# -

CHAPTER

25

Analysis of Sampled-Data Control Systems

The analysis of sampled-data control systems is complicated by the presence of both continuous and discontinuous (discrete) components in the control loop. Just as with continuous systems, a block diagram of the feedback control system can serve as the basis for analyzing the dynamic behavior of a digital system. The main focus of this chapter is the application of block diagram algebra to obtain the closed-loop transfer function and the analysis of the resulting characteristic equation to evaluate control system stability. However, care must be exercised in the development of the closed-loop transfer function. In particular, the rules for block multiplication used for continuous systems must be modified when sampling operations are included. Several techniques based on analysis of the characteristic equation of the closed-loop transfer function are presented which can be used efficiently to determine loop stability. Again, as with continuous systems, the stability of a sampled-data control system is a necessary condition for operability. 25.1 OPEN-lOOP BLOCK DIAGRAM ANALYSIS If G(s) is the transfer function between a continuous input and continuous output, the situation where both x(t) and y(t) are sampled is shown in Fig. 25.1. If the output is actually not physically sampled, a fictitious sampler can be inserted for purposes of analysis, as shown in Fig. 25.2. This procedure is employed because in z-transform analysis only values of the output at the sampling instants have any validity. It is usually assumed that both samplers act synchronously and have the same sampling period. The sampled input signal x"'(t) (see Eq. 24-1) is a series of impulses. Then X"'(s), the Laplace transform of x"'(t), has been shown in Eq. 243 to be
X"'(s) ~
11~0

x(ntlt)e

-/1;:"{5

(25-1)

or in z-transform notation,
X(z) ~
11=0

x(nLlt)Z-n

(25-2)

590

25.1 Open-Loop

Block Diagram Analysis

591

x(t)

)(~
tJ.t

y ~*(t)
Y*(s)

X(s)

X*(s)

Figure 25.1

Block diagram with sampled input and output signals.

r-- ~--~ I
"\.;
tJ.t
x(t)

Y *(t) y*(s)

)(
tJ.t

x*(t) X*(s)

I

X(s)

y~t) y(s)

Figure 25.2

The sampled input signal passes through the continuous process G(s), yielding a continuous output signal yet) (or Y(s)). Therefore, by the definition of the transfer function, yes)

= G(s)X*(s)

(25-3 )

Next consider the sampling of the output signal. As before, we define Y* (s) as the Laplace transform of y* (t), the impulse sequence obtained by sampling yet) (see Eq. 25-1). We will use an asterisk to denote the star transform, which indicates the impulse operator acting on the Laplace transform of a signal. It will be applicable to transfer functions as well as inputs and outputs. For example, to find the sampled signal Y*(s), the star transform is applied to both sides of Eq. 25-3 to obtain Y*(s)

= [G(s)X*(s)]*

= G*(s)X*(s)

(25-4)

Since X*(s) is a series of impulses, G(s)X*(s) represents a sequence of impulse responses for a continuous-time process. The star transform applied to this signal indicates the sampling of G(s)X*(s) to yield Y*(s). The proof of (25-4) using frequency-domain analysis has been given by Franklin and Powell [1, p. 86] and is a key result for block diagram analysis of sampled-data control systems. Note also that (25-4) is equivalent to Y(z)

= G(z)X(z)

(25-5)

where G(z) is the pulse transfer function and z = es':'t. This relation was derived using a time-domain analysis in Chapter 24 (see Eq. 24-77 and Fig. 24.3). Pulse Transfer Functions of Systems in Series Consider the two systems shown in Figs. 25.3 and 25.4 and derive the pulse transfer function Y(z)/X(z) for each system. For Fig. 25.3, yes) Consequently, application

= =

G1(s)GZ(s)X*(s) gives

(25-6)

of the star transform Y*(s)

[G1(s)GZ(s)]*X*(s)

(25-7)

Since vet), the output from Gj, is not sampled, the product of G1GZ operates on X*(s) to form Y*(s). Then from the definition of the z-transform it follows that Y(z) X(z)

=

Z [G1(s)GZ(s)]

= G1GZ(z)

(25-8)

592

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

x(t)

X(s)

A~
t:.t

t:.t-,--X _
I _
X*(s)

Y*(s)

y*(t)

~

y(t)

Y(s)

Figure 25.3

Two continuous

systems in series with sampled input and output signals.

x(t)

)(~
t:.t

r -X~ I
t:.t

Y*(s)

~

y(t)

X(s)

X*(s)

Y(s)

Figure 25.4

Two continuous

systems in series with a sampler in-between.

Note that
GjG2(z)

Eq. 25-8 continues use of the notation introduced on pg. 576 that ~ Z[Gj(s)G2(s)].j It is important to note that in general Gj(z)G2(z)

#

GjG2(z)

A special case of this equation order hold. Now consider

was derived in Ecj. 24-107, where G j was the zero-

the series of blocks shown in Fig. 25.4. Here yes)

=

G2(s)V*(s)

(25-9)

Using Eq. 25-4,

V* (s )
Combining gives yes) Next apply the star transform

Gns)x*(s)

(25-10)

= G2(s)Gns)x*(s)

(25-11)

to both sides of the equation:

Y*(s) or in terms of z-transforms Y(z) Thus, in this case

=

G2(s)Gj(s)x*(s)

(25-12)

(25-13)

Y(z) X(z)

=

G2(z)Gj(z)

(25-14)

The term G2(z)Gj(z) indicates that the z-transforms of Gj and G2 are obtained separately, then multiplied. The result is independent of the order of multiplication.
EXAMPLE 25.1

Show that if Gj(s) and G2(s) are both first-order

models, G2(z)Gj(z)

