32 Mathematical Models Models of Sorne Control System Components 33

We form y as the sum of lhe constant y, resulting from Mg and a variation 8y d8

B-; Kh
resulting from /(t); that is,

y =
YJ + 8y {2.14)

Using (2.14) in (2.10) with y, from (2.13) yields the linear differential equation for T(t) 8
the variation 8y as (b)

d 2 (8y) d(8y) T(s) (le.}

M + B -- + K 8y = /(t) (2.15)
dt 2 . dt

Solving (2.15) for d 2(l;y)/dt 2 and taking Laplace transforms throughout with ali
initial conditons set to zero gives

B . K 1 (e)
s2 1;Y(s) = - -s8Y(s) - - 8Y(s)
M
+ -F(
M
s) (2.16)
M -

l'
1
T(sl 1 , 1 8(s)
where 8Y(s) and F(s) are the Laplace transforms of 8y(t) and f(t), respectively.
52 + ll s + !f
\V e use (2.16) to construct the detailed block diagram in Figure .2.4c. First, - J .,
-

j
s 8 Y( s) is obtained as the output of the summer, having inputs -( B/ M )s l;Y( s ), 1
2
(d)
_:.( K/M) 8Y(s), and (1/M)F(s). We next form s 8Y(s) by passing s2 8Y(s)
through a single integration block, that is, one with transfer function 1/s. We Figure 2.5 Detailed block diagram development for a simple rotational
1
l.
mechanical system.
continue by passing s8Y(s) through a second integralion block to yield l;Y(s).
Feeding sSY(s) and 8Y(s) back through appropriate gains (B/M and K/M, Laplace transforms throughout with ali initial conditions set to zero gives
respcctively) provdes the inputs needed in (2.16) Cor the summer, as shown in
Figure 2.4c. Finally, solving for l;Y(s)/F(s) from (2.16) gives the system transfer B K l
function in Figure 2.4d. s 2 0(s) = -s8(s) - -fJ(s) + -T(s) (2.18)
J J J
The detailed block diagram of Figure 2.Sc is constructed from (2.18) in the
EXAMPLE 2.3 S!Wle way that Figure 2.4c was obtained from (2.16). We forro sO(s) by passing
s 20(s) through an integration block and then form O(s) by passing sfJ(s) through
The rotational mcchanical system of Figure 2.5a has the free-body diagram given a second integraton block, as shown in Figure 2.5c. Both sfJ(s) and fJ(s) are fcd
in Figure 2.5b. We form the algebraic sum of the viscous damping torque back through the appropriate gains ( B/J and K/J, respectively) and lhe outputs
( B d(J/ dt ). the spring lorque ( K(J), and the externally applied torque T(t) with are summed (negativcly) with (1/J)T(s), as indicated in (2.18), lo yield a detailed
the direction of ncreasing shaft angle position O as positive and sel this surn block diagram with an externally applied torque. The equivalent transfer function
equal to the product of the moment of inertia (J) and shaft acceleration (d 20/dt2 ) shown in Figure 2.5d is obtained by solving (2.18) for O( s)/T( s). Tbese equa-
to give tions are nceded later in this .section to describe the rotational mechancal parts of
1
dynamic models Cor de motors.
d 20 dO
J -2 = -B - - KfJ + T(t) (2.17)
dt dt
The Lagrangian formulation provides a unfied approach for the dynamic
Ohscrve that (Js = O is thc static operating condition an<l thus fJ itself is a time description of a broad class of physical systems [10-12J. Systems composed of
variation analogous to 8y in Example 2.2. Dividing (2.17) by J and taking several different types of interrelate<l subsystems-for example, electrical, mechan-
 VlC1lV "e.u ,u..., ........... "._. .. ,

74 Mathematical Models
TABLE 2.4 ( Continued)

I}{lu,
Must be in factored form;
o partial fraction expansion
Uncoupled form -y
2 required; each state varia
8,.
Y(s} 81
--=--+ ...+--+/3n x= ble models only one mode
U( s) S + y1 S + Yn

(-;o , of behavior for distinct ei

genvalues; time response
easily determined.

y-lJ+P.u

Must be expressed in fac

Different for the various sub tored form; subsystems
Cascada or tandem form system tormats selected.
are simulated from prod
Y(s) = J S + Z1 ) .. ( S + Zn)
uct terms and fed into next
U( s) "\ s + Y1 s + Yn subsystem; may corre
spond to physical subsys
omputer Simula/ion Diagram tems; difficult to write state
variable equations.

