You are on page 1of 52

DISCRETE-TIME SIGNALS AND SYSTEMS

Discrete-Time Signals

Recall we are interested in discrete-time continuous-amplitude signals


Continuous amplitude

Discrete amplitude

Analog signal

Continuous-time

Digital signal

Discrete-time

Fig. 1 Classification of signals

Strictly, a discrete-time signal is simply a sequence of numbers, represented


mathematically as follows

x = { x[n]}

(1)

where n and x[n ] is nth number of the sequence

However, in digital signal processing, x often comes from sampling an analog


signal xc (t ) periodically, i.e.

x[n] = xc (nT ), n

(2)

where T is sampling period

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-1

Sample index n thus has connotation of time. Suppose xc (t ) was sampled


from t = n1T to t = n2T . We then write
n

x = {x[n]}n2=n

(3)

By convention, it will be assumed x [n ] = 0 for n < n1 and for n > n2 . x is


then said to be a finite length sequence

For very long sequences, possibly infinite in length, we also use following
notation whose meaning is self-evident
x = {, x0 , x1, x2 , x3 , }

n =0

Remarks

(i)

As stated, strictly, x[n ] means nth number of the sequence x

(4)

However, by corruption of use, it can also refer to the entire sequence itself
(ii)

x[n ] is defined only for integer values of n

(iii)

x[n ] is assumed to be always finite in value, i.e.

x [n ] < ,

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

"n

(5)

2-2

Some Basic Sequences

Unit sample sequence (aka discrete-time impulse, or simply, impulse)


0, n 0
d [n ] =
1, n = 0

(6)

d [n]
1

-3

-2

-1

Unit step sequence


0, n < 0
u [n ] =
1, n 0

(7)

u[n]

-3

-2

-1

Exponential sequence
x [n ] = a n , a

(8)

x[n]

x[n]

0<a<1

-1 < a < 0

-3

-2

-1

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

-3

-2

-1

2-3

Alternating sequence (exponential sequence with a =-1)

x[n ] = (-1)n = {, + 1, - 1, + 1, - 1, + 1, }

(9)

n =0
x[n]
1

-3

-2

-1

Periodic sequence
The sequence x[n ] is said to be periodic with period N if
x [n + N ] = x [n ],

"n

(10)

The fundamental period of x[n ] is smallest positive N that satisfies Eq. (10)1

Sinusoidal sequence

x[n] = A cos(qo n + f),

A, qo , f

(11)

A is amplitude, qo is frequency, f is phase


The sinusoidal sequence x[n ] is periodic with period N if and only if

qoN = 2pk , for some

(12)

Complex exponential sequence

x[n] = Aan = A e j f a n e j qon


x[n] = Aan = A a n {cos(qo n + f) + j sin(qon + f)}
where

(13)

A = A e j f , a = a e j qo , A, a , f, qo

Thus alternating sequence has fundamental period N = 2

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-4

Basic Sequence Operations

Suppose x = { x[n]} and y = {y [n]} are two sequences


Scaling:

a x = {a x[n]}, a

(14)

Addition2:

x + y = { x[n] + y [n]}

(15)

x y = { x[n] y [n]}

(16)

Multiplication2:

x*y =

Convolution:

x [k ] y [n - k ] =

k =-

x [n - k ] y [k ]

(17)

k =-

Example 1
x = {1, 2, 3, 4}3n =0

(a)

y = {5, 6, 7}3n =1

y
2x

x +y

11

xy

10

18

28

Therefore

2 x = {2, 4, 6, 8}3n=0
x + y = {1, 7, 9, 11}3n =0
x y = {10, 18, 28}3n =1

Thus, sum and product of two sequences are given, respectively, by their sample-by-sample sum
and product

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-5

(b)

0, n < 0
x [n] = n
can be written more compactly as x[n] = Aan u[n]
Aa , n 0
An
A
n

0
u[n]
1

point-by-point
multiply

A u[n]
A
n

(c)

x[n ] = {1, 4, 0, - 1}2n=-1 can be written mathematically as x [n ] = d [n + 1] + 4d [n ] - d[n - 2]


d[ n ]

1
n

0
d [ n - 2]

n-2 = 0

n=2

d [ n + 1]

n +1= 0

n = -1

-1

Example 1(c) yields the general decomposition


x [n ] =

x [k ] d [n - k ]

(18)

k =-

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-6

Example 1 (cont.)
(d)

Suppose x[n ] = {1, 4, 0, - 1}n=-1 and y [n ] = {1, 0, - 1}n=0 . Define w [n ] = x [n ] * y [n ]


n = -1

w [-1] = x [-1] y [0] = 1 1 = 1

n=0

w [0] = x [-1] y [1] + x [0] y [0] = 1 0 + 4 1 = 4

n =1

w [1] = x [-1] y [2] + x [0] y [1] + x [1] y [0] = 1 (-1) + 4 0 + 0 1 = - 1

n=2

w [2] = x [0] y [2] + x [1] y [1] + x [2] y [0] = 4 (-1) + 0 0 + (-1) 1 = - 5

n=3

w [3] = x [1] y [2] + x [2] y [1] = 0 (-1) + (-1) 0 = 0

n=4

w [4] = x [2] y [2] = (-1) (-1) = 1

Or more conveniently, by Table method

x [n ]
1

-1

-1

-1

-1

-4

y [n ]

Adding along indicated diagonals yields


w [n ] = {1, 4, - 1, - 5, 0, 1}n4=-1

It can be shown that (see Tutorial)


M2

x = {x [n ]}

n =M1
N2

y = {y [n ]}

M 2 + N2

x * y = {w [n ]}

n =M1+N1

(19)

n =N1

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-7

Discrete-Time Fourier Transform (DTFT)

Discrete-time Fourier transform of x[n ] is defined by

DTFT { x [n ]} = X (e j q ) =

x [n ] e- jnq ,

(20)

n =-

X (e j q ) is periodic in q with period 2p since "m


X ( e j ( q + 2 mp ) ) = X ( e j q e j 2m p ) = X (e j q )

(21)

Therefore, we consider only q [-p, p ]

We call q digital frequency. Its unit is radians/sample

Inverse DTFT is given by


DTFT-1 { X (e j q )} = x [n ] =

1
X (e j q ) e+ jnq d q

2p 2p

(22)

where integral is evaluated over any 2p interval of q

For Example 2, following result on geometric series is useful

1 - aN

, a 1

an = 1 - a

n =0
a =1

N,
N -1

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

(23)

2-8

Example 2
x [n ] = d [n ]

(a)

X (e j q ) =

d[n]e- jnq = 1

n=-

x [n ] = a n u[n ],

(b)

X (e j q ) =

n =-

a n u[n ]e- jnq =

a <1

(ae- j q )
n =0

1
1 - ae- j q

1
1- a

1
1+ a

a = 0.75

tan-1( a

1- a 2 )

1
1+ a

1
1- a

a = -0.75
tan-1( a

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

1- a 2 )

2-9

x [n ] = a n ,

(c)

jq

X (e ) =

a e

- jnq

n=-

-1

an e- jnq +
n =0

1
1 - ae- j q
1- a

ae

a <1
-n - jnq

n =-

(ae- j q )

+ ae

n =0

jq m

m=1

jq

1 - ae j q

1 - 2a cos q + a

1+ a
1- a

1- a
1+ a

a = 0.75

x [n ] = - a n u [-n - 1],

(d)
X (e j q ) = -

a n u [-n - 1]e- jnq = -

n =-

a <1

-1

(ae- j q )
n =-

= - a-1e j q

n =1

-1

n
= - (a e j q ) - 1

n =0

-1 j q
-1
Last summation converges only if a e < 1 or a < 1 which contradicts the given condition

a < 1 . Therefore, given sequence does not have a DTFT

?
a = 0.75

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-10

Table of DTFT pairs


Transform X (e j q )

Sequence x [ n ]
1.

d[n ]

2.

u [n ]

1 - e- j q

pd(q - 2pk )

k =-

3.

d[n - m]

4.

a n u[n ] ,

5.