# G2Gj(z).

25.1 Open-Loop

Block Diagram Analysis

593

Solution

Referring to Fig. 25.4, let
K1 K2

G1(s)

=

TIS

+ 1 and

G2(s)

= -T2S

+
K2

(25-15)

Then by use of Table 24.1, we obtain
G1(z)

=

K1

(1 -

e

_

1
1_ -

1
T2

M 'IZ-I)

Tl

and

G2(z)

(1 -

e-MIT2z-1)

and
K2K1
T2Tl(1 e-MIT2z-I)(1 e-MITlz-l)

(25-16)

Referring to Fig. 25.3, we find
K2K1 (T2S

+

l)(TIS

+ 1)

Using Table 24.1 again yields
G2G1(Z)

=

Z[

(T2S

+ K2Kl l)(TIS
K2Kl(e-MIT2 -

+

1) ]
e-il.tITl)z-I e-MITlz-l)

(25-17)

(T2 -

Tl)(l

e-MIT2z-1)(1

Therefore, by inspection of (25-16) and (25-17),
G2(Z)G1(z)
7" G2G1(Z)

(25-18)

It should be clear on physical grounds that the two systems in Figs. 25.3 and 25.4 will have different pulse transfer functions betweeen X and Y. In Fig. 25.3 the input to the second transfer function is a continuous signal, v(t), while in Fig. 25.4 the input is a sampled signal, v* (t), which is a series of pulses. Thus, we would expect the output signals yet) and probably their sampled counterparts y*(t) to be quite different.

EXAMPLE

25.2

Examine the influence of the zero-order hold on the continuous output responses y(t) in Figs. 25.3 and 25.4 by letting G1(s) = H(s) = (1 - e-s::'t)/s and G2(s) = K/(TS + 1). The input signal x(t) is a unit step input.
Solution

Figure 25.3 represents the normal usage of the ZOH cascaded with a process transfer function. The step input x(t), after being sampled to form a series of unit impulses x*(t), is then sent through the zero-order hold, which reconstructs the unit step function again. In mathematical terms,
V(s)

=

G1(s)X*(s)

(25-19)

594

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

K
y

Figure 25.5
Time

Response of a first-order system plus zero-order hold to a train of unit impulses (Example 25.2).

Because X(z)

1/(1 -

Z-I)

and Z-I =

e-SC1I,

then

1V(s) ----

e-sM

1
---

1
s
(25-20)

-

s

1 - e -sM

Therefore,
yes)

=

Gz(s) V(s)

K
TS

1 + 1S

(25-21)

which is the transform for the step response of a first-order system. The corresponding transient response is shown in Fig. 25.5. In this case the unsampled output is independent of I1t because the input is a step function. Next apply the same analysis to Fig. 25.4. This type of configuration would never be implemented in a process control application and is presented here only to illustrate the properties of the star transform. Equation 25-11 gives the appropriate expression for yes); substituting for Gz, Gf, and X* yields (25-22) The sampled version of the hold circuit was calculated in Eq. 24-106, namely
H*(s)

= H(z)

= 1

(25-23)

The sampled step input (l/s)* is simply a series of unit impulses. Equation 25-22 therefore becomes
yes)

= -- + 1 (1 + e-st:.t + TS

K

e-25M

+ ... )

(25-24)

Inversion of Y(s) to the time domain yields the unit impulse response for the output signal yet). At each sampling instant the impulse is repeated and causes a sudden increase (jump) in the value of yet), using the most recent value of yet) as the initial condition. The resulting time-domain response is shown in Fig. 25.6.

y

KIT

0_

°
Time

Figure 25.6

Response of a first-order system without hold device to a train of unit impulses (Example 25.2).

25.1 Open-Loop

Block Diagram Analysis

595

EXAMPLE

25.3

To further illustrate the mathematical effect of the sampling operation, compare the responses of the processes depicted in Figs. 25.3 and 25.4 for

1
s

Assume Xes) is a unit step input (= lis) and ideal samplers are used. Determine the solution yet) for O::s t::S 6.
Solution

is a series of impulses of unity strength. When this signal passes through G)(s) (an integrator), the output is a series of steps as shown in Fig. 25.7. This signal V(s) enters G2(s), a first-order transfer function, yielding Y)(s); y)(t) is a series of exponential rises. Figure 25.7 shows that the rises are cumulative, with a change in slope at each sampling instant. On the other hand, if a sampler is placed between G)(s) and G2(s), a different pattern emerges, as shown in Fig. 25.8. V*(s), the sampled signal based on V(s), is a series of impulses in the time domain. When V*(s) forces G2(s) to yield Y2(s), an exponential decay after each impulse is obtained for h(t) (see Fig. 25.8), although the initial value of each

In Fig. 25.3, x*(t)

x

: [
o

,

2

,

,
Time

4

,

,

6

,

1
2
Time

1
4

6

I

4,.,

6

6 4

2

v

2

00

4 2

o

o

2
Time

4

6

Figure 25.7

Responses for Example 25.3 based on Fig. 25.3.

596

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

J

I
4
Time

2

6

x'

:~ o

1
2
Time

1
4

6

I

6

v

4 2

Time

6
4

v*
2

o

o

2
Time

4

6

6
4 Y2 2 o

o
Time

4
6 Figure 25.8 Responses for Example 25.3 based on Fig. 25.4.

decay transient grows in a monotonic fashion. The calculation of each response is left as an exercise for the reader. Note that the sampled values of yet) will be different for each case. For example, at t = 5,
yj(5)
5.91

(Fig. 25.7) (Fig. 25.8)

yi(5)

6.00

Pulse Transfer Functions of Systems in Parallel

Consider the block diagrams in Figs. 25.9 and 25.10. Note that in Fig. 25.10 the sampler is not present in the lower branch. We will derive relations between C( z)

25.1 Open-Loop

Block Diagram Analysis

597

E(s)

r - ....x t,.t--"I
C(s)

C*(s)

Figure 25.9

Addition in a block diagram with two samplers.

