Z Transform Properties
Z Transform Properties
I. Introduction
In the last lecture we reviewed the basic properties of the z-transform and the corresponding region of convergence. In this lecture we will cover
Stability and causality and the ROC of the z-transform (see Lecture 5 notes)
Comparison of ROCs of z-transforms and LaPlace transforms (see Lecture 5 notes)
Basic z-transform properties
Linear constant-coefficient difference equations and z-transforms
Evaluation of the inverse z-transform using
Direct evaluation (not done in detail in this course)
Partial fraction evaluation
Evaluation using long division
Evaluation using Taylor series
Spring, 2015
This relation plays a big role in dealing with difference equations, as will be discussed below.
Multiplication in time by a complex exponential
n
1
a x n X --- with ROC a R x
a
Note that if a is purely real, this corresponds to a circularly-symmetric expansion or contraction of the zplane. If a is purely imaginary, this corresponds to a rotation of the z-plane.
Convolution in time
x n h n X z H z with ROC R x R h
This, of course, is just like things are with DTFTs.
General multiplication in time
z 1
1
x n w n -------- X W --- d
2j
This is again a contour integral, which we cannot evaluate using the techniques developed in this class.
Hence, we will normally use DTFTs rather than z-transforms whenever we need to consider the z-transforms of multiplied time functions. We will, however, make use of the relationship that describes the ztransforms of a function multiplied by a complex exponential.
Differentiation of the Z-transform
Spring, 2015
dX z
nx n z -------------dz
This is easily proved by differentiating the z-transform equation with respect to z. It plays an important role
in dealing with systems where multiple poles occur in the same location n the z-plane.
h[n]
x[n]
y[n]
Difference equations in discrete time play the same role in characterizing the time-domain response
of discrete-time LSI systems that differential equations play for continuous-time LTI systems.
In most general form we can write difference equations as
N
ak y n k =
k=0
bm x n m
m=0
where (as usual) x n represents the input and y n represents the output. Since we can set a 0 equal to 0
without any loss of generality, we can rewrite this as
N
y n =
k=1
ak y n k =
bm x n m
m=0
Comments:
In this representation we characterize the present output of an LSI system as a linear combination of
past outputs combined with a linear combination of the present and previous inputs.
The difference equations alone do not uniquely specify the system. Initial conditions are needed as
well. Normally we assume initial rest (i.e. the output is zero before the input is applied). Otherwise
the system would be neither linear nor shift-invariant, as discussed in class.
These equations can be solved analytically, just as in the case of ordinary differential equations. Normally the solution involves obtaining the homogenous solution (or the natural frequencies) of the system, and the particular solution (or the forced response). Details and examples of this are provided in
the text.
We can solve these equations using the procedure of iteration. For example, consider the equation
Spring, 2015
1
1
3
y n --- y n 1 --- y n 2 = 3x n --- x n 1
4
8
4
or,
3
1
1
y n = --- y n 1 + --- y n 2 + 3x n --- x n 1
4
8
4
Lets obtain the unit sample response h n for this equation via iteration. As we did in class a few days
ago, this can be done by setting up the table:
x[n-1]
x[n]
y[n-2]
y[n-1]
y[n]
3/8
h[n]
x[n]
y[n]
ak y n k =
k=0
bm x n m
m=0
M
k
ak z Y z =
k=0
Xz
m=0
N
Y z
bm z
k=0
ak z
= X z
bm z
m=0
Spring, 2015
bm z
Yz
m=0
H z = ----------- = --------------------------N
Xz
k
ak z
k=0
In other words, if we have an LSI system that is characterizable by a linear constant-coefficient difference
equation, the z-transform of the unit sample response, which we refer to as the system function, will always
be the ratio of two polynomials in z
ference equation. Virtually all of the systems that we will encounter in this course will be of this form.
For example, in the case of the difference equation we had looked at previously,
3
1
1
y n --- y n 1 --- y n 2 = 3x n --- x n 1
4
8
4
we obtain
3 1
3 --- z
4
Yz
H z = ----------- = ------------------------------------Xz
1 1 1 2
1 --- z --- z
8
4
Now, the zeros of the systems are by definition the values of z that cause the numerator of H z to equal
zero (i.e. the roots of the numerator polynomial in z). Similarly, the poles of the system are the values of z
that cause the denominator of H z to go to zero (or the roots of the denominator polynomial in z). We can
2
obtain the poles and zeros of our example by multiplying numerator and denominator by z .