1 r.:-1 - -
-

r together,
s in the numerator and denominato
By arbitrarily grouping first term
terms together, we obtain
2.4.4 Cascade or Tandem Form second terms together, and third
Y(s) s+z1 s+z2 s+z3 {2.96)
--
et both the nume rator and denominator of (2.85)
be represented in factored -=K ( - -) ( --) ( s + YJ )
U( s) s + y 1 s + y 2
,orm to yield fo rming
tion dia gram of Figure 2.25b consists of
K(s + z 1 }(s + z 2 )(s + zJ The cascade, or tandem, simula in Figure 2.25a from (2.96). The dia grams
Y(s} (2.95) nec ted sub syst em blocks sho wn
the con
U(s} = (s + y }(s + y2 }(s+ y3)
144 Feedback Characteristics
Problems 145
Equating (3.125) and (3.126) yields the same equations as in (3.124) a nd, 13. J. J. DiStefano III, A. R. Stubberud, and I. J. Williams. Feedback and
consequently, the same controller parameters. Control Systems, Schaum's Outline Series. New York: McGraw-Hill,
1967.
14. T. E. Fortmann and K. L. Hitz. An lntroduction to Linear Control Systems.
New York: Marce} Dekker, 1977.
1~. T. Takahashi. Mathematics of Automatic Control. English Translation by
Holt, Rinehart, and Winston. New York: Holt, Rinehart, and Winston,
SUMMARY 1966.
lfi. H. L. Harrison and J. G. Bollinger. Introduction to Automatic Controls, 2nd
We have described closed-loop system behavior and representation in a quant1t,1 ed. Scranton, Pa.: Intemational Textbook Co., 1969, Chapter 8.
tive manner in this chapter. As a continuation of the mathematical modeling 111 17. W. H. Bode. Network Analysis and Feedback Amplifier Design. New York:
components in Chapter 2, we have emphasized the modeling of the compl1t11 Van Nostrand, 1945.
system. We discussed feedback properties and their application to systems h11th lk R. Tomovic. Sensitivity Ana/ysis of Dynamic Systems. New York: McGraw-
generally and in specific examples. We presented gain versus bandwidth tradcPII Hill, 1963.
steady-state error analysis, and system sensitivity concepts and described twt 1~,
J. B. Cruz, Jr., ed. System Sensitivity Ana/ysis. Stroudsburg, Pa.: Dowden,
control laws (PID and SVFB). Hutchinson, and Ross, 1973.
P. M. Frank. Introduction to System Sensitivity Theory. New York: Academic
Press, 1978.