E(s)

rI
E(s)

---""'"
C(s)

'v
At

-

C*(s) ~

Figure 25.10

Addition in a block diagram with one sampler removed.

and E(z) for each diagram. From Fig. 25.9, C(s) Cj(s)

+ +

C2(s)

(25-25)
G2(s)E*(s)

G1(s)E*(s)

+

(25-26) (25-27)

[Gj(s)

G2(s)]E*(s)

To find the sampled output signal, take the star transform:
C*(s)

= [Gj(s) + = [Gj(z) +

G2(s)]E*(s)

(25-28)

In terms of z-transforms,

C(z)

G2(z)]E(z)

(25-29)

Hence, the overall pulse transfer function between C and E is C(z)
E(z)

= Gj(z) +

G2(z)

(25-30)

Now suppose that one sampler is removed, as shown in Fig. 25.10, C(s) Cj(s) +
Gj(s)E*(s) C2(s)

(25-31)
G2(s)E(s)

+
+

(25-32)

Thus,
C*(s)

=

Gj(s)E*(s)

[G2(s)E(s)]*

(25-33)

Expressing the relation in z-transforms yields C(z)

=

Gj(z)E(z)

+

G2E(z)

(25-34)

Note that Eq. 25-34 implies that a pulse transfer function of the form C(z)/E(z) cannot be obtained for the system in Fig. 25.10. In contrast, the system in Fig. 25.9 does possess a pulse transfer function as defined in Equation 25-30. Because

598

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

of the useful properties of transfer functions, the arrangement in Fig. 25.9 is preferred for analysis. However, Fig. 25.10 is applicable to one important case in digital process control, namely load changes, where the continuous load affects the process but is not sampled. This topic is discussed next.
25.2 DEVELOPMENT OF CLOSED-LOOP TRANSFER FUNCTIONS

Using the block multiplication and addition rules given above, we can analyze the block diagram of a sampled-data feedback control system, such as that shown in Fig. 25.11. This diagram is a simpler version of Fig. 21.7. In this diagram the Laplace transform variable, s, is used to denote continuous signals and transfer functions, and the z-transform variable z indicates the sampled representation of the signals as well as pulse transfer functions. Figure 25.11 is analogous to the continuous-time block diagram for feedback control, Fig. 10.8. The discrete-time block labeled D(z) represents the pulse transfer function for the digital controller. The controller output passes through the zero-order hold H(s) before entering the process. This signal is the input to the final control element, possibly preceded by a transducer, which are combined in the block G v in Fig. 25.11. Other arrangements are possible, such as the stepping motor-valve combination discussed in Chapter 21. The error signal E (z) is the difference between the sampled set point, R(z) and the sampled measurement B(z). The closed-loop transfer function C(z)/ R(z) relates sampled values of the controlled variable C(z) to the sampled values of the set point R(zl- Measurement gain Km converts the physical set point R to an internal set point R, as discussed in Chapter 10. In other words, R(z) = KmR(z). Since e(t) is a continuous signal that may not be actually sampled, a fictitious sampler is shown for the purpose of discrete-time analysis. The analysis of the closed-loop system given below mainly utilizes Laplace transforms, followed by conversion to z-transforms.
Set-Point Change

Consider first the servomechanism case where measurement signal is related to P* by
B(s)

L(s)

-

O.

From Fig. 25.11 the (25-35)

=
B*(s)

H(s)Gu(s)Gp(s)Gm(s)P*(s)

Therefore,

B*(s)

can be found by applying the star transform:

=

[HGvGpGm]*

P*(s)

(25-36) H(s )Gu(s) Gp(s) Gm(s).

where [HGvGpGm]* denotes the star transform of the product The digital controller output P* (s) is related to E* (s) by
P*(s)

=

D*(s)E"(s)

(25-37)

Figure 25.11

Block diagram of a general digital feedback control system.

25.2 Development

of Closed-Loop

Transfer Functions

599

At the comparator, the error is a sampled signal:
E*(s)

=

R*(s)

-

B*(s) B*(s)

(25-38)

Now substitute (25-38) for
P*(s)

E*(s)

in (25-37) and then replace
B*(s)] [HGvGpGm]*

by (25-36):
(25-39)

=
=

D*(s)[R*(s) D*(s)[R*(s)

P*(s)]

(25-40)

Rearranging and solving for P* (s) gives
P* _

(s) Recognize that C(s) is related to C(s)

1+
P*(s)

D*(s)R*(s) [HGvGpGm]*D*(s)

(25-41)

by
(25-42)

=

H(s)Gu(s)Gp(s)P*(s)

Using the star transform for the fictitious sampler,
(25-43)

Therefore,
" [HGvGp]* D*(s)R*(s) D*(s)

C' (s) = 1 + Substituting for the setpoint, function:
R * (s)

(25-44)

[HGvGpGm]* KmR*

=

(s), we obtain the closed-loop transfer

C* (s)
R*(s)

(25-45)

In z-transform notation,

C(z)
R(z)

HGuGp(z)KmD(z)

1+

(25-46)

HGvGpGmCz)D(z)

The characteristic equation for the closed-loop control system is obtained by setting the denominator of (25-46) equal to zero: 1 +
HG(z)D(z)

=0

(25-47)

where HG(z) is defined as HGvGpGm(z). The roots of the characteristic equatior determine the stability of the sampled-data control system, as discussed in Sect.
25.3 .•

A special case of (25-46) occurs when Gm(s) = Km. Because the z-transforrr is a linear operator, the constant can be simply incorporated in it. Then C(z)
R(z) HGvGp(z) KmD(z) HG(z) D(z) D(z)

1+

HGvGp(z)KmD(z)

1+

HG(z)

(25-48:

where HG(z) is HGvGpKm(z). Note the similarity to the closed-loop transfeJ function defined for continuous-time systems in Chapter 10, Eq. 10-37.

