Review

LCCDE
The Discrete-Time Fourier Transform (DTFT) of a signal x(n) is
defined by:

X
j
x(n)e jn
X (e ) =
n=

CHUM Pharino

and the Inverse DTFT is given by

Z
1
x(n) =
X (e j )e jn d
2

Institute of Technology of Cambodia

Phnom Penh, Cambodia
[email protected]

This is decomposition of x(n) as linear combination of complex

exponential (sinusoid).

November 6, 2016
This material is property of author and is for sole and exclusive use of his students. It is not for publication, nor it is to be sold,
reproduced, or generally distributed.

The author (ITC)

lecture #06

November 6, 2016

1 / 23

Application of DTFT I

The author (ITC)

lecture #06

November 6, 2016

2 / 23

Application of DTFT II
Inverse System: Using Fourier Transforms, it is easy to find inverse
system for signal reconstruction problems, when they exist.
Example: The frequency response if a system is

Performing convolution: The Fourier transform provides a useful

tool to aid in performing convolutions.
Example: The unit sample response of a system

H(e e ) =

1 + 0.25e j
1 0.5e j

Find an input to the system that will produce the output

h(n) = an u(n)

y (n) = (n)

Find the system response to input x(n) = b n u(n).

Solution:


y (n) = F 1 H(e j )X (e j )

Solution:
G (e j ) = H 1 (e j ) =

1 0.5e j
1 + 0.25e j

There are some serious practical issues with this solution. Can you
think of some?
The author (ITC)

lecture #06

November 6, 2016

3 / 23

The author (ITC)

lecture #06

November 6, 2016

4 / 23

Application of DTFT III

Application of DTFT IV
Rewriting, we have
(
)
(M
)
N
X
X
j
j jk
j
j jk
Y (e ) 1
a(k)Y (e )e
= X (e )
b(k)X (e )e

Solving difference equations

y (n) =

N
X

a(k)y (n k) +

k=1

M
X

k=1

b(k)x(n k)
y (n) =

k=0

k=0
N
X

a(k)y (n k) +

k=1

b(k)x(n k)

k0

Solution

Recall the Shift Properties:



y (n) = F 1 Y (e j )

If x(n) X (e j ) then x(n n0 ) e jn0 X (e j )

where
Y (e j ) =

using this properties in the LCCDE:

Y (e j ) =

M
X

N
X

a(k)Y (e j )e jk +

k=1

M
X

PN
jk
k=0 b(k)e
P
X (e j )
n
1 k=1 a(k)e jk

Note: This solution assumes that there are no initial conditions.

Example: Solve the following LCCDE assuming zeros initial rest
conditions

b(k)X (e j )e jk

k=0

y (n) 0.25y (n 2) = x(n) 0.5x(n 1)

The author (ITC)

lecture #06

November 6, 2016

5 / 23

LSI Systems and LCCDEs

N
X

lecture #06

November 6, 2016

6 / 23

Example

An important subclass of LSI systems are those whose input, x(n),

and output, y (n) are related by an N th -order LCCDE
y (n) =

The author (ITC)

a(k)y (n k) +

k=1

M
X

Consider the linear shift-invariant system characterized by the

second-order linear constant coefficient different equation

b(k)x(n k)

y (n) = 1.3433y (n1)0.9025y (n2)+x(n)1.4142x(n1)+x(n2)

k0

The frequency response is (again, be careful of signs in the

denominator)

The frequency response may be written down by inspection.

PM
jk
Y (e j )
k=0 b(k)e
H(e ) =
=
P
jk
X (e j )
1 N
k=1 a(k)e
j

H(e j ) =

1 1.4142e j + e 2j
1 1.3433e j + 0.9025e 2j

Note: H((e j ) is a rational function of e j

Note: Be careful of the signs of the coefficients in the denominator.

The author (ITC)

lecture #06

November 6, 2016

7 / 23

The author (ITC)

lecture #06

November 6, 2016

8 / 23

More on LCCDE

FIR Filters

General difference equation:

y (n) =

N
X

a(k)y (n k) +

k=1

M
X

y (n) =
b(k)x(n k)

This is called an FIR (Finite Length Impulse Response) filter.

Special case: All of the coefficients a(k) are equal to zero.

h(n) =
y (n) =

M
X

b(k)x(n k)

b(k)(n k)

H(e j ) =

k=0

M
X

b(1)

6
5 b(0)

h(n) = b(n)

b(5)
b(2)
b(4)

4
3
b(3)

b(k)e jk

k=0
-2

The author (ITC)

h(n)

The unit sample response is Finite-in-Length and the frequency

response is a polynomial in e j
M
X

h(k)(n k)

k=0

k=0

h(n) =

b(k)x(n k)

k=0

k=0

M
X

M
X

lecture #06

November 6, 2016

9 / 23

Infinite-Length Unit Sample Response Filters

-1

The author (ITC)

10

lecture #06

November 6, 2016

10 / 23

More on LCCDE

Computation Algorithm

When one or more coefficient a(k) are non-zero, there is feed back
and the unit sample response is infinite in length.

y (n) =

Example:

z
N
X

}|
a(k)y (n k) +

k=1

|
y (n) = ay (n 1) + x(n)
H(e j ) =

h(n) = an u(n)

M
X

{
b(k)x(n k)

k=0

{z

Feedback
PM
jk
k=0 b(k)e
P
jk
N
k=1 a(k)e

{z

FeedForward

Numerator Polynomial
Denominator Polynomial

For FIR filters, there is no feedback, and the denominator polynomial

is equal to one.

