Chapter 4

Stability: z-Domain &

Time-Domain Analysis

EE4L005 DIGITAL CONTROL SYSTEMS

N. C. Sahoo
 Introduction
• The concept of stability plays an important role in
the analysis and design of control systems. An
unstable system is virtually useless.
• In many situations, we may start out with an
unstable system with one of the design objectives
being to achieve stability.
• Even linear systems have many different types of
stability, but to introduce the concept, we can state
that the stability is the condition wherein the system
output must be bounded for any bounded input.
• As the outputs of discrete-time control systems are
usually functions of continuous time, it is necessary to
evaluate the system performance in time-domain.
• However, the outputs are measured only at sampling
instants. Depending upon the chosen sampling
period and its relation to the time-constants of the
system, the discrete-time representation may or may
not be accurate.

• In other words, there may be large discrepancy

between the true output c(t ) and the fictitiously sampled
signal c* (t ) , so that the latter is not a realistic
representation of the behavior of the system behavior.
• Similar to continuous-time control systems, the time
response of discrete-time systems are also
characterized by maximum overshoot, rise time,
settling time etc.
 • Analysis and design methods using these time-
domain specifications are best carried out using the
complex s- or z-domain.

• Analytical tools such as root-locus diagrams are also

applicable to discrete-time systems.
 Stability
• Zero-State Response: The output response of a
discrete-time system that is due to the input only is
called the zero-state response; all the initial
conditions of the system are set to zero.
• Zero-Input Response: The output response of a
discrete-time system that is due to the initial
conditions only is called zero-input response; all the
inputs of the system are set to zero.
• From the principle of superposition, when a system is
subjected to both inputs and initial conditions, the
total output response is:
Total response =
zero-state response + zero-input response
• BIBO Stability: A system is bounded-input-bounded-
output (BIBO) stable if, for any bounded input, the
output is also bounded.
• Zero-Input Stability: A system is zero-input stable or
simply stable if the zero-input response of the
system, subject to finite initial conditions, reaches
zero as time approaches infinity; otherwise, the
system is unstable.
Mathematically, zero-input stability requires two
conditions to be satisfied. They are:
c( k )  M  
lim c(k )  0
k 
 Theorem: For a linear discrete-time system, the BIBO
and zero-input stability require that the roots of the
characteristic equation be inside the unit circle z  1 in
the z-plane.
Proof: For a system with input u (k ) and output c(k ) ,
k
c(k )   h(k  m)u (k )
m 0

where h(k ) is the pulse response of the system.

k k
 c(k )   h(k  m)u (k )   h(k  m) u (k )
m 0 m 0
Since the input is assumed to be bounded, the stability
condition requires that
Roots should be inside

k 0
h( k )  
unit circle in z-plane.
Let n characteristic equation roots of the system be
represented by zi , i  1,2,, n .
If m of the n roots are simple, and the rest are of multiple
order, then
m n
c(k )   Ki ( zi )k   Ki k i ( zi ) k
i 1 i m1
where K j are constants.
Since the term i control the response as k   , to
k
( z )
satisfy the two conditions for zero-input stability, the
magnitude of zi must be less than one.

• In other words, the roots of the characteristic

equation must all be inside the unit circle z  1 in
the z-plane.
 Stable and unstable
regions in z-plane

• If any characteristic equation root is not inside the

unit circle, the system is unstable.

• For practical reasons, we often refer to the situation

when the characteristic equation has one simple root
on the unit circle as marginally stable or marginally
unstable.
 Example:

10( z  0.5) All poles are inside the unit

G( z) 
z ( z  0.15)( z  0.82) circle. So, the system is stable.

z One pole is outside the unit

G( z) 
( z  1.5)( z  0.5) circle. So, the system is unstable.

K ( z  1) One pole is on the unit circle.