( n + 1)a n u [n ] ,

6.

(n + r - 1)! n
a u[ n ] ,
n !(r - 1)!

7.

1,

0 ,

8.

q n
sin qc n
q
= c sinc c
p
pn
p

1,
q qc

0 , qc < q p

9.

2p

e- j qm

a <1

1 - ae- j q
1

a <1

(1 - ae- j q )2
1

a <1

(1 - ae- j q )r

sin(q(N1 + 21 ))

n N1
n > N1

sin(q 2)

d(q - 2pk )

k =-

2p

10. e j qo n

d(q - qo - 2pk )

k =-

11. cos qo n

{d(q - qo - 2pk ) + d(q + qo - 2pk )}

k =-

12. sin qo n
13. Periodic square wave

1,
n N1

0 , N1 < n N 2

x [n + N ] = x [n ]

14. Impulse train

d(n - kN )

k =-

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

p
{d(q - qo - 2pk ) - d(q + qo - 2pk )}
j k =-
2p

k =-

ak =

ak d (q - 2Npk )

sin 2Npk (N1 + 21 )


N sin 22pNk

2p
d (q - 2Npk )
N k =-

2-11

Properties of DTFT
Let X (e j q ) and Y (e j q ) be the DTFTs of x [ n ] and y [ n ] respectively:

x[n]
X (e j q )
DTFT

y [n ]
Y (e j q )
DTFT

ax[n] + by [n]
aX (e j q ) + bY (e j q )
DTFT

1.

Linearity

2.

Time shifting

3.

Frequency shifting

4.

Conjugation

x * [n]
X * (e- j q )

5.

Time reversal

x[-n]
X (e- j q )

6.

Time expansion

7.

Convolution

8.

Multiplication

9.

Time differencing

x[n - no ]
X (e j q ) e- j qno
DTFT

x[n] e j qo n
X (e j (q-qo ) )
DTFT

DTFT

DTFT

x[n k ], n = multiple of k
DTFT

X (e jk q )
x( k ) [n] =
0,
n multiple of k

x[n] * y [n]
X (e j q )Y (e j q )
DTFT

DTFT

x[n ] y [n]

x[n] - x[n - 1]
(1- e- j q ) X (e j q )
DTFT

10. Accumulation

1
X (e j x )Y (e j (q-x ) ) d x
2p 2p

DTFT

x[k ]

k =-

1
1- e- j q

X (e j q ) + p X (e j 0 )

DTFT

d(q - 2pk )

d
X (e j q )
dq

X (e- j q ) = X * (e j q )

12. x [ n ] real
13. x [ n ] real and even

X (e j q ) real and even

14. x [ n ] real and odd

X (e j q ) purely imag and odd

15. Parsevals relation

x [n ] 2 =

n=-

n=-

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

k =-

j
n x[n]

11. Frequency differentiation

x [n ] y * [n ] =

2
1
X (e j q ) d q

p
2
2p

1
X (e j q )Y * (e j q ) d q
2p 2p

2-12

Example 3
x[n]
3

1
-4 -3 -2 -1 0

DTFT of above sequence can be found as follows


Method 1

Decompose x[n] as follows

2
1
-4 -3 -2 -1 0

-4 -3 -2 -1 0

From Entry 7 of DTFT table and linearity property


X 1( e j q ) =

Method 2

sin q (4 + 21 )
sin (q 2)

+ 2

sin q(1 + 21 )
sin (q 2)

Decompose x[n] as follows

1
-4 -3 -2

-1 0

From Entry 7 of DTFT table, and time-shifting and linearity properties


X 2 (e j q ) =

sin q(1 + 21 )
sin (q 2)

= 2

e + j 3q + 3

sin q(1 + 21 )
sin (q 2)

sin q(1 + 21 )
sin (q 2)

cos 3q + 3

sin q(1 + 21 )
sin (q 2)

e- j 3 q

sin q(1 + 21 )
sin (q 2)

It can be shown (exercise) X1(e j q ) = X 2 (e j q )

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-13

z-Transform

z-transform of x[n ] is defined by the Laurent series


{ x [n ]} = X ( z )
= + x [-2] z 2 + x [-1] z + x [0] + x [1] z-1 + x [2] z-2 +
=

x [n ] z-n ,

(24)

n =-

Recall from Eq. (20)


jq

X (e ) =

x [n ] e- jnq ,

n =-

Therefore, z-transform generalises DTFT:3


X (e j q ) is simply X ( z ) evaluated around the unit circle z = e j q

Region in z-plane where X ( z ) is finite is its region of convergence (ROC)

Clearly, if X (e j q ) exists, then ROC of X ( z ) must include unit circle

ROCs consist of rings in z-plane centered about the origin

Inverse z-transform is given by

-1 { X ( z )} = x[n ] =

1
2p j

C X (z) z

n -1

dz

(25)

where C is a closed contour in ROC of X ( z )

In the same way Laplace transform generalises the Fourier transform

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-14

Example 4

In the following, we re-work Example 2 for z-transforms


x [n ] = d [n ]

(a)

X (z) =

d[n] z-n = 1 and converges everywhere

n=-

x[n] = anu[n]

(b)

Im

X (z) =
=

an u[n] z-n =

n=-

an z-n

n=0

(az

unit circle

-1 n

n=0

1
1 - az-1

-1
provided az < 1

Re

z-plane

i.e., z > a

x [n ] = a n

(c)

X (z) =

a n z-n =

n=-

=
=

-1

n=0

n=-

a-n z-n
Im

(az-1)n + (az)n

n=0

an z-n +

n=1

1
1- az-1

+
- 1

1- az

(1- az ) + (1- az-1 ) - (1- az-1 )(1- az )


(1- az

-1

)(1- az )

1/a
a

Re

1 - a2
(1+ a2 ) - a( z-1 + z )

z-plane

-1
provided az < 1 and az < 1

i.e., a < z <

1
a

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-15

x[n] = - an u[-n - 1]

(d)

Im

X (z) =

-1

n =-

-a n z-n = - (a-1z )n
n =1

1
= -
- 1
1
1- a z

1
=
1- az-1
-1
provided a z < 1

Re

z-plane

i.e. z < a

Comparing Examples 4(b) and 4(d), we see that


X ( z ) is not unique different sequences can have the same X ( z )

X ( z ) s are distinguishable only by their ROCs

Beware:
ROC of X ( z ) , denoted by R X , is plotted in z-plane, not z-1-plane although
by convention, X ( z ) is given in terms of z-1

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-16

Sequence

ROC

x[n]
All z
except z = 0

1.

n
x[n]
All z
except z =

2.

n
x[n]
All z
except z = 0 and

3.

n
x[n]
R < |z|

4.

x[n]
|z| < R+

5.

n
x[n]
R < |z| < R+

6.

Fig. 2 Regions of convergence of z-transforms

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-17

Table of z-transform pairs


Sequence
1.

d[n ]

2.

u [n ]

3.

-u [-n - 1]

4.

d[n - m]

5.

anu[n]

6.

-anu[-n -1]

7.

n a n u[n ]

8.

-n anu[-n - 1]

9.

[cos qo n ]u[n ]

10. [ sin qo n ]u[n]


n
11. r cos qo n u[n]

n
12. r sin qo n u[n]

an , 0 n N - 1
13.

otherwise

0,

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

Transform

ROC

All z

1
1 - z-1
1
1 - z-1

z -m
1
1 - az-1
1
1 - az-1

az-1
(1 - az-1 )2
az-1
(1 - az-1 )2
1 - [cos qo ] z-1
1 - [2cos qo ] z-1 + z-2

[sin qo ] z-1
1 - [2cos qo ] z-1 + z-2
1 - [r cos qo ] z-1
1 - [2r cos qo ] z-1 + r 2 z-2
[r sin qo ] z-1
1 - [2r cos qo ] z-1 + r 2 z-2

1 - aN z-N
1 - az-1

z >1

z <1

All z except 0 (if m > 0 )


or (if m < 0 )
z > a

z < a

z > a

z < a

z >1

z >1

z > r

z > r

z >0

2-18

Properties of z-transform
Let X ( z ) with ROC R X and Y ( z ) with ROC RY be z-transforms of x [ n ] and y [ n ]
respectively:

x[n]
X ( z ), R X

y [n]
Y ( z ), RY

1.

