Copyright S. K. Mitra
QuadratureMirror Filter Bank
In many applications, a discretetime signal
x[n] is split into a number of subband
signals by means of an analysis filter
bank
The subband signals are then processed
Finally, the processed subband signals are
combined by a synthesis filter bank
resulting in an output signal y[n]
]} [ { n v
k
2
Copyright S. K. Mitra
QuadratureMirror Filter Bank
If the subband signals are
bandlimited to frequency ranges much
smaller than that of the original input signal
x[n], they can be downsampled before
processing
Because of the lower sampling rate, the
processing of the downsampled signals can
be carried out more efficiently
]} [ { n v
k
3
Copyright S. K. Mitra
QuadratureMirror Filter Bank
After processing, these signals are then up
sampled before being combined by the
synthesis filter bank into a higherrate
signal
The combined structure is called a
quadraturemirror filter (QMF) bank
4
Copyright S. K. Mitra
QuadratureMirror Filter Bank
If the downsampling and upsampling
factors are equal to or greater than the
number of bands of the filter bank, then the
output y[n] can be made to retain some or
all of the characteristics of the input signal
x[n] by choosing appropriately the filters in
the structure
5
Copyright S. K. Mitra
QuadratureMirror Filter Bank
If the upsampling and downsampling
factors are equal to the number of bands,
then the structure is called a critically
sampled filter bank
The most common application of this
scheme is in the efficient coding of a signal
x[n]
6
Copyright S. K. Mitra
TwoChannel QMF Bank
Figure below shows the basic twochannel
QMF bankbased subband codec
(coder/decoder)
7
Copyright S. K. Mitra
TwoChannel QMF Bank
The analysis filters and have
typically a lowpass and highpass frequency
responses, respectively, with a cutoff at t/2
) (z H
0
) (z H
1
8
Copyright S. K. Mitra
TwoChannel QMF Bank
Each downsampled subband signal is
encoded by exploiting the special spectral
properties of the signal, such as energy
levels and perceptual importance
It follows from the figure that the sampling
rates of the output y[n] and the input x[n]
are the same
9
Copyright S. K. Mitra
TwoChannel QMF Bank
The analysis and the synthesis filters are
chosen so as to ensure that the reconstructed
output y[n] is a reasonably close replica of
the input x[n]
Moreover, they are also designed to provide
good frequency selectivity in order to
ensure that the sum of the power of the
subband signals is reasonably close to the
input signal power
10
Copyright S. K. Mitra
TwoChannel QMF Bank
In practice, various errors are generated in
this scheme
In addition to the coding error and errors
caused by transmission of the coded signals
through the channel, the QMF bank itself
introduces several errors due to the
sampling rate alterations and imperfect
filters
We ignore the coding and the channel errors
11
Copyright S. K. Mitra
TwoChannel QMF Bank
We investigate only the errors caused by the
sampling rate alterations and their effects on
the performance of the system
To this end, we consider the QMF bank
structure without the coders and the
decoders as shown below
12
Copyright S. K. Mitra
Analysis of the TwoChannel
QMF Bank
Making use of the inputoutput relations of
the downsampler and the upsampler in the
zdomain we arrive at
), ( ) ( ) ( z X z H z V
k k
=
)}, ( ) ( { ) (
/ / 2 1 2 1
2
1
z V z V z U
k k k
+ = k = 0, 1
) ( ) (
2
z U z V
k k
=
^
13
Copyright S. K. Mitra