G( z ) 
( z  1)( z 2  0.25 z  0.1) So, the system is marginally
stable.
Stability Test for Discrete-Time Systems
• The Routh-Hurwitz criterion, which is used for
stability test of linear continuous-time systems with
respect to characteristic equation roots in s-plane,
cannot be applied directly to z-plane due to the
difference in stability boundary.
• A bilinear transformation is used to transform the
unit-circle in z-plane into the imaginary axis in
another complex plane so that Routh-Hurwitz
criterion can be applied to discrete-time systems.

• There is also a direct method (Jury’s Test) for testing

the location of roots with respect to unit circle in z-
plane.
A. Bilinear Transformation Method – Extension of
Routh-Hurwitz Criterion:
• The Routh-Hurwitz criterion, or the Routh tabulation
of the coefficients of characteristic equation, is a
convenient way of finding location of the roots with
respect to left and right halves of the s-plane for
stability tests in linear continuous-time systems.

The z-transformation, i.e., z   sT , transforms the

imaginary axis of s-plane onto the unit circle z  1 in
z-plane; thus Routh-Hurwitz criterion is not effective for
discrete-time systems.
• We can attempt to transform the unit circle in z-plane
back to the imaginary axis in another complex plane,
but we cannot use the inverse z-transform, since it
would transform an algebraic characteristic equation
in z into a transcendental equation of s, which still
cannot be tested with the Routh-Hurwitz criterion.
• Transformations that are algebraic and transform
circles in z-plane onto vertical lines in a complex-
variable plane, say the r-plane, are of the following
form:
ar  b
z
cr  d
where a, b, c, and d are real constants.
• One such transformation that transforms the interior
of the unit circle z  1 onto left-half of r-plane is:
r 1
z
r 1
which is referred to as the r-transformation.

Mapping of
unit circle in z-
plane onto the
imaginary axis
in r-plane by r-
transformation
• Thus, once the characteristic equation in z is
transformed into r-domain, the Routh-Hurwitz
criterion can again be applied to the equation in r
in normal manner.
Example:
Consider the characteristic equation of a discrete-time
control system given by
F ( z )  z 3  1.25 z 2  1.375 z  0.25  0
Transforming F (z) into the r-domain:
 1.875 r 3  3.875 r 2  4.875 r  1.125  0
• Since all the coefficients of transformed
characteristic Eq. do not all have the same sign, the
system has at least one root in the right half r-plane.
• This means that F (z) has at least one root outside the
unit circle in z-plane.
• Nevertheless, the Routh table for transformed
characteristic Eq. is given below.
r3  1.875 4.875
r2 3.875 1.125
r1 5.419 0
r0 1.125

• Since there is one sign change in the first

column, the transformed characteristic Eq. has
one root in right half r-plane, or F (z) has one root
outside the unit circle in z-plane.
Singular Cases:
• In Routh tabulation in s-domain, the following two
singular cases are often encountered.

1. The first element in any one row of the table is

zero, but the others are not, before the
tabulation is completed.
2. The elements in one row of the table are all
zeros, before the tabulation is completed.

• The first case is corrected by replacing the zero

element in the table by an arbitrarily small positive
number  , and then proceeding with the
calculation.
 • In the second singular case, it is an indication that
one or more of the following conditions exist:
(i)Pairs of real roots with opposite signs (symmetry of
roots about the jꞷ axis of s-plane)
(ii)Pairs of imaginary roots
(iii)Complex-conjugate roots forming symmetry about
origin of s-plane
• The auxiliary equation A(s)=0 is formed by using
the coefficients of the row just above the row of
zeros in Routh table. The auxiliary equation also
satisfies the characteristic equation, and the
tabulation is continued by replacing the row of
zeros by the coefficients of dA( s ) ds.
• For the first singular case in r-domain, the same
remedy is applied by replacing the zero element in
the first row by an arbitrary small number .
• For the second singular case in r-domain, where a
row of zeros occur in the Routh tabulation of the
equation in r, looking from the transformation
relationship between s and z and r, the following
conditions may exist:
1) Pairs of real roots in z-plane that are inverse
of each other; e.g., z1  2 and z 2  1 z1  0.5 .
2) Pairs of roots on the unit circle z  1 including
at z =1 and z = 1 simultaneously.
• The second condition corresponds to roots on the
imaginary axis in r-plane. The remedy to this
singular case is similar to that for s-domain. They
are:
 1) Form the auxiliary equation A(r )  0 using the
coefficients of the row in Routh table just above
the row of zeros.
2) Perform dA(r ) dr  0 .
3) Replace the row of zeros in the tabulation with
the coefficients of dA( r ) dr .
4) Continue with the Routh tabulation in usual
manner.
Example: Consider the equation
F ( z )  z 3  3.3z 2  3z  0.8  0
Transforming F (z) into r-plane,

