KOM 3781

Discrete-Time Control Systems

Veysel Gazi
Lecture 2: Discrete-Time
Systems and the z-Transform
Outline

• Example
• Approximation of derivatives and the integral
• Discrete-Time Systems
• Transform Methods
• Properties of the z-Transform
• Finding z-Transforms
• Solution of Difference Equations
• The Inverse z-Transform
Example
r + e u y
Gc(s) G(s)
Reference Error Control Output
input - input

H(s)

• Consider the above CT control system

• Assume that we have designed the controller Gc(s) using
CT techniques as (a lead compensator)
s+3
Gc ( s ) = 5
s+6
• We would like to implement Gc(s) on a computer
• Let us ignore for now the A/D and the D/A converters.
• Assume that the initial conditions (of controller) are zero.
Example (cont’d)
r + e u y
Gc(s) G(s)
Reference Error Control Output
input - input

H(s) s+3
Gc ( s ) = 5
U ( s ) = Gc ( s ) E ( s ) s+6
s+3
U (s) = 5 E ( s )  ( s + 6)U ( s) = 5( s + 3) E ( s)
s+6
• The s in Laplace (frequency) domain corresponds to
derivative in time domain
d d d
s  u (t ) + 6u (t ) = 5 e(t ) + 15e(t )
dt dt dt
• Assume that the initial conditions (of controller) are zero.
Approximation of the first derivative

y T Backward difference
y(k-1)
approximation of the 1st
derivative (T => sampling period):
y(k)
dy y (k ) − y (k − 1)
k–1 k t =
dt T
(average slope)

• Alternative approximations are also possible.

• Forward difference
• Central difference
Example (cont’d)
d d
u (t ) + 6u (t ) = 5 e(t ) + 15e(t )
dt dt

u (k ) − u (k − 1) e(k ) − e(k − 1)
+ 6u (k ) = 5 + 15e(k )
T T

1
u (k ) = ( u (k − 1) + (5 + 15T )e(k ) − 5e(k − 1) )
(1 + 6T )

• The current control input u(k) can be calculated as a

function of the previous input u(k-1) and the current and
previous errors e(k) and e(k-1).
• Can easily be implemented on a digital computer.
Example
r + e u y
Gc(s) G(s)
Reference Error Control Output
input - input

H(s)

• Consider again the above CT control system

• Assume that we have designed the controller Gc(s) using
CT techniques as (an integrator)
2
Gc ( s ) =
s
• We would like to implement Gc(s) on a computer
• Let us ignore for now the A/D and the D/A converters.
• Assume that the initial conditions (of controller) are zero.
Example
r + e u y
Gc(s) G(s)
Reference Error Control Output
input - input

H(s)
2
U ( s ) = Gc ( s ) E ( s ) Gc ( s ) =
s
• The 1/s in Laplace (frequency) domain corresponds to
integral in time domain
2 1
U (s) = E ( s)
s s
 
t
u (t ) = 2  e( )d
0

• Assume that the initial conditions (of controller) are zero.

Approximation of the integral
(Trapezoidal rule)
The integral (area in yellow) between two
samples may be approximated by various
methods. Δ A(k)

error y(k)
y
y(k-1) A1(k)
Δ A(k)
A(k-1)
y(k-1)×T
T t
k-1 k
y (k ) − y (k − 1) Area of the
A1 (k ) = T
2 trapezoid – area
y (k ) + y (k − 1) between two
A(k )  y (k − 1)  T + A1 (k ) =  T samples.
2
Approximation of the integral
(Trapezoidal rule)
The integral (area in yellow) between two
samples may be approximated by various
methods. Δ A(k)

error y(k)
y
y(k-1) A1(k)
Δ A(k)
A(k-1)
y(k-1)×T
T t
k-1 k

y (k ) + y (k − 1)
A(k ) = A(k − 1) + A(k )  A(k − 1) + T
2
Approximation of the total area up to step k.
Example (cont’d)
t
u (t ) = 2  e( )d
0

e(k ) + e(k − 1)
u (k ) = u (k − 1) + 2T = u (k − 1) + T (e(k ) + e(k − 1))
2
• Once again, the current control input u(k) can be
calculated as a function of the previous input u(k-1) and
the current and previous errors e(k) and e(k-1).
• Can easily be implemented on a digital computer.
• Other, more complicated controllers may be designed
similarly.
• Contains simple (adition/subtraction,
multiplication/division operations.
Approximation of the first derivative
T Backward difference
y y(k-1) approximation of the 1st
derivative (T => sampling period):
y(k) dy y (k ) − y (k − 1)
=  yk
k–1 k t dt T
(average slope - backward)