REFERENCES

1. B. C. Kuo. Automatic Control Systems, 4th ed. Englewood Cliffs. N 1

Prentice-Hall, 1982, pp. 7-11. l*flOBLEMS
2. R. C. Dorf. Modern Control Systems, 3rd ed. Reading, Mass.: Adcl1,11n
Wesley, 1980, Chapter 3. lfi l. 1) 3.1 Show a schematic diagram and then determine which of the five
3. J. L. Melsa and D. G. Schultz. Linear Control Systems. New 'l'tJ1~ characteristics of Table 3.1 are most important in each of the
McGraw-Hill, 1969, Chapter 3. following closed-loop system operations.
4. J. J. D'Azzo and C. H. Houpis. Linear Control System Analysis ami /J111
Conventional and Modern, 2nd ed. New York: McGraw-11111 , l'lkl a. Automobile automatic speed control.
Chapter 6. b. Anti-aircraft artillery (with sensor).
5. K. Ogata. Modern Control Engineering. Englewood Cliffs, N.J.: Prenl1n l l 11I c. Mid-course correction of an unmanned interplanetary
1970, Chapter 5. spacecraft.
6. C. T . Chen. Analysis and Synthesis of Linear Control Systems. Ncw \';1~ d. Voice-activated tape recording system.
Holt, Rinehart, and Winston, 1975, Chapters 6 and 7. 1 1) .l.2 Describe different feedback systems for which each of the five
7. A. P. Sage. Linear Systems Control. Champaign, Ill.: Matrix P11hl1 \ II characteristics of Table 3.1 are of major importance. In particular,
1978, Chapters 4 and 5.
1 describe one system for whicl the fust characteristic is espe~ lly
8. W. R. Perkins and J. B. Cruz, Jr. Engineering of Dynamic S y.1111111 N important, another for which the second is a primary consideration,
York: Wiley, Sections 3.6, 6.8, 11.7, 11.8, and 13.3. a nd so forth.
9. F. H. Raven. Automatic Control Engineering, 3rd ed. New York M1 t ,1 11
Hill, 1978, Chapter 4. 1) .\..\ Describe three applications as altematives to those in Table 3.1 to
10. V. W . Eveleigh. Introduction to Control Systems Design. Nl' W ' ' " ' ' illustrate communication-at-a-distance, tracking operations, and
McGraw-Hill, 1972. power amplification. lndicate which feedback characteristics are of
11. J. G. Truxal. Introductory System Engineering. New York : MrC:rnw ii1f1 maJor importance for each application.
1972. 1 t4 Ske tch frequency response curves of system gain versus frcqmncy w
12. W L Drogan. Modcm Control Theorv. 2nd ed . Englewood Cli lh, N 1 for the ,\ystem of Example 3.1 in Figure 3.4a. l.c t K 1, a nd sklfr h
Prc nltc.:c llall, 11)8'1 c m vl''> for fl O, O.5, and 1.
.J

168 Time Response Second-Order Systems 169

Using (4.25), we may write 12
1
100
e"UH= ( ) (4.30)
%OS 101-
1
1
or
1
77{
{17
- In (
100
)
%OS .
(4.31)
"'"'
;:
\ 1

Squaring both sides of (4.31) and solving for f gives 1

,ij
e

100 z 4
In ( --)
%OS
{= (4.32)
' 2 100
.,, + ln2 ( --) 2
%OS

Using (4.32) for %OS= 20%, we find that f= 0.456. Since we require the %OS to

,_
o.____,__...___,.____.__..____,__...._.___.____.
be less than or equal to 20%, then f must obviously be greater than or equal to O 0.2 0.4 0.6 0.8 LO 1.2 1.4 1.6 1.8 2.0
0.456. This conclusion may be obtained by examining (4.32) in greater detail or,
more simply, by observing from Figure 4.4a that the %OS is a monotonically
r_
decreasing function of lnterpreting this result in terms of K A by using (4.29) Figure 4.6 Normalized curves of rise time, delay
time, and settling time for the second-order sys
yields tem of Figure 4.2.
1
t= --,-- 0.456 (4.33) and y2 in (4.7) from (4.8) as 0.5 and 2, respectively. Using these values in (4.10)
2/IOKA. gives y(t) as

which gives the range of K A as O < K A 0.120. We show later in this section y(t) = 1 - ie-051 + }e-2, (4.34)
that other parameters, such as a tachometer feedback gain, may be adjusted to
meet the %OS performance specification. In such a case, the amplifier gain KA Let the value of t for which y(t) = 0.9 be denoted as t0.9. Consequently, from
may be set at a much higher value to satisfy other design specifications. (4.34), t 0.9 satisfies

0.9 = 1 - ie-o.510 + }e-2, (4.35)