Linearity

2.

Time shifting

3.

Frequency shifting

4.

Conjugation

5.

Time reversal

6.

Time expansion

7.

Convolution

8.

Multiplication

9.

Time differencing

ax[n] + by [n]
aX (z ) + bY ( z ), contains R X RY
x[n - no ]
X ( z ) z-no , R X except for possible addition or deletion
of origin or

x [n ] zon
X

zo R X

x * [n]
X * ( z* ), R X

x[-n]
X ( z-1), 1 R X

x[r ], n = rk

x( k ) [n ] =

X ( z k ), R1Xk

0,
n

rk

x[n] * y [n]
X ( z )Y ( z ), contains R X RY

x[n] y [n]

1
2p j

C X (x )Y ( zx ) 1x d x , contains RX RY , C in ROC

x[n] - x[n - 1]
(1- z-1) X ( z ), contains R X { z > 0}

10. Accumulation

( zz ),

x[k ]

k =-

1- z-1

X ( z ), contains R X { z > 1}

-z
n x[n]

11. Differentiation in z

d
X (z), RX
dz

X ( z ) = X * ( z* )

12. x [ n ] real

x [n ] 2 =

1
2p j

C X (z) X

x [n ] y * [ n ] =

1
2p j

C X (z)Y

13. Parsevals relation

n=-

n=-

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

( z1 ) z1 dz , C in RX
*

( z1 ) z1 dz , C in RX RY
*

2-19

All z-transform properties can be proved from the definition Eq. (24), but proof
can be tedious
Appendix I gives a proof of time-shifting property

Meaning of zo R X , 1 R X and R1X k


Consider

(a)

x [n ] = a n u[n ]

X (z ) =

1 - az -1

RX = { z > a }

Frequency Shifting Property. It follows from the definition Eq. (24) that
x [n ]zon = (azo )n u [n ]

(azo )n z-n =

n =0

1
1 - azo z

-1

1- a

-1

( )
z
zo

provided azo z-1 < 1

(b)

{z

zo R X =

i.e.

> azo } =

{z

> zo a }

Time Reversal Property. Likewise


x [-n ] = a-n u[-n ]

a-n z-n =

n=-

am zm =

m=

(az )m = 1 - az

m=0

provided az < 1

1 RX =

i.e.

(c)

{z

< 1 a}

Time Expansion Property. Finally


x[r ], n = rk
x(k ) [n ] =
0, n rk

x[0]z 0 + 0 z-1 + + 0 z-(k -1)

+ x[1]z-k + 0 z-(k +1) + + 0 z-(2k -1)


+ x[2]z-2k + 0 z-(2k +1) + + 0 z-(3k -1) +
=
=

x[r ](zk )-r

r =0

ar (zk )-r

r =0

i.e.

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

R1X k =

{z

> a

1k

1
1 - az

-k

provided az-k < 1

2-20

Linearity property states

ax[n ] + by [n ]
aX ( z ) + bY ( z ), contains R X RY
Significance of word contains explained in Appendix II due primarily to
pole-zero cancellation or sequence cancellation

Significance of word contains in convolution, multiplication, time-differencing


and accumulation properties can be explained similarly

Example 5
n

( 21 )

x[n ] = 3n u[-n - 1] +

(a)

-1

3n u[-n - 1]

Entry 6:

( 21 )

Entry 5:

u [n ]

1 - 3 z -1

1
1 z-1
2

1-

u[n ]
z < 3

z >

1
2

Linearity Property:
X (z) =

-1
1 - 3z

-1

1 z-1
2

1-

1
2

< z <3

x[n ] = (- 21 ) u[n - 1]

(b)

Observe:

Entry 5:

n-1

x[n ] = (- 21 ) u[n - 1] = (- 21 )(- 21 )


n

(- 21 )

u [n ]

1
1+

1 z -1
2

u[n - 1]

1 - (- 21 )z-1

z > - 21

Time-Shifting Property:
n-1

(- 21 )

u[n - 1]

1
1+

1 z -1
2

z-1,

z >

1
2

Linearity Property:
X ( z ) = (- 21 )

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

z -1
1+

1 z -1
2

z >

1
2

2-21

Example 6
X ( z ) = X 1( z ) + X 2 ( z ) =

1
1 - 0.5z

-1

1
1 - 2z-1

0.5 < z < 2

Suppose X1( z) and X 2 ( z ) have ROCs R1 and R2 and inverse transforms x1[n] and x 2 [n ]
respectively. We see from Entries 5 and 6 of z-transform table, X1( z ) and X 2 ( z ) can each have
two possible inverse transforms
X1( z )

X 2 (z)

(0.5) u[n ],
x1[n ] =

(0.5) u[-n - 1],

2 u [n ],
x2 [n ] =

2 u [-n - 1],

z > 0.5
z < 0.5
z >2
z <2

It thus follows there are 22 = 4 ways of forming x [n ] = x1[n ] + x2 [n ] , depending on R1 and R2


R1

R2

R1 R2

z > 0.5

z >2

z >2

z > 0.5

z <2

0.5 < z < 2

z < 0.5

z >2

z < 0.5

z <2

z < 0.5

As can be seen, only Entries 2 and 3 satisfy R X R1 R2 . However, we must discard Entry 3 since

it leads to x[n] = -(0.5)n u[-n -1] + 2n u[n] which does not have a z-transform. In other words, Entry 3
contradicts the problem statement, that is, there is an X ( z ) . Hence, from Entry 2

x[n] = x1[n] + x2 [n] = (0.5)n u[n] - 2n u[-n - 1]

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-22

Discrete-Time Systems

Systems can be:

continuous-time

or

discrete-time

linear

or

non-linear

time-invariant

or

time-varying

stable

or

unstable

causal

or

non-causal

with memory

or

memoryless

invertible

or

non-invertible

We focus on discrete-time linear time-invariant stable causal systems

Let be a discrete-time system with input x[n ] and output y [n ]

y [n] = { x[n]}
x[n]

(i)

(26)
y[n]

is linear if, for any a, b


{ax1[n] + bx2 [n]} = a { x1[n]} + b { x2 [n]}

(ii)

is time-invariant if, for all no


{ x[n - no ]} = y [n - no ]

(iii)

(27)

(28)

is stable in the bounded-input-bounded-output (BIBO) sense if and only if


for all input sequences satisfying

x[n] Bx < ,

"n

(29)

y [n ] By < ,

"n

(30)

we can find a By such that

(iv)

is causal if, for all no , y [no ] depends only on { x[n], n no }

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-23

Example 7
y [n ] = x [n ] + x [n - 1]

(a)

Linearity
{ax1[n ] + bx2 [n ]} = (ax1[n ] + bx2 [n ]) + (ax1[n - 1] + bx2 [n - 1])
= a ( x1[n ] + x1[n - 1]) + b ( x2 [n ] + x2 [n - 1])
= a { x1[n ]} + b { x2 [n ]}

\ System is linear
Time-Invariance

{ x[n - no ]} = x[n - no ] + x[n -1- no ]


y [n - no ] = x [n - no ] + x [n - no - 1]

{ x[n - no ]} = y [n - no ]

Clearly

\ System is time-invariant

y [n ] = x [-n ]

(b)

Linearity

{ax1[n] + bx2 [n]} = ax1[-n] + bx2 [-n]


= a { x1[n]} + b { x2 [n]}

\ System is linear
Time-Invariance

{ x[n - no ]} = x[-n - no ]
y [n - no ] = x [-( n - no )] = x [-n + no ]

\ System is not time-invariant

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-24

Suppose is linear time-invariant (LTI). Define

h[n] = {d[n]}

(31)

h[n ] is unit sample response, or impulse response, of

(i)

It follows from Eqs. (18), (27) and (28) that

y [n] = { x[n]} = x[k ] d[n - k ] = x[k ] {d[n - k ]}

k =-
k =-

x[k ] h[n - k ]

k =-

y [n ] = x [n ] * h[n ] = h[n ] * x [n ]

i.e.

