proposition which will be used for Theorem 3.4.

Proposition 2.7. If a filter bank has H0 (z) and H1 (z) as its analysis filters with
downsampling by two and has F0 (z) and F1 (z) as its synthesis filters with upsam-
pling by two, then the output X̂ (z) of the filter bank with its input X (z) is given
by
X (z)[H0 (z)F0 (z) + H1 (z)F1 (z)] + X (−z)[H0 (−z)F0 (z) + H1 (−z)F1 (z)]
X̂ (z) = .
2
Particularly, if F0 (z) = H1 (−z) and F1 (z) = −H0 (−z), the output X̂ (z) is given
by
X (z)[H0 (z)F0 (z) + H1 (z)F1 (z)] X (z)[H0 (z)H1 (−z) − H0 (−z)H1 (z)]
X̂ (z) = = .
2 2
Proof. Let a(n) be the input and b(n) be the output of the upsampler by two.
Then
∞
X ∞
X ∞
X
B(z) = b(n)z −n = b(2m)z −2m = a(m)z −2m = A(z 2 ).
n=−∞ m=−∞ m=−∞

Let c(n) be the input and d(n) be the output of the downsampler by two. Then
∞
X ∞
X
D(z) = d(n)z −n = c(2n)z −n .
n=−∞ m=−∞

∞ ∞ ∞
1 1 1 1 X − n2
X
−n − n2
X
(C(z +C(−z )) = (
2 2 c(n)z + c(n)(−1) z ) = c(2n)z −n .
2 2 n=−∞ n=−∞ n=−∞

Thus
1 1 1
D(z) = (C(z 2 ) + C(−z 2 )).
2
Wavelets and Filter Banks 59

The output of H0 (z) and downsampler is

1
[H0 (z 0.5 )X (z 0.5 ) + H0 (−z 0.5 )X (−z 0.5 )].
2
The output of H1 (z) and downsampler is

1
[H1 (z 0.5 )X (z 0.5 ) + H1 (−z 0.5 )X (−z 0.5 )].
2
The output of upsampler and F0 (z) is

1
[H0 (z)X (z)F0 (z) + H0 (−z)X (−z)F0 (z)].
2
The output of upsampler and F1 (z) is

1
[H1 (z)X (z)F1 (z) + H1 (−z)X (−z)F1 (z)].
2
Thus the proposition is proved. 

Definition 2.8. Suppose the filter bank has X (z) as its input and has X̂ (z) as its
output. If X̂(z) = kz −d X (z) for some constant k and d ∈ Z, we say the filter bank
has a perfect reconstruction –hereafter denoted by PR– property. A polynomial
P(z) which satisfies

P(z) − P(−z) = kz −2l−1 for some constatant k and l ∈ Z

is called a valid polynomial. Given a filter H0 (z), any filter H1 (z) such that P(z) =
H0 (z)H1 (−z) is a valid polynomial is called a complementary filter. A Laurent
series is said to belong to the Wiener Class W if its coefficients sequence is in l 1 .
P L−1 P L−1
Theorem 2.9. Let H0 (z) = n=0 h 0 (n)z −n and H1 (z) = n=0 h 1 (n)z −n be fil-
ters such that hh 0 (n −2k), h 0 (n −2l)i = δkl and h 1 (n) = (−1)n h 0 (L −1−n). Then
the filter bank which has H0 (z) and H1 (z) as its analysis filters
P L−1 with downsampling
P L−1
two and has G 0 (z) = n=0 h 0 (L − 1 − n) and G 1 (z) = n=0 h 1 (L − 1 − n) as
its synthesis filters with upsampling two has a perfect reconstruction property.

Proof. See [7]. 

Daubechies and other researchers (see [4], [5], and [7]) characterized the rela-
tionships between wavelets and filter banks. The followng theorem is a variant of
Vetterli’s work (see [7]).
60 I NHEUNG C HON

Theorem 2.10. Let P ∈ W be a Laurent series such that

1X 1+z N
P(z) = pk z k = ( ) S(z),
2 k 2
where N is some positive integer and S ∈
P Wsatisfies S(1) = 1. P Let S(z) =
α
s
k k z k
and B := max |z|=1 |S(z)|. If |P(z)|2
+|P(−z)| 2
= 1, |z| = 1, k |sk ||k| <
∞ for some α > 0, and B < 2 N−1 , then the scaling function φ and wavelet ψ
obtained from P(z) generates perfect reconstruction finite impulse response filter
bank.
P
Proof. Let φ(x) := k∈Z pk φ(2x − k). Then φ is an orthonormal P scaling
function that generates an MRA of L (R) (see [3]). Let ψ(x) := k∈Z (−1) p−k+1 φ(2x−
2 k