8.1r 3  0.9r 2  0.9r  0.1  0

The Routh tabulation of transformed equation is:
r3 8 . 1  0 .9
r2 0 .9  0 . 1
r1 0 0
• There is a row of zeros and Routh test cannot be
continued further. Thus, the auxiliary equation is
2
formed by using the coefficients of r row:
A(r )  0.9r 2  0.1  0
• Taking the derivative of A(r ) with respect to r,
dA( r )
 1 .8 r
dr
• The coefficients of r 1 row are now filled with the
coefficients of dA(r ) dr and the remaining Routh
tabulation is:
 r1 1 .8 0
r0  0 .1
• Since one sign change occurs in the first column of
Routh table, the characteristic equation has one root
in the right half r-plane, or F (z ) has one root outside
unit circle in z-plane.
• Solving for the roots of the auxiliary equation, we
have r  0.333 and  0.333 , which are two of the
roots of the transformed characteristic equation.
These two roots correspond to two roots of F (z)
at z  2.0 and  0.5 ,respectively. The other root
of F (z) is at z  0.8 .
 B. Jury’s Stability Test
• One of the first direct methods devised for testing the
location of roots of a polynomial in z with respect to
the unit circle in z-plane is the Schur-Cohn
criterion.
• This criterion gives the necessary and sufficient
conditions for the roots to lie inside the unit circle in
terms of the sign of Schur-Cohn determinants.

• But, this criterion is very cumbersome for equations

higher than second order.

• Another much simplified test procedure was

designed by E. I. Jury. This test, Jury’s stability
test, is widely used and is discussed here.
Jury’s Test Procedure:

Given the n-th order characteristic equation:

n 1
F ( z )  an z  an1 z
n
   a2 z  a1 z  a0  0
2

where a0 , a1 ,,a n are real coefficients.

Assume that an is positive, or that it can be made positive

by changing the signs of all the coefficients of F (z) .

• The Jury’s Table is made using the coefficients of

the polynomial F (z) .
 Row z0 z1 z2  z n-k  z n1 zn
1 a0 a1 a2  a nk  a n 1 an
2 an a n 1 a n2  ak  a1 a0
3 b0 b1 b2  bn k  bn 1
4 bn 1 bn 2 bn 3  bk  b0
5 c0 c1 c2  cn2
6 cn2 c n 3 cn4  c0
    
2n  5 p0 p1 p2 p3
2n  4 p3 p2 p1 p0
2n  3 q0 q1 q2

• Note that the elements of (2k  2) -th row (k  0,1,2,)

consist of the coefficients of the (2k  1)-th row written
in reverse order.
• The elements in the Jury’s table are defined as:
a0 ank b0 bn1k c0 c n  2 k
bk  ck  dk 
an ak bn1 bk cn  2 ck

p0 p3 p0 p1
q0  q2 
p3 p0 p3 p2

• The necessary and sufficient conditions for the

polynomial F (z) to have no roots on or outside the
unit circle in z-plane are:
F (1)  0
 0 ; n even
F (1)
  0 ; n odd
 a0  a n 
b0  bn1  
c0  c n  2 
 (n  1) constraints
d 0  d n 3 
 

q0  q 2 
• For 2nd order system, n  2 , the Jury tabulation
contains only one row. Therefore, the requirements
listed in the above equations are reduced to
F (1)  0
F (1)  0
a0  a 2
Example: Consider the characteristic equation
F ( z )  z 2  0.25  ( z  0.5)( z  0.5)  0

The coefficients are: a2  1 a1  0 a0  0.25

The stability tests for this 2nd order system are:
(i) F (1)  1  0.25  0.75  0
(ii) F (1)  1  0.25  0.75  0
(iii) a0  0.25 , a2  1 ; So, a0  a2 .