(32)

Eq. (32) shows an LTI system is completely characterised by h[n ]

z-transform of Eq. (32) is given by

(ii)

where

Y (z) = H (z) X (z)

(33)

H ( z ) = {h[n]} 4

(34)

Hence an LTI system is also completely characterised by H ( z ) . Indeed, the


system is often identified as H ( z )
H ( z ) is called transfer function of the system. It is also called system
function

x[n]

h[n]

y[n]

X(z)

H(z)

Y(z)

Fig. 3 Time- and transform-domain representations of a system

(iii) Frequency response of system is given by the DTFT

H (e j q ) = H ( z ) z=e j q

(35)

Alternatively, since X ( z ) = 1 if x[n ] = d[n ] , so by Eq. (33), Y (z ) x[n ]=d[ n ] = H (z) . But Y ( z ) x[ n ]=d[ n ] =
y [n] x[n]=d[n] and by definition, y [n] x[n ]=d[n ] = h[n] . Therefore, it follows that {h[n ]} = H ( z )

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-25

(iv) It can be shown an LTI system is stable if and only if5

h[n ] <

(36)

n =-

But

h[n ] =

n =-

h[n ] z-n

n =-

z =1

Therefore, a stable H ( z ) must include the unit circle in its ROC

(v)

As can be seen from Eq. (32), an LTI system (stable or unstable) is causal if
and only if h[n ] = 0 for n < 0

(vi) If is LTI and causal, and x[n ] is also causal, then Eq. (32) simplifies to
y [n ] =

x [ k ] h[ n - k ] =

k =0

h[k ] x[n - k ]

(37)

k =0

Suppose an LTI system has impulse response h[n ] = {h0 , h1, h2 , h3 }n =0 and
2
input x [n ] = {a, b, g }n =0
3

Because of linearity, output response can be found by adding three impulse


response sequences, fired off at n = 0 , n =1 and n = 2 , and scaled by
a, b and g respectively, as shown below

n=0

n =1

n=2

n=3

ah0

ah1

ah2

ah3

b h0

b h1

b h2

b h3

g h0

g h1

g h2

g h3

ah2
+b h1
+g h0

ah3
+b h2
+g h1

b h3
+g h2

g h3

b
g
y [n ]

ah0

ah1
+b h0

n=4

n=5

Observe that, skewing above table 45 clockwise yields precisely the convolution table of pp. 7, thus demonstrating principle of convolution table method

See A. V. Oppenheim and R. W. Schafer, Discrete-Time Signal Processing, 3rd ed., Prentice-Hall,
2010, pp. 59-60

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-26

Interconnected Systems

It can be shown (exercise) from Eqs. (32) and (33) that:

Cascaded Systems

Parallel Systems

h1[n]

h2[n]

h1[n]

h2[n]

h1[n]

h2[n]

h1[n] * h2[n]

h1[n] + h2[n]

H1(z)H2(z)

H1(z) + H2(z)

Fig. 4 System interconnections

Example 8

x[n]

w[n]

A(z)

B(z)

y[n]

C(z)

Defining w [n ] as shown and taking z-transforms of x [ n ] , y [ n ] and w [n ] , we get, at summer output


W ( z ) = A( z ) X ( z ) + B( z ) C ( z ) W ( z )

\ W (z) =

A( z )
X (z)
1 - B( z ) C ( z )

Y ( z ) = B( z ) W ( z )

But

\ Y ( z ) = B( z )

Hence

H (z) =

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

A( z )
X (z)
1 - B( z ) C ( z )

Y (z)
A( z ) B( z )
=
X (z)
1 - B( z ) C ( z )

2-27

Linear Constant-Coefficient Difference Equations

Linear constant-coefficient difference equations (LCCDEs) describe an


important class of discrete-time LTI systems6

An Nth order LCCDE is defined by


N

ak y [n - k ] =

k =0

bm x[n - m]

(38)

m =0

where ak and bm are constants, and a0 normally equals 1

Taking z-transform of Eq. (38), we get


N

ak Y ( z ) z

-k

k =0

Y (z) =

bm X ( z ) z-m

m =0
M

or

m =0
N

bm z-m

ak z

-k

X (z)

(39)

k =0

Transfer function of LCCDE system (see Eq. (33)) is given by


M

H (z) =

Y (z)
=
X (z)

m =0
N

ak z

k =0

=
-k

bm -m
z
a
m =0 0

bm z-m

a
1 + k z-k
a
k =1 0

(40)

LCCDE systems form a special class of LTI systems. Can you think of an LTI system that cannot be
described by an LCCDE?

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-28

Re-write H ( z ) as follows
M

bk z

H (z) =

k =0
N

-k

bk z

ak z

-k

k =0

k =0
N

ak z

=
N -k

and

k =0

numerator
Roots of
polynomial are called
denominator

Hence H ( z ) has

zN bM - p z p

M -k

p =0
N

aN -p z

(41)
p

p =0

zeros

of H ( z )

poles

M zeros and N poles at non-zero locations in z-plane

N > M
N - M zeros

at z = 0 if

M - N poles

M > N

Beware!
Although H ( z ) is often expressed as a ratio of two polynomials in z-1 , its
poles and zeros are found by expressing H ( z ) as a ratio of two polynomials
in z

Example 9
H (z) =

Consider

1
1 - az-1

It is often said H ( z ) has no zeros and only one pole at

1 - az-1 = 0

z=a

or

Above statement is not correct since re-writing H ( z ) as follows


H (z ) =

1
1 - az

-1

z
z
=
z
z - a

we see, apart from the pole at z = a , H ( z ) actually has a zero at z = 0

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-29

Let x[n ] be a known input sequence. Output sequence y [n ] can be found by:

(a)

solving LCCDE formally through finding the homogeneous and particular


solutions

(b)

using z-transforms

y [n] = -1 {Y (z)} = -1 {H(z) X (z)}


(c)

determining the linear convolution

y [n] = x[n] * h[n] = h[n] * x[n]


(d)

(43)

assuming y [no - 1], , y [no - N ] are known, compute the forward recursion
N

ak
y [n - k ] +
a
k =1 0

y [n ] = -

or assuming
recursion

bm
x [n - m ],
a
m =0 0

y [no ], , y [no - N + 1]
N -1

ak
y [n - k ] +
k =0 aN

y [n - N ] = -

(42)

n = no , no + 1,

(44)

are known, compute the backward


M

bm
x [n - m ],
m = 0 aN

n = no , no - 1,

(45)

Clearly, to solve Eq. (42), RH must be known


Likewise, to solve Eq. (43), RH must also be known since choice of h[n ]
depends on RH
In next section, it will be shown how, given an LCCDE, we can determine all
possible h[n ] s

Above comment implies knowing LCCDE is not enough more information is


required

Therefore, for example, suppose LCCDE system is causal and stable


(i)

h[n ] causal ROC of H ( z ) has form R- < z

(ii)

h[n ] stable ROC includes unit circle

h[n] causal and stable H ( z ) must have all its poles inside unit circle

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-30

Example 10

Consider the LCCDE


y [n ] = a y [n - 1] + x [n ]

As can be readily verified


H (z) =

1
1 - az-1

Clearly, H ( z ) has a pole at z = a . Therefore, there are two possible ROCs: { z > a } and { z < a } ,
and from z-transform table
h[ n ] = a n u [ n ]