Analysis of the TwoChannel
QMF Bank
From the first and the last equations we
obtain after some algebra
The reconstructed output of the filter bank
is given by
)} ( ) ( ) ( ) ( { z X z H z X z H
k k
+ =
2
1
)} ( ) ( { ) ( z V z V z V
k k k
+ =
2
1
^
) ( ) ( ) ( ) ( ) ( z V z G z V z G z Y
1 1 0 0
+ =
^ ^
14
Copyright S. K. Mitra
Analysis of the TwoChannel
QMF Bank
From the two equations of the previous
slide we arrive at
The second term in the above equation is
due to the aliasing caused by sampling rate
alteration
) ( )} ( ) ( ) ( ) ( { ) ( z X z G z H z G z H z Y
1 1 0 0
2
1
+ =
) ( )} ( ) ( ) ( ) ( { z X z G z H z G z H + +
1 1 0 0
2
1
15
Copyright S. K. Mitra
Analysis of the TwoChannel
QMF Bank
The inputoutput equation of the filter bank
can be compactly written as
where T(z), called the distortion transfer
function, is given by
and
) ( ) ( ) ( ) ( ) ( z X z A z X z T z Y + =
)} ( ) ( ) ( ) ( { ) ( z G z H z G z H z T
1 1 0 0
2
1
+ =
)} ( ) ( ) ( ) ( { ) ( z G z H z G z H z A
1 1 0 0
2
1
+ =
16
Copyright S. K. Mitra
AliasFree Filter Bank
Since the upsampler and the downsampler
are linear timevarying components, in
general, the 2channel QMF structure is a
linear timevarying system
It can be shown that the 2channel QMF
structure has a period of 2
However, it is possible to choose the
analysis and synthesis filters such that the
aliasing effect is canceled resulting in a
timeinvariant operation
17
Copyright S. K. Mitra
AliasFree Filter Bank
To cancel aliasing we need to ensure that
A(z) = 0, i.e.,
For aliasing cancellation we can choose
This yields
where C(z) is an arbitrary rational function
0
1 1 0 0
= + ) ( ) ( ) ( ) ( z G z H z G z H
) (
) (
) (
) (
z H
z H
z G
z G
=
0
1
1
0
), ( ) ( ) ( ), ( ) ( ) ( z H z C z G z H z C z G = =
0 1 1 0
18
Copyright S. K. Mitra
AliasFree Filter Bank
If the above relations hold, then the QMF
system is timeinvariant with an input
output relation given by
Y(z) = T(z)X(z)
where
On the unit circle, we have
)} ( ) ( ) ( ) ( { ) ( z H z H z H z H z T + =
0 1 1 0
2
1
) (  ) (  ) ( ) ( ) (
) ( e e  e e e e j j j j j j
e X e e T e X e T e Y = =
19
Copyright S. K. Mitra
AliasFree Filter Bank
If T(z) is an allpass function, i.e.,
with then
indicating that the output of the QMF bank
has the same magnitude response as that of
the input (scaled by d) but exhibits phase
distortion
The filter bank is said to be magnitude
preserving
d e T
j
=  ) ( 
e
0 = d
 ) (   ) ( 
e e j j
e X d e Y =
20
Copyright S. K. Mitra
AliasFree Filter Bank
If T(z) has linear phase, i.e.,
then
The filter bank is said to be phase
preserving but exhibits magnitude distortion
 oe e 
e
+ = = ) ( )} ( arg{
j
e T
 oe
e e
+ + = )} ( arg{ )} ( arg{
j j
e X e Y
21
Copyright S. K. Mitra
AliasFree Filter Bank
If an aliasfree filter bank has no magnitude
and phase distortion, then it is called a
perfect reconstruction (PR) QMF bank
In such a case, resulting in
In the timedomain, the inputoutput
relation for all possible inputs is given by
= dz z T ) (
) ( ) ( z X dz z Y
=
] [ ] [ = n x d n y
22
Copyright S. K. Mitra
AliasFree Filter Bank
Thus, for a perfect reconstruction QMF
bank, the output is a scaled, delayed replica
of the input
Example  Consider the system shown below
2 2
2 2
1
z
1
z
+
] [n x
] [n y
23
Copyright S. K. Mitra
AliasFree Filter Bank
Comparing this structure with the general
QMF bank structure we conclude that here
we have
Substituting these values in the expressions