The author (ITC)

lecture #06

November 6, 2016

11 / 23

The author (ITC)

lecture #06

November 6, 2016

12 / 23

Pseudo Code - FIR Filter

A Picture

% Initialization
x1 = 0; x2 = 0; x3 = 9; x4 = 0; x5 = 0
read b0, b1, b2, b3. b4, b5
% Loop
x0 = get next input
% Compute output
y = b0*x0 + b1*x1 + b2*x2 + b3*x3 + b4*x4 + b5*x5
% Delay
x5 = x4; x4 = x3; x3 = x2; x2 = x1; x1 = x0
Repeat

The author (ITC)

lecture #06

November 6, 2016

x(n)

b(1)

b(2)

b(3)

...

b(q)

x(0)

x(1)

x(2)

x(3)

...

x(n q)

y (n)

This is a simple tapped delay line

13 / 23

Pseudo Code - IIR Filter

The author (ITC)

lecture #06

November 6, 2016

14 / 23

A Picture

% Initialization
x1 = 0; x2 = 0; x3 = 0; x4 = 0; x5 = 0
y1 = 0; y2 = 0; y3 = 0; y4 = 0; y5 = 0
read b0, b1, b2, b3. b4, b5
read a0, a1, a2, a3. a4, a5
% Loop
x0 = get next input
% Compute output
y = b0*x0 + b1*x1 + b2*x2 + b3*x3 + b4*x4 + b5*x5
y = y + a1*y1 + a2*y2 + a3*y3 + a4*y4 + a5*y5
% Delay
x5 = x4; x4 = x3; x3 = x2; x2 = x1; x1 = x0
y5 = y4; y4 = y3; y3 = y2; y2 = y1; y1 = y
Repeat
The author (ITC)

b(0)

lecture #06

November 6, 2016

x(n)

b(0)

b(1)

b(2)

b(3)

...

b(q)

x(0)

x(1)

x(2)

x(3)

...

x(n q)

y (n p)

...

y (3)

y (2)

y (1)

y (0)

a(p)

...

a(3)

a(2)

a(1)

a(0)

y (n)

This is not very efficient. It requires M + N storage registers.

later we will see more efficient implementation of this filter that
required only max(M, N) storage registers (canonic realization).
Illustration the power of re-order the computations.
Same idea is used in the FFT (Fast Fourier Transform) algorithm.
15 / 23

The author (ITC)

lecture #06

November 6, 2016

16 / 23

Interconnection of System - Cascade I

x(n)

h1 (n)

Interconnection of System - Cascade II

y (n)

h2 (n)

Note that the log magnitude of the cascade is the sum of the log
magnitude of individual systems,
20 log10 |H(e j )| = 20 log10 |H1 (e j )| + 20 log10 |H2 (e j )|

A cascade is equivalent to a single linear shift-invariant system with a

unit sample response

and the phase and group delay are additive

() = 1 () + 2 ()

h(n) = h1 (n) h2 (n)

() = 1 () + 2 ()
and a frequency response
H(e j ) = H1 (e j )H2 (e j )

The author (ITC)

lecture #06

November 6, 2016

17 / 23

Interconnection of System - Parallel

x(n)

h1 (n)

The author (ITC)

lecture #06

November 6, 2016

18 / 23

Interconnection of System - Feedback I

+

+
+

y (n)
x(n)

+
+

w (n)

y (n)

f (n)

h2 (n)
g (n)

A parallel network is equivalent to a single linear shift-invariant

system with a unit sample response.

This network may be analysed as follows. With

w (n) = x(n) + g (n) y (n)

y (n) = f (n) w (n)

h(n) = h1 (n) + h2 (n)

in the transform domain we have

Therefore, the frequency response of the parallel network is

W (e j ) = X (e j ) + G (e j )Y (e j )

H(e j ) = H1 (e j ) + H2 (e j )
The author (ITC)

lecture #06

November 6, 2016

19 / 23

The author (ITC)

lecture #06

Y (e j ) = F (e j )W (e j )
November 6, 2016

20 / 23

Interconnection of System - Feedback II

Interconnection of System:
Forming New Filters

and
W (e j ) = X (e j ) + G (e j )F (e j )W (e j )

Using an ideal lowpass filter, by interconnecting systems, how can you

design a highpass filter? How about a bandpass filter?

Solving for W (e j )

|Hhp (e j )|

X (e j )
W (e ) =
1 F (e j )G (e j )
j

with
Y (e j ) = F (e j )W (e j ) =

F (e j )X (e j )
1 F (e j )G (e j )

(a) Ideal highpass filter.

it then follows that

|Hbp (e j )|

H(e j ) =

F (e j )
1 F (e j )G (e j )

The author (ITC)

lecture #06

November 6, 2016

21 / 23

The author (ITC)

lecture #06

(b) Ideal bandpass filter.

End of Today Lecture

The author (ITC)

lecture #06

November 6, 2016

23 / 23

November 6, 2016

22 / 23