or
4.2.6 Rise, Delay, and Settling Times
- O 1 - 3 e -o.5ro. + l3 e-2,. -
/(t0.9 ) - -O (4.36)
Curves of rise time (t,), delay time (tJ ), and settling time (ts ) versus the damping
coefficient f are presented for O < f 2 in this subsection. Instead of plotting t,, We may solve (4.36) numerically, for example, by using Newton's method, 1 to
td and ts versus f directly, it is more enlightening to forro normalized curves of obtain t0.9= 5.18. Sirnilarly, t0.1, which satisfies y(t01 )= 0.1, may be determined
wnt ,, w,,t J , and w"t, versus {, as shown in Figure 4.6. Points on these curves can
be obta(ned by setting w" equal to 1 and then determining t,, td and t, according
to their defimtions. The resulting normalized curves are valid for any value of w,,. 1
Ncwton's method is useful for solving algebraic equations of the form /( a) = O. Thc mcthod is based
As an example of how to forro these curves, suppose we wish to compute the on a recursive relationship a,H 1 = aN - /( N )//'( N ), which often converges to an acceptable
value of w"t, for the oerdamped case where f= 1.25 and w,,= l. We find y1 wlution from a suitably chosen starting value a0 in as few as three or four iterations.
 Stability Definitio.n and Properties 235

5
-
Stability Analysis
in the s-Plane
/\ dynamic systcm is said to be asymptotically stable if it tends to resume its given
1 1wrating condition after a small perturbation in values of the system state
v,11 tables. For example, an unpowered roller coaster tends to move to a low point
1l1111l.\ its track in the absence of prohibitive frictional forces. An automobile
' , 11w,c" control permits the vehicle to resume its preset speed following a Figure 5.1 Regions of stability and instabil-
ity in the s-plane for linear time-invariant
11101111.:ntary period of braking or acceleration. An automatic pilot brings the
systems.
,111 1 raft back to its in tended course after temporary disturbances due to unex-
pn tcd turbulence. Stability analysis enables the control engineer to ensure that The roots of the characteristic equation are not only the eigenvalues of the matrix
, IH,c systems behave properly when such perturbations occur. A but also the poles of the transfer function between input and output. Thus, the
lhc stability analysis problem is inherently related to the design problem for roots of the characteristic equation for a closed-loop system are the closed-loop
lllll'ar time-invariant systems. The transform solution to both problems for such
poles.
\Y\k1m is to place the closed-loop poles in desired locations in the s-plane. The
w 11 c~ponding state variable approach p roduces a suitable time response for the Definition 5.1. The system in (5.1) is
/.y1, h! m state variables. This chapter begins a two-chapter study of stability
.111aly~is. Rcsults in the s-plante are presented here and compared in Chapter 6 Asymptotically stable if and only if ali the roots of the characterist ic
w11h frccuency-domain stability analysis procedures. equation for A lie in the left-half s-plane.
Fo llowing a brief description of the definition and properties of stability, we Marginally stable if sorne of the roots occur as a single root at thc origin
p, cM.:nt a criterion to determine stability by using only the characteristic poly- in the s-plane or as single pairs of roots at points on the jw-axis ami 1he
n,11111al cocfficients in a convenient table. Construction rules for the root locus remaining roots are in the left-half s-plane.
nwthod are first described for negative feedback systems and then extended to the Unstable if one or more roots occur in the right-half s-plane or if
po~11iv1: feedback case. Root contours, obtained by varying a parameter other multiple roots or pairs of roots occur at a poinl on the jw-axis.
than the o pen-loop gain, form a generalization of the root locus method.
Figure 5.1 indicates these regions of stability in the s-plane.
In terms of the impulse response function g(t) for linear timc-invariant
systems, asymptotic stability is guaranteed if and only if
5.1 STABI LITY DEFINITION ANO PROPERTIES
foa,1g('T) ld'T; 00 (5 .3)
1 or the linear time-invariant system described by
In words, a linear time-invariant system is asymptotically stable if and only if the
x = Ax (5.1) impulse response is absolutely integrable over the infinite range (O, oo).
It is instructive to consider the associated stability problcm for the nonlincar
the charm.:teristic equation may be expressed as
system [1-8] described by
det (s/ - A) = O (5.2) i = f{x) (5.4)
 236 Stability Analysis in the s-plane
The Routh-Hurwitz Criterion 237
Wc may determine the equilibrium state x,. from
5.2.1 Hurwltz Test
f(x,.) = O (5 .5)
Let the characteristic polynomial of the transfer function of the system be written
as
where x ~ is simply the static operating point determined by linearization in
Section 1.4. As before, we define 8x as
/(s) = Sn + a,. _ Sn - l + +2S 2 + S + o (5 .8)
6x = x(t) - x,. (5.6)