k). Then
hφ j,k , φ j,l i = δk,l , j, k, l ∈ Z,
hφ j,k , ψ j,l i = 0, j, k, l ∈ Z,
hψ j,k , ψl,m i = δ j,l δk,m , j, k, l, m ∈ Z.
That is, ψ is an orthonormal wavelet. Hence
Z ∞ Z
1 ∞ 1
hφ(2x−l), φ(2x−k)i = φ(2x−l)φ(2x−k)dx = φ(y−l)φ(y−k)dy = δk,l .
−∞ 2 −∞ 2
X X 1X
hφ(x+k), φ(x+l)i = h pn φ(2x+2k−n), pm φ(2x+2l−m)i = pn+2k pn+2l .
n m
2 n
P
Since hφ(x + k), φ(x + l)i = δkl , 12 n pn+2k pn+2l = δkl . Let H0 (z) be a finite
length filter such that h 0 (n) = √pn2 . Then
pn+2k pn+2l 1X
hh 0 (n + 2k), h 0 (n + 2l)i = h √ , √ i = pn+2k pn+2l = δkl .
2 2 2 n
Thus H0 (z) is an orthogonal filter.
(−1)n p L−1−n
Let H1 (z) be a finite length filter such that h 1 (n) = √
2
. Then
p L−1−n−2k p L−1−n−2l 1X
hh 1 (n+2k), h 1 (n+2l)i = h √ , √ i= p L+1−n−2k p L+1−n−2l = δkl .
2 2 2 n
Thus H1 (z) is an orthogonal
P L−1 filter. P L−1
Let G 0 (z) = n=0 h 0 (L − 1 − n) and G 1 (z) = n=0 h 1 (L − 1 − n). By
Theorem 2.9, the filter bank which has H0 (z) and H1 (z) as its analysis filters with
downsampling two and has G 0 (z) and G 1 (z) as its synthesis filters with upsam-
pling two has a perfect reconstruction property. 
Wavelets and Filter Banks 61

3 Complementary Filters
In section 2, we prove that a Laurent series under some assumptions generates
orthonormal scaling function φ and orthonormal wavelet ψ and generated φ and ψ
lead to perfect reconstruction filter bank. In this section, we consider a generation
of wavelets using perfect reconstruction filter banks as the reverse problem of sec-
tion 2. Since it is required to find high pass complementary filter with respect to a
given lowpass filter for a generation of wavelets, we find sufficient conditions for
a unique existence of even length N complementary filter in a quadrature mirror
filter bank and we find all higher degree symmetric filters of length N + 4m which
are complementary to a given symmetric filter of even length N .

Definition 3.1. The filters H0 (z) and H1 (z) which satisfy H1 (z) = H0 (−z), that
is, |H1 (eiω )| = |H0 (ei(π−ω) )|, are called quadrature mirror filters and are abbrevi-
ated by QMFs. That is, |H1 (eiω )| is a mirror image of |H0 (eiω )| with respect to the
quadrature frequency 2π 4
. The filter bank which has H0 (z) and H1 (z) = H0 (−z)
as its analysis filters with downsampling by two and F0 (z) = H0 (z) and F1 (z) =
−H1 (z) as its synthesis filters with upsampling by two is called a quadrature mir-
ror filter bank and is abbreviated by QMFB.

Theorem 3.2.

(1) A real valued sequence {an } ∈ l 1 has generalized linear phase iff it is sym-
metric or antisymmetric (with respect to the phase of the symbol of {an }).

(2) A real valued finite sequenceP{an } ∈ l 1 with support [0, N ] has linear phase
N
iff a N−n = an and A(z) = n=0 an z n has only zeros of even order on the
unit circle.

Proof. See [3]. 

Theorem 3.3. In the generalized linear phase case, if a filter H0 (z) of an even
length N has a complementary filter H1 (z) of length N , then H1 (z) is unique.

Proof. By Theorem 3.2, we consider the following four cases.