T Forward difference
y y(k+1) approximation of the 1st
derivative (T => sampling period):
y(k) dy y (k + 1) − y (k )
=  yk
k k+1 t dt T
(average slope – forward)
Approximation of the second derivative
The backward difference approximation of the 2nd
derivative is
y (k ) − y (k − 1) y (k − 1) − y (k − 2)
−
d y yk − yk −1
2
T T
 =
dt 2 T T
(average change in
y (k ) − 2 y (k − 1) + y (k − 2) slope - backward)
=   2
y k
T2

The forward difference approximation of the 2nd derivative

is
y (k + 2) − y (k + 1) y (k + 1) − y (k )
−
d 2 y yk +1 − yk T T
 =
dt 2 T T
y (k + 2) − 2 y (k + 1) + y (k ) (average change in
= 2
  yk
2
slope - forward)
T
Approximation of the third derivative
The backward difference approximation of the 3rd
derivative is
y (k ) − 3 y (k − 1) + 3 y (k − 2) − y (k − 3)
 yk =
3

T3

The forward difference approximation of the 3rd

derivative is
y (k + 3) − 3 y (k + 2) + 3 y (k + 1) − y (k )
 yk =
3

T3
• The backward difference approximations are more
suitable since they are based on past values (and not
future values).
• Central difference approximations (not shown) are also
possible.
Approximation of the derivative
• Note that the coefficients in
the derivative are similar to
those in the binomial
expansion!

• The same pattern continues in

the higher ordered
derivatives as well.
Approximation of the integral
(Rectangular rule – left side)
The integral (area in yellow) between two
samples may be approximated by various
methods.
Assume that x(k) holds (i.e.,
the function is constant). for
the whole interval.

A(k + 1)  A(k ) + x(k )  T

Approximation of the
Rectangular rule – left side. total area up to step k+1.
Approximation of the integral
(Rectangular rule – right side)
The integral (area in yellow) between two
samples may be approximated by various
methods.
Assume that x(k+1) holds (i.e.,
the function is constant) for
the preceding interval.

A(k + 1)  A(k ) + x(k + 1)  T

Approximation of the
Rectangular rule – right side. total area up to step k+1.
Approximation of the integral
(Trapezoidal rule)
The integral (area in yellow) between two
samples may be approximated by various
methods. Δ A(k)

error y(k)
y
y(k-1) A1(k)
Δ A(k)
A(k-1)
y(k-1)×T
T t
k-1 k

y (k ) + y (k − 1)
A(k ) = A(k − 1) + A(k )  A(k − 1) + T
2
Approximation of the total area up to step k.
Discrete-Time Systems
r+ e A/D u y
Computer D/A Plant
-
Sensor

• Given an input signal e(t), the analog to digital

converter (A/D) samples the signal at discrete
instances of time, usually with fixed sampling period
T. e(t) e(k)
A/D

e(k)
(k-2)T kT (k+1)T k

(k-1)T
Discrete-Time Systems

e(k)
(k-2)T kT (k+1)T k

(k-1)T

• The discrete signal e(k) is in the form

e(0), e(1T), e(2T), e(3T), …, e((k-1)T), e(kT), e((k+1)T), …
• For simplicity we usually omit T from the expression
e(0), e(1), e(2), e(3), …, e(k-1), e(k), e(k+1), …
• k shows in a sense the index of the sample
Discrete-Time Systems
• Linear constant-coefficient difference equations describe
the behavior of linear time-invariant discrete-time
systems. Consider a system with input, e(k), and output,
u(k).
e(k) u(k)
Discrete-Time
System (DTS)

• The system behavior is described by recurrence

equations (difference equations).
• The compensators for digital control systems are
implemented on microprocessors as difference equations.
• Given an input, e(k), the solution of the difference
equation provides the output, u(k).
Difference Equations
• A generalized difference equation has the form

u (kT ) +  k −1u ((k − 1)T ) + +  k − nu ((k − n)T )

=  k e(kT ) +  k −1e((k − 1)T ) + +  k − n e((k − n)T )

• Omitting the sampling interval, T one obtains

u (k ) +  k −1u (k − 1) + +  k − n u ( k − n)
=  k e(k ) +  k −1e(k − 1) + +  k − n e( k − n )

• The current value u(k) can be expressed as

u (k ) = −  k −1u (k − 1) − −  k − n u ( k − n)
+  k e(k ) +  k −1e(k − 1) + +  k − n e( k − n )
• This is a recurrence equation also referred to as
difference equation.
Difference definitions
• The recurrence equation
u (k ) = −  k −1u (k − 1) − −  k − n u ( k − n)
+  k e(k ) +  k −1e(k − 1) + +  k − n e( k − n )
• Can be expressed in terms of the differences

uk = u (k ) − u (k − 1) Is called the first difference.