The regulator case (where R(s) = 0) can be derived using a procedure similar tc the one above. For simplicity consider the case when Gm(s) = Km. Equatior

600

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

25-35 must be modified to include the effect of the load change:
B(s)

=

H(s)Gv(s)Gp(s)KmP*(s)

+
+

GL(s)KmL(s)

(25-49)

Applying the star transform yields
B*(s)

=

Km[HGvGp]*P*(s)

Km[GLL]*

(25-50)

Because C(s) = then
C*(s) H(s)GuCs)Gp(s)P*(s)

+ G L(s)L(s) +
[GLL]*

(25-51)

=

[HGvGp]*P*(s)

(25-52)

= - D*(s)B*(s), we can multiply (25-50) by - D*(s) and solve for Next, substitute that result for P*(s) in (25-52), which yields after rearrangement
Since
P*(s). P*(s)

(25-53) In terms of z-transforms, G LL(z)

1+

HG(z)D(z)

(25-54)

Because the disturbance input is not a sampled signal, we are unable to define a repreclosed-loop transfer function for the discrete-time case. The term GLL(z) sents the z-transform of the output from GL(s), caused by a continuous input L(s) (note that GLL(z) ¥- GL(z)L(z)). If L(s) is specified to be a particular input (e.g., L(s) = 1/s), then C(z) can be determined. Note, however, that the characteristic equation for (25-54) is the same as (25-48); thus the same stability analysis applies to both set-point and load changes.

EXAMPLE

25.4

For the system in Fig. 25.11, let
K e-8S
Gp(s) D(z)

K

= =

p 7pS+ Kc

1
H(s)

GL(s)

= __ L 1 7LS+
e -s!:.t

1= ---

(25-55)

s

where e = N 6.t and N is an integer. Derive C( z) for a unit step change in load when the set point is held constant, that is, L(s) = l/s and R(s) = O. Identify the characteristic equation.
Solution

To calculate the step response, derive the z-transforms indicated in Eq. 25-54. Thus,
Z[GL(S)L(S)]=Z(~!)=Z(KL- 7S

+ 1s

s

s + 117 KL)

(25-56)

25.2

Development

of Closed-Loop

Transfer Functions

601

Using Table 24.1,
KL(l
Z[GL(s)L(s)]

- a)z-I

=
e -M/T .

KLC

_1 rl

-

1 _ 1arl)

(1 - rl)(l

- az-I)

(25-57)

where a Similar!y,

=

Z[H(s)Gp(s)]

= Z (1 -

se-SM

TS Kp

+

1 e -as)

(25-58)

(25-59)

Kp

z -N-I

1 _ az -I
(25-60)

1-

a

Substituting (25-57) and (25-60) into (25-54) with D(z) gives
KL(1 - a)z-I

C(z)

(1 - z-I)(1 -

az-I) a)

(25-61)

1+
After rearrangement, C
(z)

Z-N-I

1-

KcKp(1

az-I

= 1 - (1 + a)z-I

+

az-2

+

KL(l
KcKp(1

- a)z-I
-

a)z-N-I

- KcKp(l

-

a)z-N-2 (25-62)

The characteristic equation is obtained by setting the denominator polynomial in (25-62) equal to zero. Note that the order of the characteristic equation depends on the magnitude of the time delay N. For example,
N Order of Polynomial

o

2
3
4

1
2 etc.

Because N is assumed to be an integer, the time-delay term merely increases the order of the polynomial. Contrast this situation to that for a continuous-time closedloop system, where a time delay yields a transcendental function in the denominator (see Chapter 11). The implication here is that the number of roots of Eq. 25-62 is bounded (equal to the order of the polynomial), while such is not the case for a continuous-time system. In addition, special techniques exist for analyzing such polynomial equations, as is discussed in the subsequent section.
EXAMPLE 25.5

Assuming proportional control of a first-order process, determine the closed-loop transfer function for a set-point change, and evaluate the effect of Kc on the

602

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

response. Let Km = 1, Gv = 1, Gp = Kp/(TpS + 1), and D = Kc- What is the final value of c(nllt) for R(z) = 1/(1 - Z-I), a unit step change in set point?
Solution

The process transfer function is
HG(z) Kp(l

- a)rl 1 - az - 1

(25-63)

where

a

=

e-M1,p.