H (z) =

h[n ] = - a n u [-n - 1]

z > a

1
,
1 - az-1

H (z) =

z <a

1
,
1 - az-1

We show below how we can find both forms of h[ n ] recursively


Suppose y [-2] = 0 and x[n ] = d[n] . By the forward recursion
y [n ] = a y [n - 1] + x [n ]

Thus

y[n-1]

x[n]

y[n]

-2

-1

a2

a2

a3

y [n ] = h[n ] = {0, 0, 1, a, a 2 , a3 , }

n =-2

= a n u [n ]

For the backward recursion, suppose y [2] = 0 and x[n ] = d[n] . Re-arranging LCCDE as follows

y [n -1] =

Thus

1
( y [n] - x[n])
a

y[n]

x[n]

y[n-1]

-1 a

-1

-1 a

-1 a

-2

-1 a

-1 a

-3

-1 a

-1 a

y [n ] = h[n ] = {, - a-3 , - a-2 , - a-1, 0, 0, 0}

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

n =-

= - a n u[-n - 1]

2-31

Partial Fraction Expansion (PFE)

Recall Eq. (40). Factorising H ( z )


M

bk z-k

H (z) =

k =0
N

b0
a0

ak z

(1 - ck z-1)

-k

k =1
N

(46)

(1 - dk z

-1

k =1

k =0

non-zero zeros given by c1, , cM

we see:

non-zero poles given by d1, , dN

If dk1 = d k2 = = d ks for some k1 k2 ks , then we say dk1 is an sth


order pole, or a repeated pole with multiplicity s

Suppose H ( z ) has only one repeated pole (with multiplicity s), and poles are
so indexed that d1, , dN -s are simple poles while dN -s +1 is the repeated
pole
PFE of H ( z ) is given, with dN -s +1 = = dN , by
M

H (z) =

N(z)
=
D( z )

bk z-k

k =0
N

b0 (1 - ck z-1)
=

ak z-k

k =0

H (z) =

k =1
N

a0 (1 - d k z-1)
k =1

N -s
s
An
Cm
N ( z ) M -N
= Br z-r +
+

1
-1 m
D( z )
r =0
n =1 1 - d n z
m =1 (1 - d N -s +1z )

(47)

where residues Br , An and Cm are given by


An = (1 - d n z-1) H ( z )

z =dn

N(z)
N

a0 (1 - d k z
k =1
k n

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

(48)
-1

)
z =dn

2-32

Cm

d s -m

s
-1

(1 - d
w
)
H
(
w
)
=

N -s +1

(s - m )!(-dN -s +1)s -m dw s -m
w =dN--1 s +1
1

Cm =

d s -m

s -m
s -m
(s - m )!(-dN -s +1)
dw

N (w -1)


N -s

a

0 (1 - d k w )
k =1

w =dN--1 s +1

(49)

and Br are found by long division of N ( z ) by D( z ) with division process


stopping when degree of remainder is less than N

Can readily extend above procedure to H ( z ) s with more than one repeated
pole

h[n ] is found with following z-transforms

z-r =
= d [n - r ]

1
1 - az-1

a nu[n ] ( k -11)!

=
=
-a nu[-n - 1] 1

( k -1) !

(n + 1) a nu[n ] ( k -11)!

=
=
-(n + 1) a nu[-n - 2] 1
(1 - az-1)2
( k -1)!

1 ( n + 1)( n + 2) a n u [n ] 1

2
( k -1)!

(1 - az-1)3
- 1 (n + 1)(n + 2) a nu[-n - 3] ( k -11)!

1
(1 - az-1)k

1 ( n + 1)( n + 2) ( n + k - 1) a n u [ n ]

( k -1)!

=
=
- 1 (n + 1)(n + 2) (n + k - 1) a nu[-n - k ]
( k -1)!

Since it is necessary that h[n ] 0 as n for H ( z ) to be stable, above


z-transform pairs show causal poles of a stable H ( z ) must lie strictly inside
unit circle and anitcausal poles must lie strictly outside unit circle7

This is consistent with a result derived in pp. 26 where it is shown ROC of a stable LTI system must
include the circle unit

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-33

Example 11

H(z) =

8 - 7.5z-1 + 4.03z-2 - 0.926z-3 + 0.174z-4 - 0.018z-5

(1 - (0.2 + j 0.4)z-1)(1 - (0.2 - j 0.4)z-1)(1 - 0.3z-1)

8 - 7.5z-1 + 4.03z-2 - 0.926z-3 + 0.174z-4 - 0.018z-5


1 - z-1 + 0.53z-2 - 0.156z-3 + 0.018z-4

H ( z ) has M = 5 non-zero zeros, N = 4 non-zero poles and M - N = 1 pole at z = 0 . Two of the

non-zero poles form a complex conjugate pair at d1,2 = 0.2 j 0.4 , and the other non-zero pole is at
d3 = 0.3 but with a multiplicity of s = 2 . Thus PFE will have form

H (z) =

M -N

Br z-r +

r =0
1

N -s

1- d

n=1
2

= Br z-r +
r =0

(i)

An
nz

-1

An

n=11 - d n z

-1

(1 - d

m=1
2

Cm
N -s +1z

-1 m

(1 - 0.3mz-1)m

m=1

B0 and B1 found as follows.


-z-1 + 1
0.018z-4 - 0.156z-3 + 0.53 z-2 - z-1 + 1 -0.018 z-5 + 0.174 z-4 - 0.926z-3 + 4.03z-2 - 7.5 z-1 + 8
-0.018z-5 + 0.156z-4 - 0.53z-3 + z-2

- z- 1

0.018 z-4 - 0.396 z-3 + 3.03 z-2 - 6.5z-1 + 8


0.018 z-4 - 0.156 z-3 + 0.53z-2 - z-1 + 1
- 0.24 z-3 + 2.5 z-2 - 5.5z-1 + 7

Hence B1 = -1 and B0 = 1
(ii)

A1 and A2 found as follows.


A2 = (1 - d 2 z-1 ) H ( z )

8 - 7.5z

-1

z =d 2

+ 4.03 z-2 - 0.926 z-3 + 0.174 z-4 - 0.018 z-5

(1 - (0.2 + j 0.4)z-1)(1 - 0.3z-1)

= 1 - j2

2
z=0.2- j 0.4

A1 = A2* = 1 + j 2
(iii)

C1 and C2 found as follows.


C2 =

d 2-2

2
-1
(1
d
w
)
H
(
w
)
,

2-2
2-2

dw
w =1 d3
(2 - 2)!(-d3 )

d3 = 0.3

8 - 7.5w + 4.03w 2 - 0.926w 3 + 0.174w 4 - 0.018w 5

= (1)

1
(0.2
0.4)
1
(0.2
0.4)
+
j
w
j
w
(
)(
)

w =1 0.3

= 4

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-34

C1 =

d 2-1

(1 - d3w )2 H (w -1 )
,

2-1
2-1

w =1 d3
(2 - 1)!(-d3 ) dw

d3 = 0.3

d
1

(-0.3) dw

8 - 7.5w + 4.03w 2 - 0.926w 3 + 0.174w 4 - 0.018w 5


+
1
(0.2
j
0.4)
w
1
(0.2
j
0.4)
w
(
)(
)

w =1 0.3

d
1

(-0.3) dw

8 - 7.5w + 4.03w 2 - 0.926w 3 + 0.174w 4 - 0.018w 5

2
+
1
0.4
w
0.2
w

w =1 0.3

8 - 7.5w + 4.03w 2 - 0.926w 3 + 0.174w 4 - 0.018w 5


1

(-1)(-0.4 + 0.4w )
(-0.3)
(1 - 0.4w + 0.2w 2 )2

- 7.5 + 8.06w - 2.778w 2 + 0.696w 3 - 0.09w 4

1 - 0.4w + 0.2w

w =1 0.3

=1

Thus
H ( z ) = 1 + (-1) z-1 +

1+ j 2
1 - (0.2 + j 0.4)z-1

1- j 2
1 - (0.2 - j 0.4)z-1

1
1 - 0.3 z-1

4
(1 - 0.3z-1 )2

Suppose H ( z ) is causal. Then, impulse response is given by8


h[n ] = d[n ] - d[n - 1]
+ (1 + j 2)(0.2 + j 0.4)n u [n ] + (1- j 2)(0.2 - j 0.4)n u[n ]
+ (0.3)n u[n ] + 4(n + 1)(0.3)n u[n ]
= d[n ] - d[n - 1] + 2Re {(1 + j 2)(0.2 + j 0.4)n } u[n ] + (4n + 5)(0.3)n u[n ]

Exercise: Write a MATLAB program to evaluate first 10 terms of h[ n ] .