(i) H0 (z) and H1 (z) are symmetric
P(z) = H0 (z)H1 (z) = [h 0 (0)h 1 (0)] + [h 0 (1)h 1 (0) − h 0 (0)h 1 (1)]z −1 + · · · +
[h 0 (0)h 1 (1) − h 0 (1)h 1 (0)]z −(2N−3) − h 0 (0)h 1 (0)z −(2N−2) . Thus P(z) is antisym-
metric except the central term. P(z) − P(−z) = 2[h 0 (1)h 1 (0) − h 0 (0)h 1 (1)]z −1 +
· · · + 2[h 1 (1)h 0 (0) − h 0 (1)h 1 (0)]z −(2N−3) = 2z −2l−1 . Hence P(z) − P(−z) is an-
tisymmetric except the central term. We have the following system of equations
62 I NHEUNG C HON

where h 0 (0), h 0 (1), . . . h 0 (N − 1) are given and h 1 (0), h 1 (1), . . . h 1 (N − 1) are

unknowns.

h 0 (1)h 1 (0) − h 0 (0)h 1 (1) = 0

h 0 (2)h 1 (0) + h 0 (0)h 1 (2) − h 0 (1)h 1 (1) = 0
...
−h 0 (2)h 1 (0) − h 0 (0)h 1 (2) + h 0 (1)h 1 (1) = 0
h 1 (1)h 0 (0) − h 0 (1)h 1 (0) = 0

This system consists of N equations but actually consists of N2 equations by anti-

symmetry. By symmetry of H1 (z), the number of unknowns is N2 . Thus the system
consisting of N2 equations has N2 unknowns. Hence the system has a unique solu-
tion if the solution exists.
(ii) H0 (z) is symmetric and H1 (z) is antisymmetric
P(z) = H0 (z)H1(−z) = [h 0 (0)h 1 (0)] + [h 0 (1)h 1 (0) − h 0 (0)h 1 (1)]z −1 + · · · +
[−h 0 (0)h 1 (1)
+h 0 (1)h 1 (0)]z −(2N−3) + h 0 (0)h 1 (0)z −(2N−2) . Thus P(z) is symmetric except the
central term. P(z)−P(−z) = 2[h 0 (1)h 1 (0)−h 0 (0)h 1 (1)]z −1 +· · ·+2[−h 1 (1)h 0 (0)+
h 0 (1)h 1 (0)]z −(2N−3) = 2z −2l−1 . Hence P(z)− P(−z) is symmetric except the cen-
tral term. We have the following system of equations where h 0 (0), h 0 (1), . . . h 0 (N −
1) are given and h 1 (0), h 1 (1), . . . h 1 (N − 1) are unknowns.

h 0 (1)h 1 (0) − h 0 (0)h 1 (1) = 0

h 0 (2)h 1 (0) + h 0 (0)h 1 (2) − h 0 (1)h 1 (1) = 0
...
h 0 (2)h 1 (0) + h 0 (0)h 1 (2) − h 0 (1)h 1 (1) = 0
−h 1 (1)h 0 (0) + h 0 (1)h 1 (0) = 0

This system consists of N equations but actually consists of N2 equations by sym-

metry. By antisymmetry of H1 (z), the number of unknowns is N2 . Thus the system
consisting of N2 equations has N2 unknowns. Hence the system has a unique solu-
tion if the solution exists.
(iii) H0 (z) and H1 (z) are antisymmetric
The proof is similar to that of the case (i)
(iv) H0 (z) is antisymmetric and H1 (z) is symmetric
The proof is similar to that of the case (ii) 

Theorem 3.4. If a symmetric filter H0 (z) in QMFB has an even length N and
Wavelets and Filter Banks 63

the greatest common divisor of H0 (z) and H0 (−z) divides 2z −2l−1 for some l ∈ Z,
then H0 (−z) is a unique complementary filter of length N with respect to the filter
H0 (z).

Proof. From Proposition 2.7, X̂ (z) = X (z)[H0 (z)H1(−z)−H 2

0 (−z)H1 (z)]
. Since H1 (z)
X (z)[H0 (z)H0 (z)−H0 (−z)H0 (−z)]
= H0 (−z) in QMF, X̂ (z) = 2
. It is well known (see [1])
that given a(x) and b(x),

a(x) p(x) + b(x)q(x) = c(x)

has a solution p(x) and q(x) iff the greatest common divisor of a(x) and b(x)
divides c(x). Since QMFB has a PR property iff X̂ (z) = kz −2l−1 X (z) for some
constant k and some l ∈ Z, the QMFB has a PR property iff the greatest common
divisor of H0 (z) and H0 (−z) divides 2z −2l−1 . Thus H0 (−z) is a complementary
filter of H0 (z). Since H0 (z) is symmetric and N is even, H0 (−z) is antisymmetric.
By (ii) in the proof of Theorem 3.3, H0 (−z) is a unique complementary filter. 