 2uk = uk − uk −1 Is called the second difference.

3uk =  2uk −  2uk −1 Is called the third difference.

• This is why it is also called difference equation.

Difference Equations
u (k ) +  k −1u (k − 1) + +  k − n u ( k − n)
=  k e(k ) +  k −1e(k − 1) + +  k − n e( k − n )

• Difference equation might arise as

• Result of measurements
• Description of a compensator
• Approximation to a differential equation
• Difference equations are fundamental to digital control
systems analysis and design.

• We will consider difference equations extensively in this

course.
Iterative Solution of Difference
Equations e(k) DTS u(k)
Example: Consider the 1st order linear difference equation:
u (k ) = e(k ) − e(k − 1) − u (k − 1) k 0
(what makes this 1st order?)
Given: e(k ) = 1 for k even e(k ) = 0 for k odd
e(−1) = 0 u (−1) = 0 e(k) 1
Solution:
5 k
u (k ) = 0 for k  0, and -2 -1 0 1 2 3 4

u (0) = e(0) − e(−1) − u (−1) = 1 − 0 − 0 = 1

u (1) = e(1) − e(0) − u (0) = 0 − 1 − 1 = −2
u (2) = e(2) − e(1) − u (1) = 1 − 0 − (−2) = 3
u (3) = e(3) − e(2) − u (2) = 0 − 1 − 3 = −4
u (k ) = = (−1) k (k + 1)
We can find u(k) for any value of k, but not always in closed form!
Transform Methods
Definition of z-Transform
• Consider operator, sometimes called “delay” z -1
• A complex number or variable z = r ejω (r – real)
• Similar to s in Laplace domain
• Given a sequence of values e(k),

e(k ) –> e(0), e(1), e(2), , e(  )

• The z-transform of e(k) is defined as


E( z) = [{e(k )}] =  e(k ) z − k
k =0

Note that E(z) is expressed as a power series expansion!


E( z) = [{e(k )}] =  e(k ) z − k = e(0) + e(1) z −1 + e(2) z −2 + + e() z −
k =0
Useful Identity


1
 ( x)
k =0
k
=1  x + x  x + x  x +
2 3 4 5
=
1 x
;( for x  1)

Used for expressing power series solutions in closed form!

Example: Find z-transform of the sequence:
 −1 
k
1 1 1 1
e(k ) = 1, − , , − , , ,  
2 4 8 16  2 
 1  −1  1  −2  1  −3
E( z) = 1 −   z +   z −   z +
2 4 8
1 1 1
= 1− + − +
2z ( 2z ) ( 2z )
2 3

 k
 1 
=  − 
k =0  2z 
1 z z 1 −1
= = = ;( for z  1)
 1  z + 1 z + 0.5 2
1+  
 2z  2

1
From:  ( x) k = 1  x + x 2  x3 + x 4  x 5 + = ;( for x  1)
k =0 1 x
Example Transforms
1, k  0 
The z-transform of the step function: u (k ) =  
 0 k  0 
1
u (k ) =  0 u (k ) z =  0 z = 1 + z + z + = -1
 −k  −k -1 -2

1− z
z
= ( for z −1  1)
z −1

The z-transform of the step function in Matlab.

syms k
uk=1^k;
Uz=ztrans(uk)

Uz =
z/(z - 1)
Example Transforms
The z-transform of u(t) = a t
[a t ] = ? ( sampled form is : u (t )t = kT = a kT )

U ( z ) =  0 u (kT ) z = 0 a z = 0 (a z )
 −k  kT −k  T −1 k

= 1+ a z + (a z )
−1 −1 2 1 z
T T
+ = T −1
= ( for aT z −1  1)
1− a z z − aT

The z-transform of x(t) = a t in Matlab.

syms a k T
uk=a^(k*T);
Uz=ztrans(uk)

Uz =
z/(z - a^T)
Example Transforms
Find the z-transform of u(t) = e–at : u(kT) = e–akT
U ( z ) =  0 u (k ) z =  0 e − akT z − k
 −k 