PFE procedure is implemented in MATLAB by the function residuez (in


signal processing toolbox)

Alternatively, since

1+ j 2
1- j 2
2 - 2z-1
+
=
1 - (0.2 + j 0.4)w
1 - (0.2 - j 0.4)w
1 - 0.4z-1 + 0.2z-2
it follows (exercise) from entries 11 and 12 of the z-transform tables, that

n
2 - 2z-1
-1

0.2
-1

= 2( 0.2 ) (cos qo n - 2sin qo n) u[n ], qo = cos


1
2

+
1
0.4
z
0.2
z

n
It can be readily verified using MATLAB that above sequence equals 2Re (1+ j 2)(0.2 + j 0.4) u[n]

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-35

If H ( z ) is causal, h[n ] can also be found by performing a long division of N ( z )


by D( z ) for as many terms as we wish
However, with above procedure, round-off errors will accumulate as n increases
and computed h[n ] will not be accurate for n large

Example 11 (revisited)

H(z) =

8 - 7.5z-1 + 4.03z-2 - 0.926z-3 + 0.174z-4 - 0.018z-5

(1 - (0.2 + j 0.4)z-1)(1 - (0.2 - j 0.4)z-1)(1 - 0.3z-1)

8 - 7.5z-1 + 4.03z-2 - 0.926z-3 + 0.174z-4 - 0.018z-5


1 - z-1 + 0.53z-2 - 0.156z-3 + 0.018z-4

We demonstrate here a method to find the causal h[ n ] by long division. It is instructive to compare
and contrast long division shown below with that shown in pp. 34 to find the residues B0 and B1
8 + 0.5z-1 + 0.29z-2 + 0.347z-3
1- z

-1

+ 0.53z

-2

- 0.156z

-3

+ 0.018z

-4

8 - 7.5z-1 + 4.03z-2 - 0.926z-3 + 0.174z-4 - 0.018z-5


8 - 8z-1 + 4.24z-2 - 1.248z-3 + 0.144z-4
0.5z-1 - 0.21z-2 + 0.322z-3 + 0.03z-4 - 0.018z-5
0.5z-1 - 0.5z-2 + 0.265z-3 - 0.078z-4 + 0.009z-5
0.29z-2 + 0.057z-3 + 0.108z-4 - 0.027z-5
0.29z-2 - 0.29z-3 + 0.1537z-4 - 0.0452z-5 + 0.0052z-6
0.347z-3 - 0.0457z-4 + 0.0182z-5 - 0.0052z-6
0.347z-3 - 0.347z-4

Quotient gives first 4 terms of h[ n ] , namely, {8, 0.5, 0.29, 0.347, }

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-36

Appendix I Proof of z-Transform Time-Shifting Property

Consider the sequence y [n] = x[n - no ] . It follows from Eq. (24) that

Y (z) =

n =-

y [n] z-n =

n =-

x[n - no ] z-n =

x[m] z-( m+no )

m=-

= z-no

x[m] z-m = z-no X ( z )

m=-

Following discussion pertains to ROC of y [n ]

Suppose x [n ] = a nu[n ] with X ( z ) = 1 (1- az-1) , R X = { a < z } (z-transform


table Entry 5)
If no = -1, i.e. x [n ] shifted one sample left, then z -no = z and Y ( z ) will not
converge as z . Thus, RY = RX { z } = { a < z < } , i.e. RY is
R X with deleted
But if no = 2 , i.e. x [n ] shifted 2 samples right, then z-no = z-2 and Y ( z ) will
not converge at z = 0 . But R X already has form { a < z } which excludes
z = 0 . Thus, RY = R X and there is no change in ROC

Suppose next x [n ] = -a nu[-n - 1] with X ( z ) = 1 (1- az-1) , R X = { z < a }


(z-transform table Entry 6)
If no = -1, i.e. x [n ] shifted one sample left, then z-no = z and Y ( z ) will not
converge as z . But R X already excludes z . Therefore, RY = R X
and there is no change in ROC
And if no = 2 , i.e. x [n ] shifted 2 samples right, then z-no = z-2 and Y ( z ) will
not converge at z = 0 . Thus, RY = RX {z = 0} = {0 < z < a } , i.e. RY is R X
with origin deleted

Finally, reversing above arguments, we see that, starting with a sequence such
as x[n ] = an +1u[n + 1] whose ROC is R X = { a < z < } (see above), shifting
x [n ] one sample right will result in y [n ] = x [n - 1] = a nu[n ] whose ROC is
RY = { a < z } . In other words, RY is R X with added. Likewise, shifting
x [n ] = -a n-2u[-n + 1] (whose ROC is R X = {0 < z < a } ) 2 samples left will
result in origin being added to the ROC

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-37

(a)

...

x [n ] = (-0.7) u[n ]

From Entry 5 of z-transform table,


X (z) =

1
1 + 0.7z-1

z > -0.7 = 0.7

Above ROC is consistent with Entry 4 of ROC chart

(b)

...

x [n ] = (-0.7) u [n - 1]

n -1

x [ n ] = (- 0.7)(- 0.7)

u [ n - 1]

Therefore, from Entry 5 of z-transform table, and time-shift and linearity properties

X ( z ) = (-0.7)

z-1
1 + 0.7z-1

As for RX , according to proof of time-shifting property, there should be no change, i.e.


R X = { z > 0.7} . This is consistent with Entry 4 of ROC chart

(c)

...

x [ n ] = (- 0.7) u [ n + 1]

-1

x [n ] = (-0.7)

d[n + 1] + (-0.7) u[n ]

Therefore, from Entries 4 and 5 of z-transform table, and time-shift and linearity properties
X (z) = -

and

1
1
z +
0.7
1 + 0.7z-1

R X = {all z except z } { z > -0.7 } = {0.7 < z < }

Above ROC is consistent with discussion in proof of time-shifting property, and with Entries 2
and 4 of ROC chart

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-38

Appendix II ROCs in Linearity Property

We first consider following 7 examples

(a)

Consider

x[n] = x1[n] + x2 [n] = an u[n] - bnu[n - 4], a b


Then by z-transform table Entry 5 and time-shifting property

X ( z ) = X1( z ) + X 2 ( z ) =

1
1 - az-1

b 4 z-4
1 - bz-1

where for X1( z) and X 2 ( z ) to converge, we require z > a and z > b respectively. In
other words, for X ( z ) to converge, we require z to satisfy, simultaneously, z > a and
z > b . Or in other words,
R X = R1 R2

where R1 = { z > a } and R2 = { z > b }

(b)

Suppose now a = b , i.e.

x[n] = x1[n] + x2 [n] = an u[n] - an u[n - 4]


Since R1 = R2 = { z > a } , it would now appear

R X = R1 R2 = R1 R1 = R1 =

{z > a}

But

X ( z ) = X1( z ) + X 2 ( z ) =

1
1 - az-1

a 4 z-4
1 - az-1

1 - a 4 z-4

1 - az-1
(1 - az-1 )(1 + az-1 + a2 z-2 + a3 z-3 )

1 - az-1

= 1 + az-1 + a2 z-2 + a3 z-3

which corresponds to the finite length sequence 1, a, a2 , a3

}n=0 . Therefore, by ROC chart


3

Entry 1

RX =

{ z > 0}

In other words, because of pole-zero cancellation at z = a , RX has expanded resulting in