Thus, the system is stable.

Example: F ( z )  z 3  1.25 z 2  1.375 z  0.25  0

Thus, a3  1 a2  1.25 a1  1.375 a0  0.25

Checking the first three conditions,
(i) F (1)  1  1.25  1.375  0.25  1.875  0
(ii) F (1)  1  1.25  1.375  0.25  1.125  0
(iii) a0  0.25  1
Since F (1)  0 , not all the roots are inside the unit circle
and the system is unstable. There is no need to carry
out the Jury tabulation.
 Time–Domain Performance
Specifications
• The performance of a discrete-time control system in
time domain is often measured by applying a test
signal such as unit-step function to the system input.

• For linear systems, the unit-step input provides

valuable information on transient and steady state
behavior of the system.
• In fact, maximum overshoot, delay time, rise time
and settling time are all defined for unit-step input.
The response of s system to unit-step input is called
unit-step response.
Typical unit-step response of a system
• Notice that the unit-step response of a continuous-
time system could also be of the same form as that
of a discrete-time system.
• Thus, the time-domain specifications apply equally
to both types of systems. These specifications are
described below.
Maximum Overshoot:
cmax = Maximum value of the output c(t )
c ss = Steady state value of c(t )

Maximum overshoot = cmax  c ss

cmax  c ss
Percent maximum overshoot =  100
c ss
Peak Time:
Tmax = Time at which cmax occurs
Delay Time:
Td =Time required for unit-step response to reach
50% of its final value
Rise Time:
Tr =Time required for unit-step response to rise from
10% to 90% of its final value
Settling Time:
Ts =Time required for unit-step response to reach
and stay within 5% of its final value
Typical unit-step response of a digital control
system and its sampled-data representation
• It is noted that the maximum value of the sampled
*
output c * (t ) , c max , is always less than or equal to cmax

• If the sampling period is sufficiently small, the

discrete-time response gives a sufficiently accurate
representation of true response, and the discrepancy
*
between cmax and c max is not significant.

• In general, if the sampling period is too large relative

to the oscillations of the response, c * (t ) may be
entirely erroneous.
 Prototype 2nd Order System
• Although true second-order systems rarely exist,
most of the s-domain performance parameters are
defined for such a system.
• The study of second-order system is important
since many higher-order systems can be
approximated by a second-order model if the
higher-order poles provide negligible contribution to
the transient response.

C ( s)  n2
G(s)  
E ( s ) s ( s  2n )
 C ( s) n
2
M ( s)   2
R( s) s  2n s  n2
where  = damping ratio and  n = undamped natural
frequency. The characteristic equation of the prototype
second-order system is: s 2  2 n s   n2  0
The roots of the characteristic equation are:
s1 , s2  n  jn 1   2
Comparison between Continuous-time and Discrete-
time system responses
• The objective is to highlight the effects of sampling
and how sampling frequency is chosen with respect
to the dynamics of the system.
Example:

Simplified space vehicle control system

 • The system controls the altitude of space vehicle
in one dimension. The parameters are:
Position sensor gain: K P  1.65  10 6
Rate sensor gain: K R  3.71  10 5
Amplifier gain: K = variable
Moment of inertia: J v  41822
The forward path transfer function is:
C ( s) KK P 39.453 K
G ( s)   
E ( s ) s ( J v s  K R ) s ( s  8.871)

C ( s) G( s) KK P 39.453 K
   2
R( s) 1  G( s) J v s  K R s  KK P s  8.871s  39.453 K
2
Characteristic equation is: s 2  8.871s  39.453 K  0
Comparing this with prototype second-order system,

n  39.453 K rad/s and   8.871 (2 n )

• Since the system is of second order, the

characteristic equation roots always lie in the left half
s-plane so long as all the system parameters are
positive and finite.