= 0 (e )
 − aT −1 k
z
= 1 + (e − aT z −1 ) + (e − aT z −1 ) 2 +
1 z − aT −1
= − aT −1
= − aT
( for e z  1)
1− e z z −e

The z-transform of u(kT) = e–akT in Matlab.

syms e a k T
uk=e^(-a*k*T);
Uz=ztrans(uk)

Uz =
z/(z - 1/e^(T*a))
Example Transforms
 t , t  0
The z-transform of ramp function: u (t ) =  
 0, t  0 
U ( z ) =  0 u (k ) z =  0 kTz = T  0 kz − k
  
−k −k

= T ( z −1 + 2 z −2 + 3 z −3 + 4 z −4 + ) kT , k  0 
u (k ) =  
z −1 Tz  0, k  0 
=T = ( for z −1  1)
(1 − z −1 ) 2 ( z − 1) 2

The z-transform of unit ramp in Matlab.

syms k T
uk=k*T;
Uz=ztrans(uk)

Uz =
(T*z)/(z - 1)^2
Useful Identity

1
 ( x)
k =0
k
=1  x + x  x + x  x +
2 3 4 5
=
1 x
;( for x  1)

Used for expressing power series solutions in closed form!


d  
k 

k =0
k ( x) = ( x)   ( x)  =
k

dx  k =0 
= ( x)(0  1 + 2 x  3 x 2 + 4 x 3  5 x 4 + )
d  1  x
= ( x)  = ;( for x  1)
dx  1 x  (1 x) 2

Another useful power series identity in closed form!

Properties of the z-Transform
Addition and Subtraction
e1 (k )  e2 (k ) = E1 ( z )  E2 ( z )
Multiplication by a Constant
 ae(k ) = aE ( z )
Real Translation n > 0
 e( k − n )  = z −n
E( z)
Properties of the z-Transform
Initial Value Theorem
e(0) = lim E ( z )
z →

Final Value Theorem

lim e(k ) = lim( z − 1) E ( z )
k → z →1
Properties of the z-Transform
Consider e1(k) = 0 and e2(k) = 0 for k < 0, with
corresponding transforms E1(z) and E2(z).

Addition and Subtraction e1 (k )  e2 (k ) = E1 ( z )  E2 ( z )


Proof: e1 (k )  e2 (k ) =  e1 (k )  e2 (k ) z − k
0

=  e1 (k ) z − k  e2 (k ) z − k 
0
 
=  e1 (k ) z − k    e2 (k ) z − k  = E1 ( z )  E2 ( z )
0 0
Properties of the z-Transform
Consider e(k) = 0 for k < 0, with corresponding transform E(z).

Multiplication by a Constant  ae(k ) = a E ( z )

Proof:
 
 ae(k ) =  ae(k ) z − k = a  e(k ) z − k = aE ( z )
0 0

• The Z transform is a linear operator.

• It satisfies the superposition principle.
Properties of the z-Transform
Consider e(k) = 0 for k < 0, with corresponding transform E(z).

Real Translation (time delay), where n > 0

 e( k − n )  = z − n E ( z )

Proof:
equals zero

 e ( k − n )  = e ( − n ) z 0
+ e ( − n + 1) z −1
+
+ e(0) z − n + e(1) z − n −1 +
= z − n e(0) + e(1) z −1 +  = z − n E ( z )
Example
Real Translation (time delay) 1 z
e − akT  = =
Let x(t) = e-at , x(kT) = e-akT 1 − e − aT z −1 z − e − aT

For x((k − 3)T ) = e − a ( k −3)T u[(k − 3)T ]

 z  1
e − a ( k −3)T u[(k − 3)T ] = z −3  − aT
=
 z 2 ( z − e − aT )
 z − e

from :  e( k − n )  = z − n E ( z )
Properties of the z-Transform
for:  x(k )  = X ( z )
Complex Translation
e akT x(k )  = X ( ze − aT )

Proof from definitions:

e akT x(k )  = x(0) + x(1)e aT z −1 + x(2)e 2 aT z −2 +
= x(0) + x(1)( ze − aT ) −1 + x(2)( ze − aT ) −2 +
= X ( ze − aT )
Properties of the z-Transform
Initial Value Theorem e(0) = lim E ( z ) (if the limit exists)
z →