The closed-loop transfer function from (25-47) is
KcKp(l C(z) R(z) _ a)rl

1+

az-I KcKp(l -~---- - a)z-I 1 - az - 1

1-

(25-64)

Rearranging yields C(z)
R(z)

KcKp(l

-

a)z-I

1+

(25-65)

[KcKp(l

- a) - a]z-I

If the control system is to be stable, the root of the denominator polynomial must have an absolute value less than unity (see Fig. 24.2). Otherwise C(z) will be unstable for a change in R(z). If the term [KcKp(l - a) - a] is less than one, this absolute value condition is satisfied. Solving for KcKp gives,
KK c
p

l+a <1-

a

(25-66)

Therefore, for digital control, the maximum value of Kc is a function of Kp and a; hence, it is a function of the sampling period. As Ilt approaches zero, a --,> 1, and the ratio (1 + a)/(l - a) increases, thus increasing the maximum gain for stability. In contrast to this situation, recall that for a continuous first-order process under proportional feedback control, Kc can be infinite without causing instability. Thus for digital control the sampling period !::.t must be selected judiciously so that the control loop remains stable, based on the inequality (25-66). For a stable closed-loop system the final value c(n!::.t) for a unit step change in R can be calculated by the Final Value Theorem, Eq. 25-21. Substituting into Eq. 25-65 and simplifying, we obtain

.
hm (1 - z z-->I

-I

_
)C(z) -

1+

KcKp(l - a) _ KcKp KcKp (1) - a - a Kc K p + 1

_

(2)-67)

Thus for a proportional controller the same offset occurs for both continuous-time and digital systems.
25.3 STABILITY OF SAMPLED-DATA CONTROL SYSTEMS

For a satisfactory control system design, the closed-loop system must be stable. By analogy to Section 11.1, the following definition of stability is normally used for sampled-data systems [2]:
Definition. A linear sampled-data system is stable if the output sequence {y(n!::.t)} is bounded for any bounded input sequence {x(n!::.t)}. Otherwise the system is said to be unstable.

25.3 Stability of Sampled-Data

Control Systems

603

Note that this definition contains no mention of the process response during the inters ample periods. Specifically, it is theoretically possible to have a system in which the continuous output is unbounded, even though a sampled sequence of the output remains bounded. An example of such an output is shown in Fig. 25.12. This type of hidden oscillation is rare in practice and can be readily detected by changing the sampling period or by using modified z-transforms [3]. The necessary and sufficient conditions for the stability of a linear sampleddata system can be expressed in two ways:

1. ~

Ig(nM)1

,,~o

<

00, or

2. G(z) has no poles on or outside the unit circle in the z-plane. In the above conditions G( z) is the pulse transfer function of the process and is the set of samples of the impulse response. The first condition implies that the impulse response {g(nL1t)} will be bounded for the total time span (n = 0 to 00). Long division can be used to calculate the value of2:~~o g(nL1t) but this may be time-consuming. It is preferable to use the second condition, which is related to the partial fraction expansion procedure discussed in Chapter 24. Recall in Eq. 24-49 we showed how the inverse z-transform depends on the poles of the transform (Ph pz, ... , Pm)' For a pulse transfer function G(z) and input X(z), the output signal can be calculated:
{g(mlt)} bo
Y(z)

+

b1z-1

+

bzz-z pzz

+ ... +
-1) ... (1

bkz-k
(25-68)

=

G(z)X(z)

=

11

- P1Z

-1)(1 -

- Pmz -1\ X(z)

Then, using partial fraction expansion (distinct roots are assumed),
Y(z) ---

+ --+ ... + ---1 - pZZ-1 1I - P1Z-1 + [contribution of input terms]

r1

rz

rm Pmz-1

(25-69)

where ri = partial fraction coefficient (residue). In a closed-loop system, the terms shown in Eq. 25-69 include the influence of the poles of the closed-loop transfer function, plus the contribution of the set-point and/or disturbance inputs (in brackets). The input terms appear separately from the transfer function terms in the partial fraction expansion and are assumed to be bounded. Hence, the stability of the loop can be determined by the transfer function denominator terms. As discussed in Chapter 24, the sampled output response corresponding to (25-69) is
c(nL1t)

=

r1(P1)n

+

rz(pz)"

+ ... +

rm(Pm)"

+ [input terms]

(25-70)

If the absolute value of any of the poles Pi is greater than unity, c(nL1t) ---i> x as n ---i> 00, thus violating the condition for stability. This is also true if the root is a complex number (in polar form, if P = Iplejw, then pl1 = Ipll1ejI1W).

o

ilt

2ilt

3.a.t 4.a.t 5ilt
Time

6.a.t
Figure 25.12

A hidden oscillation.

604

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

Stability Regions in the sand z Planes For both continuous-time and discrete-time systems, the condition of stability is determined by the roots of the characteristic equation (i.e., the system poles). For continuous systems the roots of the characteristic equation must have negative real parts. Graphically, the roots must lie in the left half of the s plane, shown in Fig. 25.13. Since . we can map 24.6 showed cases. Because stituting into (25-71)

the stability region in s onto the complex z plane. For example. Fig. the relation between poles in the sand z planes for several specific the s variable is a complex Eq. 25-71 gives
z

number we can write s

=

ex

+ jl3.

Sub-

=

eM(o:+j~)

(25-72) (25-73)

Thus a real value of s (13 = 0) maps to a real value of z. Similarly a complex value of s maps to a complex point in the z plane, although sometimes two complex poles in the s plane map to a single value in the z plane due to aliasing (cf. Fig. 24.6). From (25-73) the magnitude of z is (25-74)
eo:tl!

> <

1 for 1 for

ex ex ex

> <

0 0 lies in the left half of the s plane, then plane is less than unity in magnitude, stability in the z plane is the interior plane maps into the inside of the unit

and

erxtl! eo:M

= 1 for

= 0

Hence, if a root of a continuous-time system < 0 and the corresponding root in the z that is, Izl < 1. Graphically, the region of of the unit circle, since the left half of the s circle of the z plane.
ex

Stability Tests The roots of the characteristic equation can be computed using a nonlinear equation solver to determine if a pulse transfer function is stable. Alternatively, the frequency

I S plane I
Imag

I z plane
Imag

I

Unstable

Unstable Real

Stable Real

~
~~nitcirCle

--r:l

U

Figure 25.13

Stability regions in the sand z planes.