• Thus, the continuous-time system will always be

asymptotically stable.
Discrete-Time System:

• Next, let us consider that the above system is

subjected to sampled-data control and the input is
processed by a S/H device (sampling period = T).
• The transfer function of the process is:
KK P J v
G p (s) 
s(s  K R J v )
• The open-loop transfer function is:
C (s) 1   Ts KK P J v
G(s)   Gho ( s )G p ( s ) 
E (s) s s( s  K R J v )
Thus,
1 KK P  1 Jv Jv 
G( z )  (1  z )  2   
KR  s K R s K R (s  K R / J v ) 

1 KK P  Tz Jv z Jv z 
 (1  z )     K RT / J v 
KR  ( z  1)
2
K R ( z  1) K R ( z   )
• The closed-loop characteristic equation of this
system is:
z  a1 z  a0  0
2

KK P
a1  2 (TK R  J v  J v  K RT / J v )  (1    K RT / J v )
KR
 0.000012K (3.7110 T  41822  41822
5 8.871T
) 1   8.871T

 K RT / J v KK P 
a0    2  J v  (TK R  J v )  K RT / J v 
KR
 8.871T
 0.000012K[41822  (3.7110 T  41822) 5 8.871T
]
• Applying Jury’s stability test:
(i) a0  1 
8.871T 8.871T
  0.000012 K [41822  (3.7110 T  41822 ) 5
] 1
 (ii) F (1)  1  a1  a0  0  1   8.871T  0

(iii) F (1)  1  a1  a0  0 
8.871T 8.871T
2(1   )  0.000012 K [83644 (1   )
 3.7110 5 T (1   8.871T )]  0
• Since T is always positive, condition (ii) is always
satisfied.
• Stability boundaries formed by conditions (i) and
(iii) are shown below in K vs.T plane.
• Clearly for small values of T, the stability region is
governed by condition (i) and for large values of T,
condition (iii) plays the dominant role.
Stability boundaries in the parameter plane of digital
space vehicle control system
Comments on selection of K and T:
• This Figure offers clues to selection of K and T for
stable operation. However, for satisfactory operation,
more information is needed for selection of proper
values within the stable limits.
• From the knowledge of continuous-time systems, it
is well known that increasing K would reduce the
damping ratio, increase the maximum overshoot,
increase the natural undamped frequency and thus
bandwidth, and reduce steady state error if it is finite
and nonzero.

• These properties are also true for discrete-time

system.
• As for the sampling period T, Figure shows that for a
given K, increasing T will cause instability. On the
other hand, a smaller sampling period would require
a faster clock rate for the digital computer, which
means more complex hardware and cost.
• For illustration purposes, say K=1. From Figure,
maximum sampling period for stable operation is
0.674s.
• The unit-step response with T = 0.1, 0.6, and
0.674s are shown below. These responses are
shown as continuous functions by linking the
sampled responses by smooth curves, since the
true output of digital system is a function of t.
Unit-step responses for continuous-time and digital space
vehicle control system with K = 1
• The properties of unit-step responses for several
sampling periods with the ratios to  n and
bandwidth (BW) of the continuous-time system are
shown in Table below.
• The results in Table show that BW or  n of the
continuous-time system can be used as a guideline
for selection of sampling period for digital control
system.
• In this example, the performance of discrete system
will be close to that of continuous system if the
sampling frequency  s is at least 15 times the BW
or  n .
• However, for safety, the minimum sampling
frequency may be chosen to be 20 times the BW.