Proof: equals zero

lim E ( z ) = lim ( e(0) + e(1) z −1 + e(2) z −2 + )
z → z →

lim E ( z ) = e(0)
z →
Example
Find the initial value, if it exists, for

(1 − e −T ) z −1 (1 − e −T )
X ( z) = =
−1 −T −1
(1 − z )(1 − e z ) (1 − z −1 )( z − e −T )

 (1 − e −T )  (1 − e −T )
x(0) = lim X ( z ) = lim  − −  = =0
z → z → (1 − z )( z − e
1 T
)  (1)()

Properties of the z-Transform
Final Value Theorem
lim e(k ) = lim( z − 1) E ( z ) (if the limit exists)
k → z →1

Proof:

e(k ) = E ( z ) =  e(k ) z − k
k =0

e ( k − 1) = z E ( −1
z ) =  e ( k − 1)
k =0
z −k

 

 e( k ) z
k =0
−k
−  e(k − 1) z − k = (1 − z −1 ) E ( z )
k =0

 
−k 
lim   e(k ) z −  e(k − 1) z  = [e(0) − e(−1)] + [e(1) − e(0)] +
−k
=
z →1
 k =0 k =0 
e() = lim e(k ) = lim(1 − z −1 ) E ( z ) = lim( z − 1) E ( z )
k → z →1 z →1
Example
Find the final value of u(k), if it exists, for
T z z +1
U ( z) =
2 z − 0.5 z − 1

T z ( z + 1)   T z ( z + 1) 
lim u (k ) = lim( z − 1)   = lim   = 2T
k → z →1
 2 z − 0.5 ( z − 1)  z →1
 2 ( z − 0.5) 
Finding z-Transforms
z z
1 = e − akT  =
z −1 z − e − aT

Find z-transform of

f (t ) = 1 − e − t f (kT ) = 1 − e − kT

−T
z z z (1 − e )
F ( z) = [1 − e − kT ] = [1] − [e − kT ] = − =
z − 1 z − e −T ( z − 1)( z − e −T )
Finding z-Transforms
z
The z-transform of sin(ωt): e  akT  =
z − e  aT
sin(t ), t  0  e jt = cos(t ) + j sin(t )
u (t ) =  
 0, t  0 e − jt = cos(t ) − j sin(t )


sin(t ) =
2j
( e −e )
1 jt − jt
u (k ) = (
1 j kT
2j
e − e − j kT ) , k  0

U ( z ) =  0 u (k ) z
 −k
=
1
2j
( 0 e
 j kT
z −k
−  0 e − j kT z − k =

)
 1  
jT − jT
1  z z z (e − e )
=  jT − − jT  = 2 j  2 jT − jT 
2 j  z −e z −e   z − z (e + e ) + 1 
z sin(T )
= 2
z − 2 z cos(T ) + 1
Finding z-Transforms
The z-transform of cos(ωt): e  akT  =
z
z − e  aT
cos(t ), t  0  e jt = cos(t ) + j sin(t )
u (t ) =  
 0, t  0  e − jt = cos(t ) − j sin(t )

u (k ) = ( e + e − j kT ) , k  0
1 j kT
cos(t ) = ( e jt + e − jt )
1
2 2

U ( z ) =  0 u (k ) z − k =
 1
2
( 0

e j kT − k
z + 0 e
− j kT − k
z = )
 1  2 z − z (e + e ) 
jT − jT
1 z z 2
=  jT + − jT  = 2  2 jT − jT 
2 z −e z −e   z − z (e + e ) + 1 
z 2 − z cos(T )
= 2
z − 2 z cos(T ) + 1
Calculating the z-Transform from
the Laplace Transform
Example: Given below G(s), find corresponding G(z) for T=0.1:
s 2 + 4s + 3 s 2 + 4s + 3 0.375 0.25 0.375 Expressed as
G (s) = 3 = = + +
s + 6s + 8s s ( s + 2)( s + 4)
2
s s+2 s+4 partial fractions
Laplace Time z-
Substitute corresponding Transform Function Transform
z-Transforms 1 z
e − at
s+a z − e − aT
1 z
z z z 1(t )
G ( z ) = 0.375 + 0.25 + 0.375 s z −1
z −1 z − e −2T z − e −4T

z z z
= 0.375 + 0.25 + 0.375
z −1 z − 0.8187 z − 0.6703

z 3 − 1.658 z 2 + 0.6804 z
= 3 Express as ratio of two polynomials
z − 2.489 z 2 + 2.038 z − 0.5488
z-Transform Example – MATLAB
s 2 + 4s + 3
T=0.1; G (s) = 3
num=[1 4 3]; s + 6 s 2 + 8s
den=[1 6 8 0]; % no repeated poles allowed
n=length(den);
Gs=tf(num,den)
[r,p,k]=residue(num,den); % Get poles & residues
for i=1:1:n-1
pz(i)=exp(p(i)*T) % find poles in z-plane
end
[numzz,denz]=residue(r,pz,k) % substitute z-plane poles
numz=conv(numzz,[1 0]) % multiply by z
Gz=tf(numz,denz,T) % display G(z)