R X R1 R2

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-39

(c)

Consider next

x[n] = x1[n] + x2 [n] = - an u[-n - 1] - an u[n - 4]


It would appear z-transform is given by
X ( z ) = X1( z ) + X 2 ( z ) =

1
1 - az-1

a 4 z -4
1 - az-1

= 1 + az-1 + a 2 z-2 + a3 z-3

as in Example (b). But for X1( z ) and X 2 ( z ) to exist, we require respectively z < a and
z >a

which contradicts one another. In other words, X ( z ) cannot exist. As for RX

(assuming it is meaningful to speak of one!), it is given by

{z < a} {z > a} =

RX =

With reference to the linearity property, we see


R1 R2 = = R X

(d)

We next extend previous example (c)

x [n ] = x1[n ] + x2 [n ] = -a n u[-n - 1] - a n u[n ] + b n u[n ],

b a

Clearly, for X1( z) to converge, we require z < a and z > a meaning R1 = ; while for
X 2 ( z ) to converge, we require z > b , i.e. R2 = { z > b } . In other words, for X ( z ) to

converge, three conditions must be satisfied, and because of the contradiction between the first
two conditions, R X =
With reference to the linearity property, we see

R1 R2 = { z > b } = = R X

(e)

Suppose b = a in Example (d), i.e.

x [n ] = x1[n ] + x2 [n ] = -a n u[-n - 1] - a n u [n ] + a n u[n ]

x[n] = - an u[-n - 1]

Clearly

which from Entry 6 of z-transform table, has transform


X (z) =

1
1 - az-1

z < a

In this case, we see

R1 R2 = { z > a } =
while

R X = { z < a } R1 R2

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-40

(f)

Consider next

x [n ] = x1[n ] + x2 [n ] = -a n u[-n - 1] - b n u [n ] + b n u[n ],

b < a

Clearly, R1 = { b < z < a } and R2 = { b < z } . Accordingly,

R1 R2 =

{ b < z < a }{ b < z } = { b < z

< a}

But x [ n ] can be simplified as follows

x[n] = - an u[-n - 1]
whose ROC is

RX =

< a}

R X R1 R2

Hence

(g)

{z

Finally, consider

x[n] = x0 [n] + x1[n] + x2 [n] + x3 [n] + = a0d[n] + a1d[n - 1] + a2d[n - 2] + a3d[n - 3] +


Taking z-transforms, we get from the linearity and time-shifting properties

X (z ) = X 0 ( z ) + X1( z ) + X 2 ( z ) + X3 ( z ) + = 1 + a1z-1 + a2 z-2 + a3 z-3 +


with

R0 R1 R2 R3 = {all z} { z > 0} { z > 0} = { z > 0}

Now, x [ n ] as written above is simply the sample-by-sample decomposition of anu[n] . See Eq.
(18). Accordingly, from Entry 5 of z-transform table
RX = { z > a }

Difficulty here is above results appear to violate the condition


R X (R0 R1 R2 R3 )

Paradox resolved by noting linearity property applies only to a finite sum of subsequences. For
infinite sums, one has to study behaviour of the sum in the limit

Lessons

1.

In most cases, R X = R1 R2 . See Examples (a), (c) and (d)

2.

In some cases, RX can grow, i.e. R X R1 R2 , due to pole-zero cancellation


as in Example (b), or sequence cancellation as in Examples (e) and (f)

3.

Linearity property applies only to finite sum of subsequences. See Example (g)

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-41

4.

Generally, a sketch of x [ n ] can be helpful before one attempts to find X ( z ) .


See Examples (b), (e), (f) and (g)

We next consider the inverse problem, i.e., find x[n ] given X ( z ) and RX

5.

If R X = , then X ( z ) cannot be specified and no unique x[n ] can be


found. Indeed, from Examples (c) and (d), we see there is an infinite number of
sequences with R X =

6.

From Examples (e) and (f), we see that given a non-empty RX and a X ( z ) , the
inverse transform is not unique. However, all these inverse transforms are
identical algebraically

7.

R1 R2 = does not imply an inverse transform cannot be found since RX


may not be empty. See Example (e)

8.

In Tutorial Problem 19, it is shown that if X ( z ) has partial fraction expansion


X (z ) = X1(z ) + + X N (z ) =

where
and

a1

1 - b1z-1

++

aN

1 - bN z-1

(50)

a1, , aN 0

(51)

b1 bN 0

(52)

then X ( z ) will not exhibit any pole-zero cancellations

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-42

Problems

1.

Consider the continuous-time signal

xc (t ) = A cos (8pt )

2.

3.

(a)

What is the (fundamental) period of xc (t ) ?

(b)

1 sec
to yield
Suppose xc (t ) is being sampled with a sampling period of T = 16
the discrete-time sequence x[n] = xc (nT ) . Determine the fundamental period N
of x[n] , and comment on the value NT .

(c)

1 sec
and T =
Repeat Part (b) for T = 18

1
401

sec .

Suppose x = { x[n]}n=2M , y = { y[n]}n=2 N , and w = x * y = {w[n]}n=2 K .


1

Show that

K1 = M1 + N1

and

K2 = M 2 + N2

x[n] = {3, -1, 2, 0, 4,1, - 2}4n=-2

Given

y[n] = {2,1, 0, -1, 3}3n=-1

and
Find the sequences

4.

(a)

w1 = 2 x - y

(b)

w2 = x y

(c)

w3 = x * y

Use the properties of DTFT and the table of DTFT pairs to find the DTFT of the
sequence

x[n] = a n ,

a <1

(This problem was solved in pp. 10 but the solution was based on the geometric series
formula Eq. (23).)

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-43

5.

Consider the discrete-time signal


{-1, 0, 1, 2, 1, 0, 1, 2, 1, 0, -1} , -3 n 7
x[ n] =
0 ,
otherwise

Without explicitly evaluating the DTFT X (e jq ) , find

6.

(a)

X (e jq )

(b)

X (e jq )

(c)

arg X (e jq )
p

q=0
q=p

jq

(d)

-p X (e

(e)

Determine the signal whose DTFT is X (e- jq )

(f)

Determine the signal whose DTFT is Re { X (e jq )}

) dq

Prove Parsevals (Plancherels) Relation

x[n] y* [n] =

n=-

1
X (e jq ) Y * (e jq ) d q

2
p
2p

Hint. Consider the DTFT of g[n] = x[ n] * y* [-n]

7.

The frequency response of a discrete-time linear time-invariant system is known to


satisfy
H (e jq ) = H (e j (q-p ) )

Determine h[5] where h[n] is the impulse response of the system.

8.

A discrete-time linear time invariant system was subjected to the input sequence
x[ n] = ( 1 4 ) n u[n] . The output sequence was observed to have the form y[ n] =
ad[ n] + bd[n -1] where a and b are real constants. Determine a and b, and the
impulse response h[n] of the system, given it is known that H (e j p 2 ) = 1 .

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-44

9.

10.

11.

Determine the z-transform, including the ROC, for each of the following sequences. Use
the properties of z-transforms, the table of z-transform pairs, and the ROC chart.
(a)

x[ n] =

( 12 )

u[ n]

(b)

x[ n] =

( 12 )

(u[n] - u[ n -10])

(c)

x[n] = - 2-n u[-n -1]

(d)

x[n] = (- 12 ) u[-n -1] +

(e)

x[n] = a n ,

a <1

(f)

x[n] = a n ,

a >1

( 12 )

u[n]

Determine the inverse z-transform for each of the following using the properties of
z-transforms, a table of z-transform pairs, the ROC chart, and the PFE.
(a)

X ( z) =

(b)

X ( z) =

1 z -1
2
3 z -1 + 1 z - 2
4
8

1
4

1 z-1
2
3 z-1 + 1 z-2
4
8

z >

11+

11+

< z <

1
2

1
2

Two LTI systems, one with impulse response h[n] and the other with impulse response
g[n] , are cascaded together as shown below.
x[n]

w[n]
h[n]

y[n]
g[n]

Show, using the definition of convolution, that the cascaded system can be represented
by a single system with impulse response h[ n] * g[ n] .
x[n]

y[n]
h[n] * g[n]

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-45

12.