for T(z) and A(z) we get
1 1
1
1
0
1
1 0
= = = =
) ( , ) ( , ) ( , ) ( z G z z G z z H z H
1 1 1
2
1
= + = z z z z T ) ( ) (
0
1 1
2
1
= =
) ( ) ( z z z A
24
Copyright S. K. Mitra
AliasFree Filter Bank
Thus the simple multirate structure is an
aliasfree perfect reconstruction filter bank
However, the filters in the bank do not
provide any frequency selectivity
25
Copyright S. K. Mitra
An AliasFree Realization
A very simple aliasfree 2channel QMF
bank is obtained when
The above condition, in the case of a real
coefficient filter, implies
indicating that if is a lowpass filter,
then is a highpass filter, and vice versa
) ( ) ( z H z H =
0 1
 ) (   ) ( 
) ( e t e
=
j j
e H e H
0 1
) (z H
1
) (z H
0
26
Copyright S. K. Mitra
An AliasFree Realization
The relation
indicates that is a mirrorimage
of with respect to t/2, the
quadrature frequency
This has given rise to the name quadrature
mirror filter bank
 ) (   ) ( 
) ( e t e
=
j j
e H e H
0 1
 ) ( 
e j
e H
0
 ) ( 
e j
e H
1
27
Copyright S. K. Mitra
An AliasFree Realization
Substituting in
with C(z) = 1 we get
The above equations imply that the two
analysis filters and the two synthesis filters
are essentially determined from one transfer
function
) ( ) ( z H z H =
0 1
), ( ) ( ) ( ), ( ) ( ) ( z H z C z G z H z C z G = =
0 1 1 0
) ( ) ( ) ( ), ( ) ( z H z H z G z H z G = = =
0 1 1 1 0
) (z H
0
28
Copyright S. K. Mitra
An AliasFree Realization
Moreover, if is a lowpass filter, then
is also a lowpass filter and is a
highpass filter
The distortion function in this case reduces
to
) (z H
0
) (z G
0
) (z G
1
)} ( ) ( { )} ( ) ( { ) ( z H z H z H z H z T = =
2
0
2
0
2
1 2
1
2
0
2
1
29
Copyright S. K. Mitra
An AliasFree Realization
A computationally efficient realization of
the above QMF bank is obtained by
realizing the analysis and synthesis filters in
polyphase form
Let the 2band Type 1 polyphase
representation of be given by
) (z H
0
) ( ) ( ) (
2
1
1 2
0 0
z E z z E z H
+ =
30
Copyright S. K. Mitra
An AliasFree Realization
Then from the relation it
follows that
Combining the last two equations in a
matrix form we get
) ( ) ( z H z H =
0 1
) ( ) ( ) (
2
1
1 2
0 1
z E z z E z H
=
(
=
(
) (
) (
) (
) (
2
1
1
2
0
0
0
1 1
1 1
z E z
z E
z H
z H
31
Copyright S. K. Mitra
An AliasFree Realization
Likewise, the synthesis filters can be
expressed in a matrix form as
Making use of the last two equations we can
redraw the twochannel QMF bank as
shown below
   
(
=
1 1
1 1 2
0
2
1
1
1 0
) ( ) ( ) ( ) ( z E z E z z G z G
32
Copyright S. K. Mitra
An AliasFree Realization
Making use of the cascade equivalences, the
above structure can be further simplified as
shown below
33
Copyright S. K. Mitra
An AliasFree Realization
Substituting the polyphase representations
of the analysis filters we arrive at the
expression for the distortion function T(z) in
terms of the polyphase components as
) ( ) ( ) (
2
1
2
0
1
2 z E z E z z T
=
34
Copyright S. K. Mitra
An AliasFree Realization
Example  Let
Its polyphase components are
Hence
Likewise
1
0
1
+ = z z H ) (
1 1
2
1
2
0
= = ) ( , ) ( z E z E
1 2
1
1 2
0 0 1
1
= = = z z E z z E z H z H ) ( ) ( ) ( ) (
1 2
0
2
1
1
0
1
+ = + = z z E z E z z G ) ( ) ( ) (
1 2
0
2
1
1
1
1
+ = = z z E z E z z G ) ( ) ( ) (
35
Copyright S. K. Mitra
An AliasFree Realization
The distortion transfer function for this
realization is thus given by
The resulting structure is a perfect
reconstruction QMF bank
1 2
1
2
0
1
2 2
= = z z E z E z z T ) ( ) ( ) (