25.3 Stability of Sampled-Data

Control Systems

605

response of the pulse transfer function can be obtained (z = ejwt:.!), allowing Bode or Nyquist plots to be developed [2,3]. Such methods can be used for stability determination but do not readily lend themselves to traditional controller design (e.g., gain and phase margins). Either of these approaches can be time-consuming for high-order process models. Hence alternative techniques may be preferred for analyzing the stability of the characteristic equation. These techniques include 1. Modified Routh stability test 2. Jury's test 3. Schur-Cohn test

Modified Routh Stability Criterion. In Chapter 11, the Routh stability test was used to analyze the characteristic equation of a continuous-time system for unstable roots, those roots that lie in the right half of the complex s plane. For sampleddata systems, this test can also be applied to determine whether any roots of the characteristic equation (in z) lie outside the unit circle. If a simple rational transformation from z to s can be found that maps the interior of the unit circle into the left half of the complex plane, then the Routh criterion can be used directly with z-transforms. Such a transformation (or mapping) is provided by the bilinear transformation

1+
z=l_w

w

(25-75)

This complex mapping or transformation does not correspond to that for the original z-transform. It only approximates the original transformation z = eSt:.!, hence the use of w in place of s. However, the boundary separating the stable and unstable regions does map exactly [3,4] and consequently the stability determination is exact. To apply this stability test, first determine the characteristic equation of the sampled-data system (written as positive powers of z): fez) = a n zn + a n-l
zn-l

+ ...

+ a 1z + a 0 = 0

(25-76)

Note that if the time delay is an integer multiple of !:::..t, a polynomial in z will always result (see Example 25.4). Using the bilinear transformation, the characteristic equation is transformed to a function of w:
f(w)

= aII w" + aII-I

W"-1

+ ...

+ a 1w + a0 = 0

(25-77)

also yielding a polynomial in w, where ai = real constant coefficient (i = 0, 1, ... , n). Note that the af are not necessarily equal to the ai' The Routh test can then be applied directly to Eq. 25-77 to determine the number of roots that lie in the unstable right half of the w plane or, equivalently, how many roots of fez) lie outside the unit circle.

EXAMPLE

25.6

Consider the characteristic equation fez)

=

2Z3

+

Z2

+ z + 1= 0

(25-78)

Use the bilinear transformation to check stability.

606
Solution

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

Substitution

of the bilinear transformation

(25-75) into (25-78) yields

few)

=

2

el-w wy +
manipulation,
w3

+
7wz

eI-IV wy +

I-IV + C +

IV)

+ 1 = 0

(25-79)

After some algebraic

Eq. 25-79 becomes

+

+ 3w + 5 = 0

(25-80)

Examination of (25-80) shows that the necessary condition for stability is satisfied because all coefficients are positive. The corresponding Routh array is Row Row Row Row 1 2 3 4 1 7 16/7 5 3 5 0

Since no sign changes occur in column one, the system is stable. Equation 25-78 has actual roots of -0.739 and 0.119 ± 0.814j, all of which lie inside the unit circle. Jury's Stability Criteria. Jury's criteria provide a simple analytic test for stability of a closed-loop system. These criteria can be applied directly to a z polynomial and thus avoid application of the bilinear transform. Jury's method yields necessary and sufficient conditions for a polynomial in z with real coefficients to have all roots within the unit circle in the z plane. However, the number of unstable roots cannot be determined.
, aj, ao Consider the characteristic equation in (25-76). If a", all_j, all-z, ... are real, constant coefficients and all is positive, the Jury array is tabulated as shown in Table 25.1. The original coefficients {aJ are written in order of increasing subscript in the first row with the order reversed for the second row. Each succeeding pair of rows is calculated from the determinant relationships of Table 25.1, and has one less element than the previous pair. Additional rows are computed until 2n - 3 rows are obtained. The last row contains three elements, So, sj, Sz. The Jury criteria state that the necessary and sufficient conditions for the roots of Eq. 25-76 to lie within the unit circle in the z plane are

f(z = 1) > 0 f(z = -1) > 0 for n even f(z = -1) < 0 for n odd
and
laol

(25-81)

< all
> >
Ibll-11

Ibol
Ieol

Icll-zl
Idll-31

Idol>

n - 1 constraints

(25-82)

Isol

>

ISzl

A proof has been given by Jury and Blanchard [5]. Note that the first set of conditions, Eq. 25-81, should be checked before the Jury array is constructed. If (25-81) is not satisfied, the characteristic equation has

25.3 Stability of Sampled-Data

Control Systems

607

Table 25.1
Row

... zn-2 . Zo .. Cn-2 So bn-1 ... Co ao a2 a1 (3 an-2 (0 boao C2 C1 Cn-4 Cn-3 an-1 (2 (3 an-1 an (1 an-2 S1 a2 a1 S2 Z1 zn-1 (0 zn Z2 Co Cn-2 bn-1 bo an bn-2 b2 bn-3 b1

...

General Structure of the Jury Array

I

an ao

ak an-kl

So

=

(3 I (0

(0 (31

roots outside the unit circle and the Jury array need not be calculated. If Eq. 2581 is satisfied, the conditions in Eq. 25-82 must be investigated. If an inequality is violated, the rest of the array calculations are unnecessary.
EXAMPLE 25.7

Using the Jury stability criteria, determine if any roots of the characteristic equation -11825 -1-2 2 -3 -32 -31 ZO -3Zl 5 -1711 -2Z2 Z4 1 Z3 2 (25-83) fez) = 2Z4 - 3Z3 + 2Z2 - z + 1 = 0 The Row 9 > 0 of Eq. 25-81 are satisfied because conditions f( -1) = f(l) = 1 > 0 region. Solution

608

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

Applicable conditions of (25-82) are
ao bo
Co

<

a4 =? =? =?