Are the following systems linear? Are they time-invariant?


(a)

{ x[n]} = g[n] x[n] with g[n] given

(b)

{ x[n]} =

k=n
n

x[k ]

(c)

{ x[n]} = e x[ n]

(d)

{ x[n]} = ax[n] + b

(e)

{ x[n]} = x[-n +1]

13.

Example 3.12 of Oppenheim and Schafer, pp. 152, gives the transfer function of a
discrete-time LTI system that cannot be described a finite order LCCDE. Specifically,
H ( z ) = ln(1 + az-1 ) . Give other examples.

14.

When the input to an LTI system is


x[ n] =

( 12 )

u[ n] + 2n u[-n -1]

the output is
y[n] = 6 ( 12 ) u[n] - 6 ( 34 ) u[n]
n

15.

(a)

Find the transfer function H ( z ) of the system. Where are its poles and zeros and
what is its ROC?

(b)

Find the impulse response of the system for all n.

(c)

Write the LCCDE that characterises the system.

(d)

Is the system stable? Is it causal?

A causal LTI system is described by the LCCDE

y[n] =

3
2

y[n -1] + y[n - 2] + x[n -1]

(a)

Find the system function H ( z) = Y ( z) X ( z) and its ROC.

(b)

Find the impulse response of the system.

(c)

You should have found the system is unstable. Find a stable (non-causal) impulse
response that satisfies the difference equation.

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-46

16.

The transfer function of an LTI system has the pole-zero plot shown below. Specify
whether each of the following statements is true, false, or cannot be determined from
the information given.
Im

Re

z-plane

17.

Unit circle

(a)

System is stable.

(b)

System is causal.

(c)

If system is causal, then it must be stable.

(d)

If system is stable, then it must have a two-sided impulse response.

A causal LTI system has transfer function


H ( z) =

1 + z -1
(1 -

1 z -1 )(1
2

1 z -1 )
4

(a)

What is the region of convergence (ROC) of H ( z ) ?

(b)

Is the system stable? Explain.

(c)

Find the z-transform of the input x[n] , including the ROC, that yields the output

y[n] = - 13 (- 14 ) u[n] - 43 (2)n u[-n -1]


n

Hint. First find the z-transform of y[n] , including its ROC.

18.

Given

a n u[n] = =

1
1 - az-1

Use the properties of z-transform to find the z-transform of the following sequences.
(a)

(n +1)a nu[n]

(b)

(n + 1)(n + 2)a nu[n]

(c)

( n + 1)( n + 2) (n + k -1) a n u[n]

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-47

19.

Suppose
X ( z ) = X1 ( z ) + + X N ( z ) =

where
and

a1
1 - b1 z

-1

++

aN

1 - bN z-1

a1, , aN 0
b1 bN 0

Prove that X ( z ) cannot exhibit any pole-zero cancellations.

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-48

Answers

1.

2.

(a)

Fundamental period is Tp = 14

(b)

N = 4 , NT = Tp

(c)

1 N =9
For T = 18
,
and NT = 2T p . For T =

1
401

, sequence is not periodic.

Start by considering the definition of convolution

w[n] =

x[k ] y[n - k ]

k =-

Next, sketch generic plots for the finite length sequences x[k ] and y[ n - k ] and observe
what happens as n increases from - .

3.

(a)

2 x - y = {6, - 4, 3, 0, 9, -1, - 4}4n=-2

(b)

x y = {-2, 2, 0, - 4, 3}3n=-1

(c)

x * y = {6,1, 3, -1,18,1, 3, - 6,11, 5, - 6}7n=-3

4.

DTFT {a n } =

5.

(a)

(b)

(c)

-2q

(d)

4p

(e)

{-1, 0, 1, 2, 1, 0, 1, 2, 1, 0, -1}3n=-7

(f)

{- 12 , 0, 12 ,1, 0, 0,1, 2,1, 0, 0,1, 12 , 0, - 12 }n=-7

1 - a2
1 + a 2 - 2 a cos q

7.

h[5] = 0

8.

a=

16
17

,b=

4
17

, h[n] =

16 d[ n]
17

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

1 d[ n - 2]
17

2-49

9.

(a)

z >

1
,
1 - 12 z-1

1 - ( 12 ) z-10
10

(b)

11.

= 1 + 12 z-1 + + ( 12 ) z-9 , z > 0


9

1 z-1
2

z <

(c)

1
,
1 - 12 z-1

(d)

Transform does not exist.

(e)

10.

1-

1
2

1
2

1 - a2
(1+a 2 ) - a ( z + z-1 )

, a < z <

1
a

(f)

Transform does not exist.

(a)

-4 (- 12 ) u[-n -1] - 3(- 14 ) u[ n]

(b)

x[ n] = 4 (- 12 ) u[ n] - 3(- 14 ) u[ n]

We begin by noting that w[n] = k =- h[k ] x[n - k ] and y[n] = l =- g[l ] w[n - l ] .
The stated result follows by substituting the first equation into the second equation.

12.

Linearity

(a)
(b)
(c)
(d)
(e)

-1

= 1 + z -1 +

1 z -2
2!

13.

H ( z) = ez

14.

(a)

H ( z) =

(b)

h[n] =

( 34 )

(c)

y[n] =

( 34 ) y[n -1] +

(d)

System is stable and causal.

1 - 2 z-1
1n

3 z-1
4

z >

1 z -3 + ,
+ 3!

Time-Invariance

z >0

3
4
n-1

u[n] - 2 ( 34 )

u[n -1]

x[n] - 2 x[n -1]

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-50

15.

16.

17.

18.

z-1

(a)

H ( z) =

(b)

h[n] =

(c)

h[n] = - ( 52 ) 2n u[-n -1] - ( 52 )(- 12 ) u[n]

(a)

Cannot be determined

(b)

Cannot be determined

(c)

False

(d)

True

(a)

z >

(b)

H ( z ) is stable

(c)

X ( z) =

(a)

1
(1 - az-1 )2

(b)

2
(1 - az-1 )3

(c)

1-

3 z-1- z-2
2

z >2

( 52 ) 2n u[n] - ( 52 )(- 12 )

u[n]
n

1
2

1 - 12 z-1
1 - 2 z-1

( k -1)!
(1 - az-1 )k

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-51

X ( z) =

19.

a1
1 - b1 z

-1

++

aN

1 - bN z-1

It follows X ( z ) can be written, for appropriate values of c1 , , c N -1 , as follows

X ( z) =

c1 + c2 z-1 + + cN -1z-( N -1)

(1 - b1z-1 )(1 - b2 z-1 ) (1 - bN z-1 )

Suppose the numerator can be factored such that, with no loss of generality, X ( z ) has
pole-zero cancellation at z = b1 . In other words, we can write X ( z ) as follows.

X ( z) =

c1 (1 - b1z-1 )(1 + d1z-1 + + d N -2 z-( N -2) )


(1 - b1z-1 )(1 - b2 z-1 ) (1 - b3 z-1 )

Given the above expression for X ( z ) , suppose we now wish to perform a PFE. The
residue a1 is then given by

a1 = (1 - b1z-1 ) X ( z )

z=b1

c1 (1 - b1z-1 )(1 + d1z-1 + + d N -2 z-( N -2) )


(1 - b2 z-1 ) (1 - b3 z-1 )

=0
z=b1

which contradicts the original assumption that a1 0 . In other words, X ( z ) cannot


exhibit any pole-zero cancellations.

Curtin University, Australia


YH Leung (2005, 2006, 2007, 2012, 2016)

2-52

You might also like