EE518 Digital Signal Processing

Autumn 2001

University of Washington
Dept. of Electrical Engineering

Lecture 6: Properties of ROC and Inverse z-Transform

Oct 17, 2001
Prof: J. Bilmes <bilmes@ee.washington.edu>

TA: Mingzhou Song <msong@u.washington.edu>

6.1 Properties of ROC (continued)

More on Property 5
x[n] is a right sided sequence and x[n] = 0 for n < N1 . Let r0 such that

n=

|x[n]|r0n =

|x[n]|r0n <

n=N1

So
{z : |z| = r0 } ROC
Now see if
{z : |z| > r0 } ROC
Consider r1 r0 and let r1 = cr0 for c 1.

|x[n]|r1n

|x[n]|(cr0 )n

n=N1

n=N1
1

n=N1

|x[n]|cn r0n + |x[n]|cn r0n

c|N1 |

n=0

|x[n]|r0n + c|N1 |

n=N1

=A

|x[n]|r0n

n=0

|x[n]|r0n <

( {z : |z| = r0 } ROC)

n=N1

where A =

 N1
r1
r0

. So

|x[n]|r1n <

n=N1

which means
{z : |z| > r0 } ROC
Property 7. If x[n] is two sided sequence, and if the circle |z| = r0 is in ROC, then the ROC will consist of a ring in
the z-plane which includes |z| = r0 .
Why?
x[n] = xL [n] + xR [n]

6-1

6-2

where xL [n] is a left sided sequence and xR [n] is a right sided sequence. ROC for xR [n] is bounded inside by outmost
pole of XR (z). ROC for xL [n] is bounded outside by innermost pole of XL (z). Add the two together. Either the two
ROCs overlap, which is a ring shaped ROC for x[n], or there is no ROC for x[n].
Property 8. ROC must be a connected region.
Note from Property 7, any sequence is the sum of a left sided sequence and a right sided sequence. So if intersection
exists, it must be formed by
dR < |z| < dL
Hence, poles determine possible types of ROCs.
Ex: When there are two poles d1 and d2 , and |d1 | < |d2 |, there are three possible ROC candidates, i.e.
ROC1 ={z : |z| < |d1 |}
ROC2 ={z : |d1 | < |z| < |d2 |}
ROC3 ={z : |d2 | < |z|}

30
20

20log10(|H(z)|)

10
0
10
20
30
40
4
50
4

3
3

2
1

Re z

Im z

Figure 6.1: A z-transform H(z) =

6.2

12.4z1 +2.88z2
10.8z1 +0.64z2

with poles around 0.4 j0.6928 and zeros around 1.2 j1.2

Inverse z-Transform

z-Transform is

X(z) =

x[n]zn

n=

Let z = re j

X(z) =

n=

x[n](re j )n =

n=

x[n]rn e jn = X(re j ) = FT {x[n]rn }

6-3

So
FT 1 {X(re j )} = x[n]rn
Then
x[n] =rn FT 1 {X(re j )}
Z
1
=rn
X(re j )e jn d
2
Z
1
=
X(re j )(re j )n d
2
Since z = re j ,
dz = jre j d = jzd
that is,
1
d = z1 dz
j
Therefore,
1
2 j

x[n] =

X(z)zn1 dz

But r can be anything inside the ROC. The inverse z-transform equation can be written as
x[n] =
The contour integral
r, for all r such that

1
2 j

X(z)zn1 dz

(6.1)

is performed at a counter-clockwise closed circular contour centered at the origin with radius
{z : |z| = r} ROC

Another way to derive this.

Theorem 6.1 (Cauchy Integral Theorem).
1
2 j

zk dz =

k=1
k 6= 1

1
0

Proof. See Churchill & Brown (1984), Introduction to Complex Variables.

Consider
1
2 j

X(z)zn1 dz =

1
2 j

1
2 j

x[k]

x[k][n k]

k=

x[k]zk+n1 dz

c k=

zk+n1 dz

( Cauchy Integral Theorem)

k=

=x[n]
Note: the result is valid for both positive and negative values of n.
Note: for z = e j , this reduces to inverse FT, since

I
Z

1
1
n1 
X(z)z dz
=
X(e j )e jn e j je j d
2 j c
2
j
j

z=e
since dz = jre j d.
How to calculate this in general?

(6.2)

6-4

Theorem 6.2 (Cauchy Residue Theorem).

x[n] =

1
2 j

I
c

X(z)zn1 dz = [residues of X(z)zn1 at poles inside c]

(6.3)

Proof. See Churchill & Brown (1984).