36
Copyright S. K. Mitra
AliasFree FIR QMF Bank
If in the above aliasfree QMF bank
is a linearphase FIR filter, then its
polyphase components and , are
also linearphase FIR transfer functions
In this case,
exhibits a linearphase characteristic
As a result, the corresponding 2channel
QMF bank has no phase distortion
) (
0
z H
) (
0
z E ) (
1
z E
) ( ) ( 2 ) (
2
1
2
0
1
z E z E z z T
=
37
Copyright S. K. Mitra
AliasFree FIR QMF Bank
However, in general is not a
constant, and as a result, the QMF bank
exhibits magnitude distortion
We next outline a method to minimize the
residual amplitude distortion
Let be a lengthN realcoefficient
linearphase FIR transfer function:
 ) ( 
e j
e T
) (
0
z H
=
=
1
0
0 0
] [ ) (
N
n
n
z n h z H
38
Copyright S. K. Mitra
AliasFree FIR QMF Bank
Note: can either be a Type 1 or a
Type 2 linearphase FIR transfer function
since it has to be a lowpass filter
Then satisfy the condition
In this case we can write
In the above is the amplitude
function, a real function of e
) (
0
z H
] [
0
n h
] [ ] [
0 0
n N h n h =
) ( ) (
0
2 /
0
e =
e e
H e e H
N j j
~
) (
0
e H
~
39
Copyright S. K. Mitra
AliasFree FIR QMF Bank
The frequency response of the distortion
transfer function can now be written as
From the above, it can be seen that if N is
even, then at e = t/2, implying
severe amplitude distortion at the output of
the filter bank
}  ) (  ) 1 (  ) ( {
2
) (
2 ) (
0
2
0
e t e
e
e
=
j N j
jN
j
e H e H
e
e T
0 ) ( =
e j
e T
40
Copyright S. K. Mitra
AliasFree FIR QMF Bank
N must be odd, in which case we have
It follows from the above that the FIR 2
channel QMF bank will be of perfect
reconstruction type if
}  ) (   ) ( {
2
) (
2 ) (
0
2
0
e t e
e
e
+ =
j j
jN
j
e H e H
e
e T
}  ) (   ) ( {
2
2
1
2
0
e e
e
+ =
j j
jN
e H e H
e
1  ) (   ) ( 
2
1
2
0
= +
e e j j
e H e H
41
Copyright S. K. Mitra
AliasFree FIR QMF Bank
Now, the 2channel QMF bank with linear
phase filters has no phase distortion, but
will always exhibit amplitude distortion
unless is a constant for all e
If is a very good lowpass filter with
in the passband and
in the stopband, then is a very good
highpass filter with its passband coinciding
with the stopband of , and viceversa
) (
0
z H
) (
1
z H
1  ) ( 
0
~
e j
e H
0  ) ( 
0
~
e j
e H
) (
0
z H
 ) ( 
e j
e T
42
Copyright S. K. Mitra
AliasFree FIR QMF Bank
As a result, in the passbands
of and
Amplitude distortion occurs primarily
in the transition band of these filters
Degree of distortion determined by the
amount of overlap between their squared
magnitude responses
2 1/  ) (  ~
e j
e T
) (
0
z H ) (
1
z H
43
Copyright S. K. Mitra
AliasFree FIR QMF Bank
This distortion can be minimized by
controlling the overlap, which in turn can be
controlled by appropriately choosing the
passband edge of
One way to minimize the amplitude
distortion is to iteratively adjust the filter
coefficients of on a computer
such that
is satisfied for all values of e
) (
0
z H
] [n h
0
) (
0
z H
1
2
1
2
0
~ +  ) (   ) ( 
e e j j
e H e H
44
Copyright S. K. Mitra
AliasFree FIR QMF Bank
To this end, the objective function  to be
minimized can be chosen as a linear
combination of two functions:
(1) stopband attenuation of , and
(2) sum of squared magnitude responses of
and
) (
0
z H
) (
0
z H ) (z H
1
45
Copyright S. K. Mitra
AliasFree FIR QMF Bank
One such objective function is given by
where
and
and 0 < o < 1, and for some
small c > 0
2 1
1  o o  ) ( + =
e 
t
e
e
d e H
s
j
}
=
2
1
) (
}

.