Transfer function:
s^2 + 4 s + 3
-----------------
s^3 + 6 s^2 + 8 s
Transfer function:
z^3 - 1.658 z^2 + 0.6804 z
----------------------------------
z^3 - 2.489 z^2 + 2.038 z - 0.5488
Properties of the z-Transform
z-Transforms of Common Functions
Solution of Difference Equations
Using z-Transform
Example: Second order system e(k) u(k)
DTS

u (k ) + 1.2u (k − 1) + 0.35u (k − 2) = e(k ) − 0.5 e(k -1)

Take z-transform of both sides of the equation
U ( z ) + 1.2 z −1U ( z ) + 0.35 z −2U ( z ) = E ( z ) − 0.5 z −1E ( z )
Solving for U(z) gives:
1 − 0.5 z −1 z ( z − 0.5)
U ( z) = −1 −2
E( z) = 2 E( z)
1 + 1.2 z + 0.35 z z + 1.2 z + 0.35
• As with Laplace, the result is an algebraic equation.
• Given e(k) (or the corresponding E(z)), one can solve for
U(z) (and therefore for u(k)).
Solution of Difference Equations
Using z-Transform
Example: Second order system e(k) u(k)
DTS

1 − 0.5 z −1 z ( z − 0.5)
U ( z) = −1 −2
E ( z ) = E( z)
1 + 1.2 z + 0.35 z z + 1.2 z + 0.35
2

Assume that e(k) is a unit step:

z z 2 ( z − 0.5) Find
Then: E ( z ) = U ( z) = 2
z −1 ( z + 1.2 z + 0.35)( z − 1) u(k)

• From the calculated U(z) one can solve for the

corresponding u(k).
The Inverse z-Transform
• Power Series
• Partial Fraction Expansion
• Inversion Formula
• Discrete Convolution
Power Series Solution for
Inverse z-Transform
Solve the previous example using long division:
z 2 ( z − 0.5) z 3 − 0.5 z 2
U ( z) = 2 = 3
( z + 1.2 z + 0.35)( z − 1) z + 0.2 z 2 − 0.85 z − 0.35

Long division:
U ( z ) = ( z 3 − 0.5 z 2 )  ( z 3 + 0.2 z 2 − 0.85 z − 0.35) = 1 − 0.70 z −1 + 0.99 z −2 −
U ( z ) =  0 u (k ) z − k


From U(z), the corresponding u(k) can be deduced as

u (k ) = {1, − 0.70, 0.99, − 0.443, }

Partial Fraction Expansion
z 2 ( z − 0.5)
U ( z) = 2
( z + 1.2 z + 0.35)( z − 1)
• As with Laplace, the commonly used technique is to
expand in partial fractions.
• Take the partial fraction of U(z)/z (this is a key step
to remember):
U ( z) A B C
= + +
z z − 1 z + 0.7 z + 0.5
This generates elements that are easy to inverse transform.
U ( z) z ( z − 0.5) z ( z − 0.5)
= 2 =
z ( z + 1.2 z + 0.35)( z − 1) ( z − 1)( z + 0.7)( z + 0.5)
 (1)(.5)   (−0.7)(−1.2)   ( −0.5)( −1) 
 (1.7)(1.5)   (−1.7)( −0.2)   ( −1.5)(0.2) 
=   +   +   =
0.196 2.471 1.667
+ −
z −1 z + 0.7 z + 0.5 z − 1 z + 0.7 z + 0.5
Multiplication by z leads to the following:

0.196 z 2.471z 1.667 z

U ( z) = + −
z −1 z + 0.7 z + 0.5

Finally, the inverse transform is:

u (k ) = 0.196 + 2.471(−0.7) k − 1.667(−0.5) k

1
u(k) 0.99

0
k
-0.443
-0.70
MATLAB® residuez command: [r,p,k]=residuez(b,a)
[b,a] = residuez(r,p,k)
z 3 − 0.5 z 2
U ( z) = 3
z + 0.2 z 2 − 0.85 z − 0.35
>> [r,p,k]=residuez([1 -0.5 0 0],[1 0.2 -0.85 -0.35])

r =
0.1961
2.4706
-1.6667 0.1961z 2.4706 z 1.6667 z
p = U ( z) = + −
1.0000 z −1 z + 0.7 z + 0.5
-0.7000
-0.5000
k =
0