Lincarized variational equations for 6x can be expressed in the form of (5.1) for A necessary condition for asymptotic stability is that aU coefficients in (5.8) rnust
each equilibrium state x, as have the same sign and none may be zero. This test by inspection is referred to as
the Hurwitz test.
We note that a necessary condition fortmarginal stability is that the Hurwitz
6x = !!_ 1 . 6x (5 .7) test as stated above is satisfied except that all odd or ali even coefficients may be
ax x - x.
zero. If only even coefficients are present, then we might have single pairs of roots
As in (5 .2), roots o the resulting characteristic equations enable us to determine of points on the jw-axis. If only odd coefficients are present, we also have a root
the stability of each x,. For the linear time-invariant system in (5.1), there is only at thc origin (s = O). OnJy a necessary condition, the Hurwitz test for marginal
one equilibrium state (x, = O). The asymptotic stability o this single equilibrium stability may be satisfied and yet the system is unstable. Al! we have guarantecd is
that the failure of the Hurwitz test implies instability.
s tate is guaranteed if and only i ali roots of the characteristic equation lie in ihc
left-half s-plane or, equivalcntly, if the system response remains bounded for ali
bounded inputs. Therefore, we often omit reerence to the "stability of the
equilibrium state x /' and simply examine the "stability of the system." Observe
once again that the stability of a linear system depends only on the dynamics of 5.2.2 The Routh Table
thc systcm and not its input. The s-plane propcrties defined in this scction lead to
operatio nal proccdures for determining the stability of a given linear time- Hurwitz [9] derived a necessary and sufficient condition for asymptotic stability
invariant system. based on the evaJuation of certain determinants involving the characteristic
polynomial coefficients. These numcrical computations are quite cumbersome,
especially for higher-order systems. Earlier, Routh (10) had developed a tabular
form that involves only simple scquential calculations. The interpretation of the
5.2 THE ROUTH-HURWITZ CRITERION Hurwitz dcterminants by means of calculations in a Routh array or table is
reerred to as the Routh-Hurwitz critcrion.
We present here a procedure for determining the stability of a linear time- The Routh table for a general polynomial is provided in Table 5.l. The
invaria nt system by examining its characteristic polynomial directly. If we were coefficients of the characteristic polynomial in (5.8) are arranged in the first two
sornehow able to factor this polynomial, we could easily plot the root Jocations in rows of the table as shown. Assuming for the .moment that a,. _ 1 is nonzero, thc
the s-plane. According to the results of the last section, the system is asymptoti-
cally s tablc if ali roots are in the left-half s-plane. liowever, we now describe a .:o::
TABLE 5.1 General Form of the Routh Table u
criterion that <loes not require the factorization of the characteristic polynomial :;S ('\~.:
but rathcr utilizes the polynomial coefficients in an easily applied algorithm to
prcdict stability [9,10].
sn
s" - 1 ' n - 2 n - 4
H
.-~ .-
, -: r :,
n - 1 n - 3 'ii:':;,fS>
~~:.~
s" - 2 n - 5 l '-'
The Routh-Hurwitz criterion yields both necessary and sufficient conditions for Pn - 2. 1 Pn - 2.2 Pn - 2. 3 .... ;-;j
8~/(:{,,.:11!!'~~:\\ " .
s" - 3
asympto tic stability. In other words, the systern is asymptotically stable if and Pn - 3 .1 Pn - 3,2 Pn - 3. 3 i~'!"!?;"n ,\,;1~, r~ r ..-,
o nly if th; cocfficicnts o the charactcristic equation satisfy this criterion. Initially, '} i7.-f_'lf:J ~ t,'J~ p ., .
wc present a preliminary Hurwitz test as a necessary condition for asymptotic s2 P2, 1 "2.2
-~-~o.~, t~ f\ ~ '-;
,.~~;.,x;F '.:!.~
s' 11"1:zs\~Jl' r'i
stahi lity. The wc discuss the Routh-Hurwitz criterion, including two special
cases which somctimes occur in complcting the Routh table. s
P 1. 1
Po . 1
I
f~
rr. ,:.,
..