2
1
3 1
2 3
3z z ---
z 3 --- z
3z --- z
4
4
4
H z = ------------------------------------------------ = -------------------------- = ---------------------------------2 1
1
2
1
1 1 1 2
z 1
z --- z ----- z + ---
z 1 --- z --- z
8
4
4
2
4
8
1
1
1
So, in this system, the zeros are at z = 0 and z = --- and the poles are at z = --- and z = --- .
4
2
4
Pole locations and the ROC
We note that since the z-transform is infinite at the values of z corresponding to the pole locations, the ROC
cannot include the locations of the systems poles. In fact, ROCs are always bounded by circles that are
centered at the origin of the z-plane, and that pass through the locations of the poles. In this case, the potential boundaries of the ROCs are circles of radius 1/4 and 1/2. This means that there are three possible ROCs
for this system:
1
z --- This system would be unstable and have a left-sided unit sample response.
4
Spring, 2015
1
1
--- z --- This system would also be unstable and have a both-sided unit sample response.
4
2
1
z --- This system would be stable and have a right-sided (causal) unit sample response.
2
The exact ROC would be known either because it would be given or because you will know whether the
system is causal and/or stable.
V. Inverse z-transforms
Recall that the equations that define z-transforms are
X z =
x n z
n =
n1
1
x n = -------- X z z
dz
2j
c
As we know, we will not evaluate the complex contour integral for the inverse z-transform directly. Instead
we will use one of the following techniques:
Partial fraction expansion
Long division
Taylor series expansion
Partial fraction expansion
As is discussed briefly in the videotape, partial fraction expansion is a computational hack algorithm that
enables us to obtain the results that we would have obtained if we had gone through the formal procedure
of contour integration over the complex z-plane.
The partial fraction method of obtaining inverse z-transforms builds on the fact that we know that
n
1
z
a u n ----------- = -------------------- for the ROC z a and that
1
za
1 az
n
1
a u n 1 -------------------- for the ROC z a
1
1 az
Spring, 2015
Ai = H z z z
z = zi
1 1
2
1 + --- z
1
z = --4
2
3
--2
= --- = 1
3
--2
1
z = --2
1 1
A 2 = H z 1 + --- z
3 1
3 --- z
4
= ----------------------1 1
1 + --- z
1
z = --2
4
6
= --- = 2
3
1
z = --4
Hence
1
2
H z = ----------------------- + -------------------------1
1 1
11
--- z 1 + --- z
2
4
and if we are told that the system is causal, then the corresponding inverse z-transform is
1n
1 n
h n = --- + 2 ------ u n
2
4
Note that the first samples of this function starting with n = 0 are 3, 0, and 3/8, as determined previously
by iteration.
There are actually three possible inverse z-transforms transforms for this system function:
Spring, 2015
1
1 n
1 n
1. If the ROC is z --- , h n = --- + 2 ------ u n , as noted above. This system is causal and
2
4
2
stable.
1
1 n
1 n
1
2. If the ROC is --- z --- , h n = --- u n 1 + 2 ------ u n . This system is neither causal
2
4
2
4
nor stable.
1
1 n
1 n
3. If the ROC is z --- , h n = --- u n 1 + ------ u n 1 . This system is also neither
2
4
4
causal nor stable.
Partial fractions with numerator order greater than or equal to denominator order:
If the order of the numerator is too large, we can reduce it via long division. For example, if we have the
transform
3
3z + 4z + z + 5
H z = --------------------------------------------------- , we can apply long division as follows:
1
1 3z
(Using MATLAB notation and setting in plain text to control positioning:)
-z^(-2)-2z^(-1)-(2/3)
-3z^(-1)+1 )3z^(-3)+4z^(-2)+ z^(-1)+ 5
3z^(-3)+ z^(-2)
3z^(-2)+ z^(-1)
3z^(-2)- z^(-1)
2z^(-1)+ 5
2z^(-1)-(2/3)
17/3
In other words, the result of this division operation is that
3
2
1 2
17 3
3z + 4z + z + 5
H z = --------------------------------------------------- = z z --- + -------------------1
1
3
1 3z
1 3z
2
and the corresponding inverse z-transform equals n 2 n 1 --- plus the inverse z-transform of
3
17 3
-------------------- , which now has a denominator of higher order than the numerator, and which would be either
1
1 3z
n
Spring, 2015
tively.