T=0.1 % arbitrary choice for this example

Uz=tf([.1961 0],[1 -1],T)+tf([2.4706 0],[1 .7],T)-tf([1.6667 0],[1 .5],T)

Uz =
z^3 - 0.5 z^2 + 2.5e-05 z
-----------------------------
z^3 + 0.2 z^2 - 0.85 z - 0.35
Inverse z-Transform
z 2 ( z − 0.5)
Check answer using MATLAB (ztrans) U ( z) =
( z − 1)( z + 0.7)( z + 0.5)
u (k ) = 0.1961 + 2.4706(−0.7) k − 1.6667(−0.5) k
syms k z
uk=0.1961+2.4706*(-0.7)^k-1.6667*(-0.5)^k;
Uz=ztrans(uk);
pretty(simplify(Uz))

2
z (200000 z - 99994 z + 5)
----------------------------------
10000 (2 z + 1) (10 z + 7) (z - 1)

99,994/200,000 5/200,000 numerical errors

z 2 ( z − 0.49997) + 0.000025 z z 2 ( z − 0.5)

M ( z) = 
( z − 1)( z + 0.7)( z + 0.5) ( z − 1)( z + 0.7)( z + 0.5)
The Inverse z-Transform
1
e( k ) =
2 j  
E ( z )z k −1dz Inversion
Formula
=  [residues of E ( z ) z k −1 ]
poles of E ( z ) z k −1

𝚪 is any closed path in the z-plane that encloses all the

finite poles of E(z)zk-1 . There are two important cases:
k −1
Case 1: E ( z ) z has a simple pole at z = a
 residue of E ( z ) z k −1  = ( z − a) E ( z ) z k −1
z =a z =a

Case 2: E ( z ) z k −1 has a m'th order pole at z = a

1 d m −1
 residue of E ( z ) z k −1  = 
m −1 
( z − a ) m
E ( z ) z k −1

z =a (m − 1)! dz z =a
Inverse z-Transform
z 2 ( z − 0.5)
Example M ( z) =
( z − 1)( z + 0.7)( z + 0.5)

Case 1: M ( z ) z k −1 has simple poles

z ( z − 0.5) z k
(no pole
First find: M ( z ) z k −1 =
( z − 1)( z + 0.7)( z + 0.5) at z = 0)

The residue at z = 1 is
z ( z − 0.5) z k
res  M ( z ) z k −1  = ( z − 1) M ( z ) z k −1  =
z =a z =1 ( z + 0.7)( z + 0.5) z =1
1(0.5)(1k )
= = 1/ 5.1 = 0.196
1.7 *1.5
Inverse z-Transform
z 2 ( z − 0.5)
M ( z) =
The residue at z = −0.7 is ( z − 1)( z + 0.7)( z + 0.5)

z ( z − 0.5) z k
res  M ( z ) z k −1  = ( z − a ) M ( z ) z k −1 =
z =a z =a ( z − 1)( z + 0.5) z =−0.7
−0.7(−0.7 − 0.5)(−0.7) k
= = 2.471(−0.7) k
(−0.7 − 1)(−0.7 + 0.5)

The residue at z = −0.5 is

z ( z − 0.5) z k −0.5(−0.5 − 0.5)(−0.5) k

= = −1.667(−0.5) k
( z − 1)( z + 0.7) z =−0.5
(−0.5 − 1)(−0.5 + 0.7)

Summing the residues:

m(k ) = 0.196 + 2.471(−0.7) k − 1.667(−0.5) k

Inverse z-Transform
1
Example E( z) =
( z − 1)( z − 2)
Case 1: E ( z ) z k −1 has simple poles and a pole at z = 0
k −1
z
E ( z ) z k −1 = , k 0
( z − 1)( z − 2)
Two sub-cases: 1
(1) k = 0 → E ( z ) z k −1 = (three poles )
z ( z − 1)( z − 2)
 1 
e(0) =   res. of 
@ z = 0,1,2  z ( z − 1)( z − 2) 
1 1 1 1 1
= + + = −1+ = 0
( z − 1)( z − 2) z =0 z ( z − 2) z =1 z ( z − 1) z = 2 2 2
Inverse z-Transform
(2) k > 0, no pole at z = 0.
z k −1
res [ E ( z ) z k −1 ] = ( z − 1) = −1
z =1 ( z − 1)( z − 2) z =1
k −1
z
res [ E ( z ) z k −1 ] = ( z − 2) = 2k −1
z =2 ( z − 1)( z − 2) z = 2