...l
f' <..:,
 The Routh-~urwitz Criterion 239
238 Stability Analysis in the s-plane
TABLE 5.2 The Routh Table for Example 5.1(a)
ncx t row o f the table may be determned from
s 1
6 13 I
s3
11 n-21 s2 "'' - 11 12 ./4
Pn - , = (-1) n- l a,,_ 3 = a,, _ 1an-2 - (l)a,. 3
s' P1.1=9.8 ~4/
2 1
a,, _ an- 1
s Po,1=4~

1 an - 4 1 the s 2 row- is formed from the entries in the first two ~ lumns of the previous
a,,_ 1a,, _4 - (l)a,. _5
Prr-2,2 = ( -1) a,, _.
1 n - S (5 .9) two rows as
a,, _ n-1

131
P2,
1
1
= (-1) 16
::.
l2 = (- l)( 1(12) - 6(13)) _
6 - 11 (5.10)
e In, o: a11y row has been completed, the elements of the following row may be
,li 1, 1m1ncd from the previous two rows in the same manner, provided no division
l,y 1110 is implied. This possibility will be discussed in special cases later in the The entry in the second column of the s 2 row is
,1 'I t lt 111
1lu: interpretation of the Routh table is easy to remember:

P2, = ( - 1) \~ t:. cil = ( - l)(l(O) -6 6(4) ) -- 4 {5.11)

The number of roots in the right-half s-plane is 2
equal to the number of sign changes in the lirst
column of the Routh table. Observe that a zero is assumed whenever no entry appears in a particular
3
location, such as the absence of an entry in the third column of the s row.
Moreover, when the entry in the last column of a row is nonzero but the entry
l li1s condition is known as the Routh-Hurwitz criterion. We do not provide a
immediately below is m.issing (or zero), then that nonzero entry automatically
d,, 1v11tion of this criterion, but we will be using the Routh table extensively as an
appears one column earlier in the second row following, as indicated by the
o pn il tional too! for stability analysis. 4
arrows in Table 5.2. For example, the + 4 entry in the s row, column 3, has a
3
m.issing entry below it in the s row. As calculated in (5.11 ), this + 4 automati-
cally appears as p2 2 in column 2 of the s 2 row.
l XAMPLE 5.1
Continuing with.Table 5.2, we have
h>1 thc given characteristic polynom.ials, use the Routh-Hurwitz criterion to
il,tl11111nc the number of roots in the right-half s-plane. 6 121
p 1, I = ( - 1) 111 11 4 -- 11
108 = 9.8 (5 .12)
11 /( .1) ~ s 4
+ 6s 3 + 13s 2 + 12s + 4 .

h /(s) = s + 5s 4 + 2s 3 + 18s 2 + 45s + 25

5

l' . /(s) s 5 - 4s 4 + 4s 3 + 9s 2 - 12s + 4

Finally, Po. i is +4, formed from p 2, 2 asan automatic entry because P1,2 is zero.
Observe that ali entries in the first column of the Routh array are positive. Since
A:.. a p1climinary step, we observe that only the polynomials in (a) and (b) pass
there are no sign changes in this first column, the Rou th-Hurwitz criterion
thl' 1lurwitz test. Sign changes among the coefficients for the polynom.ial in (c)
guarantees that there are no roots in the right-half of the s-p lane. This result from
111d1cutc instability for the corresponding closed-loop system. Addilional informa-
the Routh table is consistent with our a priori knowledge for this fourth-order
11 0 11 , cgar<ling root locations for each of the three characteristic polynomials can
contrived polynom.ial, which has two roots at - 1 and two roots at - 2. The
he obtai ned fro m the Routh table.
student should verify that the product (s + 1) 2 (s + 2) 2 does yield the given
'I he cocf cients of the polynomial in (a) are placed in the first two rows of the
Ro uth array in Table 5.2 The entry in the first column of the next row-that is, polynomial.
'<
240 Stability Analysis in the s-plane The Routh-Hurwitz Criterion 24 1
TABLE 5.3 The Routh Table for Example 5.1(b)
R(s) :
---1..l:1
A ., j
< +
K(s - 2
1) (sl + 2,s + 2)
1
Y(s) .,