1 8

<2
':/>

> b3 > C2

3> 1
11 (violation)

Thus, the system is unstable. By factoring, the roots are -0.152 ± 0.661j and 0.902 ± 0.523j. Note that the second set of roots lies outside the unit circle.

Jury's stability test is subject to special cases similar to those that develop in the application of the Routh criteria. In the tabulation of the Jury array, a singular case can arise by either having the first and last elements of a row be zero or having a complete row of zeros. In either case, the tabulation terminates prematurely. Calculations for singular cases have been described by Ogata[2]. While similar to Jury's criteria, the Schur-Cohn stability test is not attractive since roughly twice as many determinants must be calculated. Reference 2 contains more details on this method. The Routh and Jury tests are attractive because they are analytical in natute, allowing one to calculate stability regions for an unspecified controller gain. However, if only the roots of the characteristic equation are needed (as in Examples 25.6 and 25.7), using a root-finding computer program would be more efficient.
Schur-Cohn Criteria. SUMMARY

We have developed in this chapter the key equations for design of sampled-data control systems, namely the closed-loop transfer functions for both set-point and load changes and the characteristic equation. The characteristic equation can be used to perform a stability analysis of the digital control system, using several different methods. However, as with continuous-time systems, this result represents only a limit against which candidate controllers can be checked. The actual selection of controller settings is based on performance characteristics of the closed-loop system response, as discussed in the next chapter. Hence, stability is only one consideration, although it is an important one. REFERENCES
1. Franklin, G. F, and J. D. Powell, Digital Control of Dynamic Systems, Addison-Wesley, Reading, MA, 1980. 2. Ogata, K., Discrete-Time Control Systems, Prentice-Hall, Englewood Cliffs, NJ, 1987. 3. Luyben, W. L., Process Modeling, Simulation, and Control for Chemical Engineers, McGraw-Hill, New York, 1973. 4. Saucedo, R., and E. E. Schiring, Introduction to Continuous and Digital Control Systems, Macmillan. New York, 1968. 5. Jury, E. 1., and J. Blanchard, A Stability Test for Linear Discrete Systems in Table Form. IRE Proc. 49, 1947 (1961).

EXERCISES
25.1. To determine the effect of pole and zero locations on the performance characteristics of digital controllers, calculate and sketch the responses of the following controller

Exercises

609

transfer functions to a unit step change in the error signal e(nM). can you make concerning the effect of pole-zero locations?

What conclusions

1 = 1+.f.. 1 (b) D (z) = 1 + 0 1
(a) D(z) (c) D(z)

1 - OAr!

1 + OAr!
(d) D(z)
(e) D(z)

= 1 _ OAz

1 = =
z2 _
Z2

0.8: 1

(f)

D(z)

+ 0.8z + 0.25

25.2. Derive an expression for the response y(n~t) of each system shown below to a unit step change in x(t). Are the responses the same?

System I

~

~t ~

~(:JJIL~

~;

U*(s)

System II

25.3. For the blending system shown in the drawing, it is desired to control exit concentration C3 by adjusting flow rate Q2' Composition C2 is the primary load variable. Using the information given below, do the following: (a) Draw a block diagram for the controlled system. The symbols in the drawing should be used in the block diagram as much as possible. (b) Derive (or state) the transfer function for each block in the block diagram. It is not (c) Derive an expression for the closed-loop transfer function, C3(z)/R(z). necessary to substitute individual transfer functions.

I

\$
I

v

610

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

Available Information

i. The volume of liquid in the tank can be assumed to be constant since flow rates ql, q3, Q4, and Qs are constant and Q2 is quite small. 11. The density of all streams is constant at 90 Ib/ft3. Iii. The recycle line is 68.8 ft long and has an inside diameter of 4 in. iv. The concentration transmitter (CT) has negligible dynamics. The transmitter output changes linearly from 3 to 15 psi as the concentration changes from 0 to 40 lb solute/ft3. v. The concentration controller (CC) is a digital PI controller with a"gain of 0.5 and a reset time of 5 min. The controller output goes to a zero-order hold which outputs a signal to the control valve. The sampling period is 12 s. VI. The steady-state characteristic of the control valve is given by
Q2

= 4 - 1.15 ~

ft3/min

where p is the controller output signal in psi. After a sudden change in p, the flow through the valve Q2 reaches a new steady-state value in 32 s (after five time constants). V11.The tank is 6 ft in diameter and is perfectly mixed. Vll1. The nominal steady-state values are:
Ql

= 50 ft3/min 2 ft3/min 0.5 lb solute/ft3

Zh

= 82 ft3/min
100 ft3

v

80 lb solute/ft3

IX.

Feed composition c] is constant.

25.4. An irreversible, isothermal chemical reaction 2A ~ B is carried out in the continuous stirred-tank reactor shown in the drawing. Since reactor temperature is maintained constant, only the product concentration C3need be controlled. This composition is sampled automatically every two minutes and the sample analyzed by a gas chromatograph (GC). The GC retention time of 2 min is essentially a pure time delay. The digital output signal from the GC is then sent to a digital PI controller which adjusts the feed composition c] via a transducer and a valve. All flow rates are kept constant including the recycle flow rate Q2' (a) Draw a block diagram for the closed-loop system. (b) Derive a transfer function for each block using the information given below. (c) Derive an expression for the closed-loop transfer function, C3(z)/ R(z).