• It must be noted that for K = 1, BW = n . In general,

BW is often greater than  n , so that the former is
often chosen as the reference guideline.
• A better perspective can be gained from the root-
locus of the digital control system for various values
of T in z-plane, as shown below.
• Positions of the roots for incremental values of T
are connected by continuous curves to show that,
as T varies continuously, the roots will trace out
continuous trajectories.
• The root loci show that, as T increases, the roots
move from the first and fourth quadrants into second
and third quadrants in z-plane.
• From the mapping characteristics between s- and z-
plane, this movement of roots corresponds to an
increase in the frequency of the digital control
system.
Root loci of the characteristic equation of the discrete
system when T varies (K = 1)
• When the roots become closer to the negative real
axis in z-plane, the response becomes oscillatory.

• When the roots are real and negative, the response

will be oscillatory with alternate positive and negative
values, as shown in step responses for T = 0.674s.

• On the other hand, as T becomes smaller, the roots

move closer to z = 1 in the first and fourth quadrants.

Conclusion: The roots of the digital control system

should lie in the first and fourth quadrants and very
close to z = 1 point for the digital system to emulate
the continuous-time system.
 • Now, let us take a larger K so that continuous
system is more oscillatory. K = 6; the continuous
system has a maximum overshoot of 0.389. BW =
22.49 rad/s and  n  15.38 rad/s.

• From stability analysis, the discrete system with K = 6

is stable for T < 0.0857s.
• The unit-step responses of continuous-time system
and discrete-time system with T = 0.0558s and
0.0857s are shown below.
• The performance comparison of discrete system with
various T are also shown in the Table after that.
Unit-step responses for K = 6
Performance comparison of discrete-time system for
various T (K = 6)
Root loci of the characteristic equation of discrete-time
system when T varies (K = 6)
 • The results in Table show that a sampling frequency
of at least 20 times the bandwidth of continuous
system is a good choice. For higher value of K,  n is
higher.
• From root locus, the stable range of T results in
characteristic equation roots that stay in the first and
fourth quadrants of z-plane. For T > 0.0857s, the
system is unstable.
Steady State Performance:
• The steady state performance of a continuous-time
system can be studied using final value theorem of
Laplace transform. Similarly, the steady state
performance of digital control system can be studied
by using final value theorem of z-transform.
• The steady state error for continuous-time space
vehicle control system is defined as
s2 (Jv s  KR )
ess  lim e(t )  lim sE ( s)  lim R( s )
t  s0 s0 J v s  K R s  KK P
2
• For the discrete-time system,
1 1 R( z )
*
ess  lim e(kT )  lim (1  z ) E ( z )  lim (1  z )
k  z 1 z 1 1  G( z)
where G(z) has been derived earlier.
• The comparison of steady state results in Tables
says that the steady-state errors for both systems
are identical for the same inputs.

• In general, the steady state errors for a digital

control system may also depend upon the sampling
period.
 Correlation of time-response with root
locations in s-plane and z-plane
 Complex conjugate roots on left half s-plane gives
exponentially decaying sinusoidal responses.
 Negative real axis roots correspond to monotonically
decaying responses.
 Simple conjugate roots on the imaginary axis gives
undamped constant-amplitude sinusoidal oscillations.
 Multiple-order roots on imaginary axis and roots in
right half s-plane give unstable responses.
• Although the relationship between s-plane and z-
plane has been thoroughly discussed before, the
sampling operation in digital system brings about
conditions that require special attention.
• For instance, when the sampling theorem is not
satisfied, the folding effect of sampling could distort
the true response of the system.

Poles of a second-order
continuous-time system
and its natural response
 • Now if the system is subjected to sampling with a
sampling frequency  s  21 , the sampling
operation generates an infinite number of poles in
the s-plane at s   1  j1  jns , n  0,  1,  2,.

• As shown below, the sampling operation folds the

poles back into the primary strip,   s / 2     s / 2
so that the net effect is equivalent to having a
system originally with poles at s   1  j ( s  1 ) .

• The sampled signal that demonstrates the effect

of frequency folding. The folding effect makes the
sampled system appear as if the frequency were
equal to  s  1 rather than 1 .
_

Pole locations in s- and z-plane

showing the effect of frequency folding
Root locations z-plane and corresponding time responses