~
s5
s
:,~ > 51 12 4::,
s7 > -1.6 S 25
1 143 40 /
!, 40.3 /25....-------
25~ Figure 5.2 The system for stability investigation in
Example 5.2.
The Routh table for the characteristic polynomial in (b) is shown in Table 5.3.
The succcssive formation of entries beyond the first two rows yields We note that the polynomials in (a) and (b) have the rcmaining roots in thr
left-half s-plane. Specifically, we showed above that (a) and (b) havc none and
1 451 two right-half s-plane roots, respectively. If sorne of the rcmaining roots for thcse
P.u =(-1)1~ 5 l~I = -1.6
1
()3.2 = ( - 1) 5 5 25 = 40 polynomials wcre on the jw-axjs, then the second spccial case described in thc
following subsection would be indicatcd. Bccausc this condition that ali elcments
of a row are zero is not indicatcd, wc conclude that the fourth-ordcr polynomial
5 18 1 5 251 in (a) has four lcft-half s-plane roots and the fifth-order polynomial in (b) has thc
P2.1 = ( - 1) 1- 1.6
-1.6 40 = 143
P2.2 = (-1) 1 - 1.6
-1.6
0 = 25 remaining three roots in the left-half s-plane. Therefore, since wc havc specified
that these thrce polynomials are characteristic polynomials- that is, denomina-
tors of closed-loop transfer Cunctions- thc corresponding systcm in (a) is asymp-
- 1.6 401 143 251 totically stable (ali four poles in the lcft-half s-plane) and the systems in (b) and
(e) are both unstable (al least one right-half s-plane pole in each case).
P1.1 = ( - 1) 1 143143 25 = 40.3 Po.1 = ( - 1) 40.3 _ O -_
1
25 ( 5.13)
40 3
EXAMPLE 5.2
Ob:..crve that p 2 2 and Po. 1 may be obtained automatically as indicated by the
arrows in Table 5.3. The two sign changes in the first column of this table mean
that there are two roots in the right-half o the s-plane. The factors o this Use the Routh-Hurwiti criterion to determine the range of K for stability for the
closed-loop system in Figure 5.2.
contrivcd polynomial in (b) are (s - 1 - j2)(s - 1 + j2)(s + 1)2(s + 5), which
The characteristic equation is given by
may easily be verified.
Table 5.4 shows the Routh table Cor the polynomial in (e). Four sign changes in
the first column indcate that there are four roots in the right-half of the s-plane. K(s - 1)2 =O
I + G(s)H(s) = 1 + 's + l)(s 2 + 2s + 2' (5.14)
Moreover, we can further identify that the fifth root is in the left-half of the
s-planc. If the fifth root were on the jw-axis, it would have to be at the origin,
Multiplying by the denominator of G(.v)H(s) in (5.14) and grouping terms yields
bccausc roots off the real axis must appear in complex conjugatc pairs. A single the characteristic polynomial
root a t thc origin occurs if and only if a 0 in (5.8) is zero. Thus, the single
fifth root must lie somewhere on the negative real axis. The four right-half s-plane 3
f ( s) = s + ( K + 3)s 2 + ( 4 - 2K )s + K + 2 = O (5.15)
roots may he either four real roots, two real and two complex roots, or four
The Routh array is shown in Table 5.5. We calculate P1, i as
complex roots.

TABLE 5.4 The Routh Table for Example 5.1(c) _,K~3 4x-+1:1 (-1)[(K+2)(1) - (K+3){4-2K)J
P1.1 = - -- - -- =-- - - -- -+ 3- - ----
~
S5 1 4 -12 K +3 l'i.
~

S4 -4 9 4 - 2K 2 - 3K + 10
si 1
6.25 -11 ~
K+3 (5.16)
s7
s, > 1. ~4
-23.76
The valuc of Po. 1 is easily determined as K + 2. The en tries in the first column of
.so > 4
the Routh array must be positive for asymptotic stability-that is, P2,1 > O,
242 Stability Analysis in the s-plane The Routh-:Hurwitz Criterion 243