e(k ) = −1 + 2k −1 , k  0
e(0) = 0
Inverse z-Transform
1
E( z) =
Using MATLAB ( z − 1)( z − 2)
syms k z
Ez=1/((z-1)*(z-2));
ek=iztrans(Ez,k);
pretty(ek)

ek =
k
2 kroneckerDelta(k, 0)
-- + -------------------- - 1
2 2

e(0) = 0
k −1 The same result
e(k ) = −1 + 2 , k 0
Inverse z-Transform
Case 2: E ( z ) z k −1 has repeated poles at z = a

Example
z
E( z) =
( z − 1) 2
then q −1
1 d
e( k ) = 
q −1 
( z − a ) q
E ( z ) z k −1

(q − 1)! dz z =a

1 d 2−1  z k −1 
= 2 −1 
( z − 1) 2
z 
(2 − 1)! dz  ( z − 1) 2
 z =1
d k
= ( z ) = k z k −1 = k
dz z =1
z =1
Inverse z-Transform
Case 2: Using MATLAB
z
Example E( z) =
( z − 1) 2

syms k z
Ez=z/(z-1)^2;
ek=iztrans(Ez,k)

ek =
k
Discrete Convolution
E1 ( z ) E2 ( z ) E( z) E ( z ) = E1 ( z ) E2 ( z )

Expanding
E ( z ) = [e1 (0) + e1 (1) z −1 + e1 (2) z −2 + ]  [e2 (0) + e2 (1) z −1 + e2 (2) z −2 + ]
Group by powers of z
E ( z ) = e1 (0)e2 (0) + [(e1 (0)e2 (1) + e1 (1)e2 (0))]z −1 +
[e1 (0)e2 (2) + e1 (1)e2 (1) + e1 (2)e2 (0)]z −2 +
= e(0) + e(1) z −1 + e(2) z −2 +
k k
e(k ) =  e1 (n)e2 (k − n) =  e1 (k − n)e2 (n)
n =0 n =0

e k 1
E1 z E2 z e1 k  e2 k

Convolution in discrete time is equivalent to multiplication in

the z-domain!
Discrete Convolution
z 2 ( z − 0.5)
Example E( z) = = E1 ( z ) * E2 ( z )
( z − 1)( z + 0.5)( z + 0.7)
z ( z − 0.5) −5 z 6z
E1 ( z ) = = +
( z + 0.5)( z + 0.7) z + 0.5 z + 0.7
e1 (k ) = −5(−0.5) k + 6(−0.7) k
e1 (0) = 1.00
e1 (1) = −1.70
e1 (2) = 1.69
e1 (3) = −1.433

z
E2 ( z ) = e2 (k ) = {1,1,1, }
z −1
Discrete Convolution
k
e1 (0) = 1.0
e(k ) =  e1 (k − n)e2 (n)
n =0 e1 (1) = −1.70
e1 (2) = 1.69
For k = 3
3
e1 (3) = −1.433
e(3) =  e1 (3 − n)e2 (n) e2 (n) = 1.0
n =0

e(3) = e1 (3)e2 (0) + e1 (2)e2 (1) + e1 (1)e2 (2) + e1 (0)e2 (3)

= (−1.433)(1) + (1.69)(1) − 1.70(1) + 1(1)
= −0.443
Discrete Convolution
Example Discrete Convolution (conv in MATLAB®)
e1 (k ) = −5(−0.5) k + 6(−0.7) k e1 ( k ) = {1.00, –1.70, 1.67, –1.455, ...)
e2 (k ) = {1.00, 1.00, 1.00, 1.00, ...)

for n=1:1:20; % n must be >0, k=n-1, t=(n-1)T = {0,T,2T,…}

e1k(n)=-5*(-.5)^(n-1)+6*(-.7)^(n-1);
e2k(n)=1;
end
ek=conv(e1k,e2k) Value on previous slide
ans =
Columns 1 through 6
1.0000 -0.7000 0.9900 -0.4430 0.6851 -0.1671
Columns 7 through 12
0.4607 0.0056 0.3320 0.0996 0.2642 0.1480