Theorem 3.5. Given a symmetric filter H0 (z) of even length N and its comple-
mentary filter H1 (z) of symmetric form and length N , all higher degree symmetric
filters of length N + 4m which are complementary to H0 (z) are of the form

G(z) = z −2m H1 (z) + E(z)H0(z),

where E(z) = [a1 + a2 z −2 + · · · + am z −(2m−2) + am+1 z −2m + am z −(2m+2) + · · · +

a2 z −(4m−2) + a1 z −4m ].

Proof. E(z)H0 (z) = a1 h 0 (0) + a1 h 0 (1)z −1 + [a1 h 0 (2) + a2 h 0 (0)]z −2 + · · · +

a1 h 0 (N − 2)z −(N+4m−2) + a1 h 0 (N − 1)z −(N+4m−1) . Thus E(z)H0 (z) is symmetric
about the point between N+4m 2
and N+4m−2
2
. H1 (z)z −2m = h 1 (0)z −2m +h 1 (1)z −(2m+1)+
−(2m+2) −(N+2m−1)
h 1 (2)z + · · · + h 1 (N − 1)z . Thus H1 (z)z −2m is symmetric about
the point between 2 andN+4m N+4m−2
2
. Hence H1 (z)z −2m + E(z)H0 (z) is symmet-
ric. Since H1 (z) is a complementary filter to H0 (z), z −2m H1 (z) is a complementary
filter to H0 (z). It is easily checked from E(z) = E(−z) that given H0 (z), the
solution of the equation

H0 (z)H1 (−z) − H0 (−z)H1(z) = 0

is H1 (z) = E(z)H0(z). Thus z −2m H1 (z) + E(z)H0 (z) is a complementary filter of

length N + 4m to H0 (z).
Suppose G(z) is a complementary filter of length N + 4m to H0 (z). Then
H0 (z)G
(−z) is valid and of length 2N +4m−1. Since H0 (z)E(−z)H0(−z)−H0(−z)E(z)H0(z)
64 I NHEUNG C HON

= 0, H0 (z)E(z)H0(−z) is valid and of length 2N +4m−1. Let K (z) = H0 (z)[G(−z)−

E(z)H0 (−z)]. Then K (z) is valid and of length 2N +4m−1. K (z) = h 0 (0)(g(0)−
a1 h 0 (0)) + [h 0 (1)(g(0) − a1 h 0 (0)) + h 0 (0)(a1 h 0 (1) − g(1))]z −1 + [h 0 (2)(g(0) −
a1 h 0 (0)) + h 0 (0)(g(2) − a2 h 0 (0) − a1 h 0 (2)) + h 0 (1)(a1 h 0 (1) − g(1))]z −2 + · · · +
[h 0 (N − 1)(a1 h 0 (N − 1) − g(N + 4m − 1))]z −(2N+4m−2) . By choosing a1 = hg(0) 0 (0)
,
−(2N+4m−2)
the coefficients of z and z0
are zero. Since H0 (z)G(−z) is valid and
−1 −(2N+4m−3)
a1 = hg(0)
0 (0)
, the coefficients of z and z are zero. Similarly, by choosing
proper ai for i = 2, . . . , m, we can make the coefficients of four terms 0 for each
ai . By this process, K (z) has length 2N − 1, has powers of z −1 in the range z −2m to
z −2m+2N−2 , and is valid. Since K (z) has H0 (z) as a factor and has z −2m as a factor,

K (z) = z −2m H0 (z)Q(z), where length of Q(z) is N.

By Proposition 3.3, the solution of length N is unique. Thus K (z) = z −2m H0 (z)H1 (−z).
Hence z −2m H0 (z)H1 (−z) = H0 (z)(G(−z) − E(z)H0(−z)). That is, G(−z) =
z −2m H1
(−z) + E(z)H0(−z). Thus G(z) = z −2m H1 (z) + E(z)H0(z). 

References
[1] Burton, D. M., Elementary Number Theory, Allyn and Bacon Inc., 1980.
[2] Chan, Y. T., Wavelet Basics, Kluwer, 1995.

[3] Chui, C. K., An Introduction to Wavelets, Academic Press, 1978.

[4] Daubechies, I., Orthonormal Basis of Compactly Supported Wavelets, Comm. in Pure and
Applied Math. 41 (1988), 909-996.

[5] Lee, E. H., Wavelets and Signal Processing, Master’s thesis at Seoul Women’s University,
1996.

[6] Vaidyanathan, P. P., Multirate Systems and Filter Banks, Prentice Hall, 1993.

[7] Vetterli, M. and C. Herly, Wavelets and Filter Banks: Theory and Design, IEEE Trans. Signal
Processing 40, 1057-1071.