I
I I I I v
,..... __ 1_

-- I
c 3m *
J

Exercises

611

qj

values: 10 ft3/min, 150 ft3, Cj
q2

= 5 ft3/min,

q4

= 20 ft3/min

V
C4 11. The rate of reaction

=

0.3 lb-mole A/ft3

0.2 lb-mole A/ft3 is given by
r (lb-mole

A/ft3/min)

=

15.4

C32

where C3denotes the concentration of A in the reactor in Ib-mole/ft3. 111. The transport-velocity lag (i.e., time delay) associated with the recycle line is 2 mill. jV. The GC sample valve is located a few feet from the reactor. The sample loop dynamics are negligible but the GC cycle time is 2 min. v. The control valve-transducer combination provides the following steady-state relation between controller output p (mA) and feed composition Cj (mol A/ft3):
Cj

= 0.125

-vp-=4

After a step change in p occurs, Cj essentially reaches the new steady-state value in 32 s (five time constants). VI. The output signal from the GC varies linearly from 4 to 20 mA as the measured composition changes from 0 to 0.3 lb-mole A/ft3. V11. The reactor can be assumed to be perfectly mixed. V111.The density of each liquid stream i~ approximately the same.

25.5. A distillation column has an approximate transfer function between position x B and steam flow rate to the reboiler qs of
G
p

bottoms

com-

=~
(3s

+

1}2

A proportional digital controller is used to control XB by adjusting qs' A gas chromatograph with a time delay equal to the sampling period!::lt is used to measure XB' Find the ultimate gain of the system for values of I1t = 1,5, 10, and 20 min. Assume that Gv = Gm = 1. The process time constant is in minutes.

25.6. Product from a manufacturing

process enters the warehouse shown in the drawing at a variable rate of qi lb/h. Shipments from the warehouse are qo lb/h, which can also vary with time. The production control strategy is to adjust qi on the basis of the inventory W(lb) stored in the warehouse. Because it is expensive and inconvenient to monitor the warehouse contents on a continuous basis and to make continuous

adjustments, a digital control strategy is used. In particular, a direct-acting sampleddata controller is used in conjunction with a zero-order hold device. Thus, the control law relating the manipulated variable and the sampled measurements of the inventory W* for constant set point is given by
Qi(S)

= -Kc

H(s)W*(s)

where Kc is the controller gain, H(s) is the transfer function for the zero-order hold device, W* denotes samples of the inventory W, and all variables are in deviation form.
qj

(Product from manufacturing)

W
Warehouse qo (Shipments)

612

ANALYSIS

OF SAMPLED-DATA

CONTROL

SYSTEMS

(a) Draw a block diagram for the controlled process assuming that measurement dynamics are negligible. A consistent set of units for the gain of each block can be assumed. (b) Derive an expression for W( z) that indicates how the inventory changes when a disturbance occurs in qo. (c) Use the Final Value Theorem to derive an expression for the steady-state value of W that results from a unit step disturbance in qo. (d) For what values of Ke will the closed-loop system be stable? (Assume the sampling period is M = 1 hr) (e) For what values of Ke will the closed-loop response be oscillatory? (Assume
I1t

=

1 hr)

25.7. Determine whether the following two characteristic equations have any unstable roots: (a) f[(z) = 2 +
2z2 3Z3

+

5Z4

(b) f2(z) = 1 + 3.3r1
Use the modified Routh criterion.

+ 3r2 + 0.8z-3

25.8. A closed-loop control system under proportional-only digital control has

_ 1
Gm
-

_
HGp(z) -

0.2683r1 + (1 _ 0.2636r1)(1

0.4406z-2

- 0.7658r1)

Find the value of Routh criterion.

Kc

that causes the loop to become unstable. Use the modified

25.9. A process operating under proportional-only digital control has
Gp(S)

=

3

G m = 1 and

M = 1

Using Jury's test, determine if the control system is stable. Check your results by calculation of the response to a set-point change. 25.10. Determine how the maximum allowable digital controller gain for stability varies as a function of I1t for the following system:

1
Gp(s) Ge

= (s + =
Ke

1)(5s

+ 1)

Use the modified Routh criterion. 25.11. A temperature control loop includes a second-order overdamped model
HG(z)

(0.0826 + 0.0368r1)r1 (1 - 0.894r1)(1 - 0.295z-1)

and a digital PI controller
D(z)

=

Ke(
Kem

1 + 8(1 ~

r1))

Find the maximum controller gain =

for stability. 0.8 PI1-]

25.12. The following digital control algorithm has been proposed:
p"
2en

+

1.3e"_1

+

0.7e"_2

-

-

0.4P"-2

where

P

is the controller output and e is the error signal.

(a) Derive the corresponding pulse transfer function P(z)/E(z). (b) Is the controller stable? (c) Suppose that a unit step change in e(nM) occurs. What is the steady-state value
of p?

Exercises

613

25.13. A digital controller is used to control the liquid level of the storage tank shown in the drawing. The control valve has negligible dynamics and a steady-state gain of Kv = 0.1 ft3/(min)(mA). The level transmitter has a time constant of 30 s and a steady-state gain of 4 mA/ft. The tank is 4 ft in diameter. The exit flow rate is not directly influenced by the liquid level; that is, if the control valve stem position is kept constant, q3 "" f(h). Suppose that a proportional-only digital controller and a digital to analog converter with 4 to 20 mA output are used. If the sampling period for the analog to digital converter is At = 1 min, for what values of controller gain Kc is the closed-loop system stable? Will offset occur for the proportional controller after a change in set point?