\

=
t
e e
e 
0
2
2
1
2
0 2
1 d e H e H
j j
) ( ) (
c e
t
+ =
2
s
46
Copyright S. K. Mitra
AliasFree FIR QMF Bank
Since is symmetric with respect to
t/2, the second integral in the objective
function  can be replaced with
After  has been made very small by the
optimization procedure, both and
will also be very small
 ) ( 
e j
e T
}
t
e e
e

.

\

= 
2 /
0
2
2
1
2
0 2
) ( ) ( 1 2 d e H e H
j j
1

2

47
Copyright S. K. Mitra
AliasFree FIR QMF Bank
Using this approach, Johnston has designed
a large class of linearphase FIR filters
meeting a variety of specifications and has
tabulated their impulse response
coefficients
Program 10_9 can be used to verify the
performance of Johnstons filters
48
Copyright S. K. Mitra
AliasFree FIR QMF Bank
Example  The gain responses of the length
12 linearphase FIR lowpass filter 12B and
its powercomplementary highpass filter
obtained using Program 10_9 are shown
below
0 0.2 0.4 0.6 0.8 1
80
60
40
20
0
20
e/t
G
a
i
n
,
d
B
H
0
(z) H
1
(z)
49
Copyright S. K. Mitra
AliasFree FIR QMF Bank
The program then computes the amplitude
distortion in dB as
shown below
0 0.2 0.4 0.6 0.8 1
0.02
0.01
0
0.01
0.02
0.03
e/t
Amplitude distortion in dB
2
1
2
0
 ) (   ) ( 
e e
+
j j
e H e H
50
Copyright S. K. Mitra
AliasFree FIR QMF Bank
From the gain response plot it can be seen
that the stopband edge of the lowpass
filter12B is about 0.71t, which corresponds
to a transition bandwidth of
The minimum stopband attenuation is
approximately 34 dB
s
e
t = t e 105 . 0 2 / ) 5 . 0 (
s
51
Copyright S. K. Mitra
AliasFree FIR QMF Bank
The amplitude distortion function is very
close to 0 dB in both the passbands and the
stopbands of the two filters, with a peak
value of dB 02 . 0
52
Copyright S. K. Mitra
AliasFree IIR QMF Bank
Under the aliasfree conditions of
and the relation , the
distortion function T(z) is given by
If T(z) is an allpass function, then its
magnitude response is a constant, and as a
result its corresponding QMF bank has no
magnitude distortion
) ( ) ( ), ( ) (
0 1 1 0
z H z G z H z G = =
) ( ) (
0 1
z H z H =
) ( ) ( 2 ) (
2
1
2
0
1
z E z E z z T
=
53
Copyright S. K. Mitra
AliasFree IIR QMF Bank
Let the polyphase components and
of be expressed as
with and being stable allpass
functions
Thus,
) (
1
z E
) (
0
z H
) (
0
z E
), ( ) (
0
2
1
0
z z E =
A
) ( ) (
1
2
1
1
z z E =
A
) (
0
z A ) (
1
z A
  ) ( ) ( ) (
2
1
1 2
0
2
1
0
z z z z H A A
+ =
  ) ( ) ( ) (
2
1
1 2
0
2
1
1
z z z z H A A
=
54
Copyright S. K. Mitra
AliasFree IIR QMF Bank
In matrix form, the analysis filters can be
expressed as
The corresponding synthesis filters in
matrix form are given by
(
=
(
) (
) (
1 1
1 1
) (
) (
2
1
1
2
0
2
1
0
0
z z
z
z H
z H
A
A
   
(
=
1 1
1 1
) ( ) ( ) ( ) (
2
0
2
1
1
2
1
1 0
z z z z G z G A A
55
Copyright S. K. Mitra
AliasFree IIR QMF Bank
Thus, the synthesis filters are given by