Partial fractions with multiple poles in the same location:
If we have multiple poles in the same location, the situation is more difficult. As described in the text, If we
have s poles at location d in the z-plane, the contribution of the multiple pole to the partial-fraction of the ztransform is
s
Cm
---------------------------1 m
m = 1 1 dz
where
ds m
s
1
1
1
C m = ----------------------------------------- ----------------- 1 dw H w evaluated at w = d
sm
sm
dw
s m ! d
While the use of w = z
in the equation above may seem needlessly confusing and arbitrary, it actually
is helpful because the derivatives are easier to evaluate in terms of w than in terms of z
In the general case, with some isolated poles, one set of multiple poles in the same location, and a polynomial of numerator order M and denominator order N, the z-transform would be of the form
N
MN
Hz =
r=0
Br z
Ak
1 m
--------------------- + Cm 1 di z
+
1
1 dk z
k = 1 k i
m=1
where d i is the location of the pole of multiplicity s. If there is more than one location with multiple poles,
additional series of terms similar to the latter term above will be obtained.
Let us consider a simple example with multiple poles:
1
H z = -----------------------------------------------------1 2
1 1
1 1
--- z 1 + --- z
4
2
This transfer function, of course, has a single pole at z = 1 4 and a double pole at z = 1 2 .
Using partial fractions, we would like to rewrite this equation in the form of
C1
C2
A1
H z = -------------------------- + ------------------------- + ---------------------------1
1 1
1 1 2
1 + 1
--- z 1 --- z 1 --- z
4
2
2
The residue A 1 is obtained in conventional fashion:
1 1
A 1 = H z 1 + --- z
1
= --------------------------- 1 2
1 1
1
-z
z = --
2
4
Spring, 2015
1
1
= ----- = --2
9
3
1
z = --4
s m ! d
w=d
for s = 2 , m = 1 , w = z
,w
1 2
1
1
C 1 = ------------------------------ ---------- 1 --- w -----------------------------------------------
1
1
2
2
1
1 1
1 ! 1 2 dw
--- w 1 + --- w
4
2
w=2
1 4
2
= 2 -------------------- = --2
9
1 + 1
---
For the residue C 2 we solve the same equation but with m = 2 . This produces
0
1
1
1 2
C 2 = ------------------------------ ---------- 1 --- w -----------------------------------------------
0
0
2
2
1
1 1
0 ! 1 2 dw
--- w 1 + --- w
4
2
w=2
1
2
= ------------ = --2
3
1 + --4
2
You can verify that this is the correct partial-fraction expansion by multiplying the terms back together or
through the use of the MATLAB command residuez.
As noted above, we can obtain the inverse z-transform for the final term through the use of the differentiation property for z-transforms:
dX z
nx n z -------------dz
Applying this property to the case of the decaying exponential signal produces the transform pairs
1
n
az
na u n -------------------- for the ROC z a and
1
1 az
Spring, 2015
az
na u n 1 -------------------- for the ROC z a
1
1 az
n
Hence, for an ROC of z a we can use the shift property of the z-transform to obtain
1
a n + 1 a
n+1
1
u n + 1 --------------------------1 2
1 az
Note that higher-order poles would yield components of the corresponding inverse z-transforms that
include exponentials multiplied by polynomials of a higher order, with the order of the polynomial equal to
the multiplicity of the pole at a given location minus 1.
If the ROC of the z-transform in question is z 1 2 , the inverse transform can be obtained by combining
all of the inverse transforms:
1 n
1 n
29
23
19
2 1 n
2
1
--- --- + --- --- + --- n + 1 --- u n -------------------------- + ------------------------- + --------------------------- 2
9 4
9 2
3
1
1 1
1 1 2
1 + 1
--- z 1 --- z 1 --- z
2
4
2
3+ 0z^(-1)+(3/8)z^(-2)+ .
1-(1/4)z^(-1)-(1/8)z^(-2) ) 3-(3/4)z^(-1)+
3-(3/4)z^(-1)-(3/8)z^(-2)
0z^(-1)+(3/8)z^(-2)+ 0z^(-3)
0z^(-1) - 0z^(-2) + 0 z^(-3)
(3/8)z^(-2)+0 z^(-3)+ 0z^(-4)
H z = 3 + 0z
Spring, 2015
3
+ --- z 2 +
8
1 + z 1
For example, the z-transform X z = ln ----------------- with ROC z 1 can be obtained using series expan 1 z 1
sion:
3
5
7
1 + z 1
1 z
z
z
X z = ln ----------------- = 2 z + ------- + ------- + ------- +
3
5
7
1 z 1
The corresponding inverse is obtained easily by inspection as in the case of the results of long division.