NETWORK STRUCTURES FOR FIR SYSTEMS AND PARAMETER QUANTIZATION EFFECTS

IN DIGITAL FILTER STRUCTURES

Solution 13.1
Since H(z) has only real zeros we will use only first-order
sections in the cascade form. Then Figure S13.1-1 represents one
(i) possible ordering for these sections.
y(n)

x(n)

Figure S13.l-1

(ii)
H(z)
=

For the direct form we first express H(z) as

1

7 -l 4

69

-2
8

zZ- TZ

-3 + z-4

The direct-form structure is then as shown in Figure S13.1-2

x(n)

y(n)

Figure 13.1-2

S13.1

Since the unit-sample response is symmetrical the filter

is in fact linear phase. The linear phase form is as shown in
Figure S13.1-3.
(iii)

x(n)

69

y(n)

Figure S13.l-3
(iv) The coefficients in the frequency-sampling structure are the
To

samples of the frequency response equally spaced in frequency. in the form

evaluate these frequency samples it is convenient to rewrite H(z)

H(z)

z-

or
(z+z 1 ) -

H(z) = z

(z +z)

6]

Then

f(0)

H(z) Iz1

81

4r

H(1) = H (z)

=e z=e 25

(2cos

cos 2

S13.2

. 87
H(2) = H(z) z=e
3

= e

L 2 cos O

cos

69

8J

Because of the conjugate symmetry of the Fourier transform,

j87T

H(3)

= H (5-3)

= H (2)

= e

O87T cos 52 7 cos 4 r -

691

.47r

H(4)

= H (5-4)

= H (1)

= e

42 L cosLT 5

7
-

2r

Cos 5

69
8

The frequency sampling structure, in the form of chalkboard (b) lecture 13, is
then as shown below:
H(0)
-l
z
H(1)
-j2r eZ
x(n) H 1/5

y(n)

-j 87r
5

z1

Figure S13.1-4

In this form the sections involve

complex coefficients.

By utilizing

the conjugate symmetry of the frequency samples the network can be

rearranged in terms of second-order sections with real coefficients.

S13.3

Specifically the transfer function corresponding to the recursive

part of the network of Figure S13.1-4 is given by

G(Z) = H(0)-l1 +
1-z~ l-e

H(1) .2rT
-j3z

+
l1
+

H(4)
-1 l-e z

J7

H(3) 1-e 5

The terms paired in brackets are complex conjugates, i.e.,

*i

~f
= e e

87T

H(l)

H (4) and e

.27r
= e

H(2)

.47
.67 J-a
-J-g-= e. H (3) and e

Continuing these complex conjugate terms, G(z) can be expressed

as

2Tr
2Re [H (1) ]-2z~ 1 Re [R (1) e J 5
G (Z)
= 1-z H (0)

1-2z -ZCOS 2os+ 1 1 5

-2

47T
2Re[H(2)]-2z

1Re [H(2.)eJ 5

1-2z~1c 4+r + z -2 lz Cos- 5

leading to the structure shown in Figure S13.1-5.

S13.4

H (0 )

x (n)

Re H (1)

l/5

y(n)

-z

z~

-2

Re IH(1)

e5

2 cos
2 Re H(2) Jw5

-2
z

Re e (2) e

Oos 4

cr

Figure S13.1-5

Solution 13.2

(a)

The form of the desired transfer function is

in the form

easily obtained by

expressing h(n)

h (n)

[e

e jw0nu(n)

+ e

je

jOn

u (n) ]

Thus

e $

- j$

cos$--~1cosw (0

H(z)

1 l-Ve2z-

S13.5

Thus,
a1 =

(b) is

The form of the transfer function for the coupled form network (see Problem 11.3)

H2(z)

= b z~1/[i-2a 2 Z

+ (a2 +b 2 )z2

Thus to obtain a unit-sample response of the desired form, A cos$ = 0 and A cos(w 0 -q)=-l. The coefficients a2 and b2 are given by

a2 = cosw 0

= sin0

~7

(c) For the direct form filter since the coefficient b is unity, it
is not affected by quantization. For the coefficient a1 , the
quantized value is

a1

= 1.375

The resulting transfer function is

H1 (z) = 1
1
1li +z 2 z 1+z-2z

Since this is still in the form of the desired H(z), the unit-sample
response will still be of the form of Eq. (13.2-1). With the
quantized coefficient, 2 cosw 0 is now equal to 11/8 so that

0 =

.26w

For the coupled form, the quantized coefficients are given by

S13.6

^ ^ a2 = b2

5 = T = 0.625

Thus H 2 (z) with quantized coefficients becomes

H2 (z) =

(5/8)z~ 1

5 -

25 -2

In comparing this with the desired H(z) we note that since the co
efficient of z-2 is not unity, the resulting unit-sample response will
not be of the desired form. In particular, the unit-sample response
will be of the form of a damped sinusoidal sequence, corresponding
to the fact that the poles of the coupled form system with quantized
coefficients have moved inside the unit circle. In contrast, the
poles of the direct form system with quantized coefficients remain
on the unit circle but are displaced in angle. These results are of
course consistent with the differences in the quantization grids for
the two stru'tures, as illustrated in Figures 6.49 and 6.51 of the text.

Solution 13.3

N-1

(a) H(k) = h(n) Wnk

n=O
N
=

h(n) Wnk + h(N-1-n) W N-n)k

n=O

n=O

which, because of the symmetry constraint becomes

N _

H(k) -

h(n)

_W + W;N-n)k

n=O
For k = N

S13.7

-1

h (n)
n=o

N -1 h (n)

[e-jTrn

+ e-jT

(N-l-n)

[e-

jTn _ ejTrn
0

n=0

-0

N (b) H (k) h (n) [Wnk + W = N 1-n)k

n=0

N-1

h (n) WN 2 k
n=0
N
(Nl ) k
=W

[Wk(n- N-) + W-k(nN N

2f'r
2h (n) cos [2k

~
N2 (nN1

Nn=0

)knd

The summation in the above expression is real. Let us assume for

convenience that it is also positive. If it is not, then for those values of k for which it is negative, an additional phase of 7 will be added. Then, with this assumption, N-1
2 WN

e j(k) e=
or
6 (k)
= -

T(N-l)k

NT

7rk +

S13.8

cos(Rk

7k) -

z 1 cos -k

Tk)

(c)

Hk (z) = 1-2z cos(-) N

+ Z

(-1)kcosr (1-z~1 1-2z~ cos (-) N

Thus, from Equation + z-2

(4.49) of the text.

k
H(z) = 1 -N
k=l

I (k) 12 cos( k/N) (1-z -1 )

H(0)

1-2z -1cos (2Tk/N) + z-2

+ 1-z] _i

(d)
H (0)

H (z) 21(1) 1
1/N H 2 (z)
21H 2

z1

21H(2) -N H 3 (Z) 2 1H(3)

I
0 0 -~

HN (z) 2-1 Figure S13.3-1

2|

(N 1)

S13.9

where the subnetworks Hk (z) are of the form

Cos

2 -cos 27rk -1

-1r 1

Figure S13.3-2

S13.10

MIT OpenCourseWare http://ocw.mit.edu

Resource: Digital Signal Processing

Prof. Alan V. Oppenheim

The following may not correspond to a particular course on MIT OpenCourseWare, but has been provided by the author as an individual learning resource.

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.