The realization of the magnitudepreserving
2channel QMF bank is shown below
) ( ) ( ) ( ) (
0
]
2
1
1 2
0
[
2
1
0
z H z z z z G = + =
A A
) ( ) ( ) ( ) (
1
]
2
1
1 2
0
[
2
1
1
z H z z z z G = + =
A A
56
Copyright S. K. Mitra
AliasFree IIR QMF Bank
From
it can be seen that the lowpass transfer
function has a polyphaselike
decomposition, except here the polyphase
components are stable allpass transfer
functions
  ) ( ) ( ) (
2
1
1 2
0
2
1
0
z z z z H A A
+ =
) (z H
0
57
Copyright S. K. Mitra
AliasFree IIR QMF Bank
It has been shown earlier that a bounded
real (BR) transfer function
of odd order, with no common factors
between its numerator and denominator, can
be expressed in the form
if it satisfies the powersymmetry condition
and has a symmetric numerator
) ( / ) ( ) ( z D z P z H
0 0
=
) (z P
0
1
1
0 0
1
0 0
= +
) ( ) ( ) ( ) ( z H z H z H z H
  ) ( ) ( ) (
2
1
1 2
0
2
1
0
z z z z H A A
+ =
58
Copyright S. K. Mitra
AliasFree IIR QMF Bank
It has also been shown that any oddorder
elliptic lowpass halfband filter with a
frequency response specification given by
and satisfying the conditions
and can always be
expressed in the form
) (z H
0
  ) ( ) ( ) (
2
1
1 2
0
2
1
0
z z z z H A A
+ =
,  ) (  1 1 s s
e
o
j
p
e H
p
e e s s 0 for
,  ) ( 
s
j
e H o
e
s t e e s s
s
for
t e e = +
s p
) (
p p s
o o o = 1 4
2
59
Copyright S. K. Mitra
AliasFree IIR QMF Bank
The poles of the elliptic filter satisfying the
two conditions on bandedges and ripples lie
on the imaginary axis
Using the poleinterlacing property
discussed earlier, on can readily identify the
expressions for the two allpass transfer
functions and ) (z
0
A ) (z
1
A
60
Copyright S. K. Mitra
AliasFree IIR QMF Bank
Example  The frequency response
specifications of a realcoefficient lowpass
halfband filter are given by: ,
, and
From we get
In dB, the passband and stopband ripples are
Rp = 0.0010435178 and Rs = 36.193366
t e 4 0. =
p
t e 6 0. =
s
0155 0. =
s
o
) (
p p s
o o o = 1 4
2
00012013 0. =
p
o
61
Copyright S. K. Mitra
AliasFree IIR QMF Bank
Using the Mfile ellipord we determine
the minimum order of the elliptic lowpass
filter to be 5
Next, using the Mfile ellip the transfer
function of the lowpass filter is determined
whose gain response is shown below
0 0.2 0.4 0.6 0.8 1
60
40
20
0
e/t
G
a
i
n
,
d
B
Real halfband filter
62
Copyright S. K. Mitra
AliasFree IIR QMF Bank
The poles obtained using the function
tf2zp are at z = 0, z = ,
and z =
The polezero plot obtained using zplane
is shown below
486625263 0. j
486625263 0. j
1 0 1
1
0.5
0
0.5
1
Real Part
I
m
a
g
i
n
a
r
y
P
a
r
t
63
Copyright S. K. Mitra
AliasFree IIR QMF Bank
Using the poleinterlacing property we
arrive at the transfer functions of the two
allpass filters as given below:
2
2
2
0
2368041466 0 1
2368041466 0
+
+
=
z
z
z
.
.
) ( A
2
2
2
1
7149039978 0 1
7149039978 0
+
+
=
z
z
z
.
.
) ( A