Finding residues (to be defined shortly) is often difficult. But if X(z)zn1 is a rational function of z,
X(z)zn1 =

(z)
(z d0 )s

(6.4)

If X(z)zn1 has s poles at z = d0 and (z) has no poles at z = d0 , then

Residue [X(z)zn1 at z = d0 ] =
So if s = 1 (1st order pole), then

 s1

d (z)
1
(s 1)!
dzs1 z=d0

(6.5)

Residue [X(z)zn1 at z = d0 ] = (d0 )

Ex: Find the inverse z-transform of

X(z) =
x[n] =

1
2 j

1
1 az1

|z| > |a|

zn1
1
dz =
1
2 j
c 1 az

zn
dz
c za

The radius of contour must be greater than a since the contour has to be in ROC. We will consider the contour integral
for n 0 and n < 0 separately.
1. n 0
There is only one pole at z = a, i.e.,

zn
has only one pole at z = a
za

So (z) = zn and (a) = an . Therefore

x[n] = an

n0

since s = 1 and no derivative is necessary.

2. n < 0
There are multiple order poles at z = 0, with the order depending on n. There is one pole at z = a.
When n = 1, there are a 1st order pole at z = 0 and a 1st order pole at z = a.
Residue [X(z)zn1 |n=1 at z = a] = a1
Residue [X(z)zn1 |n=1 at z = 0] = a1

( (z) = zn )
( (z) =

So residues sum to 0, i.e.,

x[1] = 0
When n = 2, there are two poles at z = 0 and one pole at z = a.
X(z)zn1 =
with

z2
1 (z)
2 (z)
=
=
z a (z a)1
(z 0)2

1 (z) = z2 and 2 (z) = (z a)1

1
)
za

6-5

So
(Pole at z = a, s = 1)
(Pole at z = 0, s = 2)

Residue [X(z)zn1 |n=2 at z = a] = 1 (a) = a2





d
d
Residue [X(z)zn1 |n=2 at z = 0] =
2 (z)
=
(z a)1 
= (z a)2 |z=0 = a2
dz
dz
z=0
z=0

Residues cancel out again, so

x[2] = 0
We can continue this for x[3], x[4],

Easier Ways to Compute Inverse z-Transform

Inspection.
Use z-transform table.
Ex:
Z

an u[n]
So
1

1
1 az1

|z| > |a|

 n
2
u[n] = x[n]

3
Z

1 23 z1
if x[n] is a right sided sequence.
Ex:
X(z) =

1
1
+
1 a1 z1 1 a2 z1

a1 < |z| < a2

so
x[n] = an1 u[n] an2 u[n 1]
since ROCs of each of the terms in X(z) must overlap for the existence of X(z).
Partial Fraction Expansion.
(There are some nice forms of z-transform that we can get their inverse relatively easily.)
When
X(z) =

P(z)
M bk zk
= k=0
Q(z) Nk=0 ak zk

(6.6)

We can factor the denominator and numerator and get

X(z) =

1
b0 M
k=1 (1 ck z )
a0 Nk=1 (1 dk z1 )

(6.7)

1. N > M and the poles are all first order poles (i.e., all the poles are unique).
N

X(z) =

Ak

1 dk z1

(6.8)

k=1

Note:
Ak = (1 dk z1 )X(z) |z=dk
since
(1 dk z1 )X(z) = Ak +

Al (1 dk z1 )
1 dl z1
l6=k,l=1
N

(6.9)

6-6

Ex:
X(z) =

2
1

3 1
+ 18 z2
4z

2
(1

1 1
1 1
2 z )(1 4 z )

A1
A2
+
1 1
1 2z
1 14 z1

and
1
A1 = (1 z1 )X(z) |z= 1 = 4
2
2
1 1
A2 = (1 z )X(z) |z= 1 = 2
4
4
So
X(z) =

4
1 12 z1

1 14 z1

x[n] = 4

 n
 n
1
1
u[n] 2
u[n]
2
4

2. N M and the poles are all first order poles (i.e., all the poles are unique).

X(z) =

MN
N
k
Ak
M
r
k=0 bk z
=
B
z
+
r

N
1
k
k=0 ak z
r=0
k=1 1 dk z

(6.10)

The first term is obtainable by long division (example later). The second term is obtainable by the same procedure
used when N > M.
In general
max(MN,0)

X(z) =

Br zr +

r=0
max(MN,0)

r=0

Br z

M1
bk zk
k=0
Nk=0 ak zk

(M1 < N)
(6.11)

Nu Si

Aik
+
1
i=1 k=1 1 di z

where
Nu , # of unique poles in denominator
Si = order of i-th pole
Br is obtained as before
 s k



di
1
si
1
(1

d
w)
X(w
)
Aik =
i
(si k)!(di )si k dwsi k
w=d 1

(6.12)

We can verify this by multiplying both sides of X(z) by (1 di z1 )si and following the above operations.
Power Series Expansion.
Ex:
then

1
1
X(z) = z2 z 1 + z1
2
2
1
1
x[n] = [n + 2] [n + 1] [n] + [n 1]
2
2

Ex: Power series.

(1)n+1 an zn Z
X(z) = log(1 + az ) =
x[n] =
n
n=1
1

(1)n+1 an
n

Long Division.
1 + 2.0z1
1 + 0.4z1 0.12z2
= 1 + 1.6z1 0.52z2 + 0.4z3 +

X(z) =

n1
n0

6-7

Then
x[n] =0
n<0
x[0] =1
x[1] =1.6
x[2] = 0.52
x[3] =0.4
..
.