We can show by the methods of the following subsection that closed-loop potes
TABLE 5.5 The Routh Array for Example 5.2
appear on the jw-axis for K = - 2 and K ::: 1.61.
s3 1 4 -2K
~ K+ 3 K+ 2
s' P1.1 ~
s Po.1

5.2.3 Special Cases for the Routh Tabl~

Pu > O, and Po. 1 > O. These conditions are satisfied by setting
The problem occurring when a zero appears in the first column of the Routh table
Pi. i = K + 3> O is discussed here. This situation ~equires a special consideration because division
by zero is indicated when blindly applying the forrner rules. Two possibilities may
-2K 2 - 3K+ 10 be encountered:
Pi . 1 = K+3
>0
1. A zero appears in the first column of a row and at least one nonzero elcmenl
Po.,= K + 2 > O (5 .17)
appears elsewhere in the same row.
2. Ali elements of a particular row are zero.
1he first and third inequalities hold for K > - 2. We sketch the numerator of p 1, 1
Vl'1 s usK in Figure 5.3 and note that this function is positive for ( - 3 - ,/89 )/4
K < ( - 3 + /89)/4. T his range is approximately - 3.11 < K < 1.61. Since we In the first case, we simply replace the zero in the first column by an arbilrary
aho require K > - 2 because of the other inequalities in (5.17), we form the small positive number e and proceed as before to complete the table. In the
111tcrsection of the a llowed regions for K . Therefore, the range of K Cor the second case, we form an auxiliary polynomial fa( s) corresponding to the last row
a,y111p to tic stability of the closed-loop system in Figure 5.2 is approximately and the particular row having aJI zeros. We then rcplace the row that would havc
contained ali zeros by the coefficients of the derivative of the auxiliary polynomial
- 2 < K < 1.61 (5.18) with respect to s. T his second possibility occurs if there are roots of the original
characteristic polynomial either on the jw-axis or symmetrically located on thc
real axis about the origin in the s-plane. These jw-axis or real-axis roo ts are also
roots of the auxiliary polynomial. The second case also occurs when a quad of
12
t
Num lp,I
four complex roots appear symmetrically about the origin in lhe s-plane. Details
of procedures for these special cases are illustrated in the following examples.

EXAMPLE 5.3

Use the Routh-Hurwitz criterion to determine the number of right-half s-plane

6 roots for the given characteristic polynomials.

a. /(s) = s 3 + s + 1
3 b. /(s) = s 3 + 2s 2 + s + 2
c. /(s) = s 5 + s 4 + 4s + 4

o We see at the outset that the polynomials in (a) and (c) fail the Hurwitz test and
-4 -2 - 1 o instability is indicated. T he Routh table can provide additional information on
K- root locations.
The Routh array in Table 5.6 for the polynomial in (a) shows the need for the
Flgure 5.3 Sketch of the numerator of p 1 1 versus K in first special case immediately in the second row. We set p2 1 = e, where e is an
(5.16) and Table 5.5.
244 Stablity Analysis in the s-plane The Routh-Hurwtz Criterion

TABLE 5.6 The Routh Array for Example 5.3(a) TABLE 5.8 The Routh Array for Example 5.3( e)

~v~7----
si 1 1 55 1
1 O

~V'
s2- >
54

si -- -4-----------~ :
s'
> s2
s >
5,
>
arhitrary small positive value. Calculating p 1,1 and Po. 1 gives 5

1
P1. 1 =( - l) I! E ~I _- 1 - -~
} We indicate thc occurrence of the second special case by a dashed line followin
E E 2
the s row. Inserting for P1.1 the coefficient of s 1 in (5.22), that is, +4, we mav
then complete the Routh arra y in Table 5. 7 by sctting Po. i = 2. Therc are no sign
P, , - (-i) J-:;,
-1/t
~I
=1
(5.19)
changes in the first column and, consequently, there are no right-half s-planl'
roots of the polynomial in (b).
The dashed line indicating the occurrence of the second special case at the .1 '
Wc note two sign changes in the first column of the Routh array because only row means that there are two roots either on the }<il-axis or symmctrically located
r ,.1 < O. Thus, there are two roots in the right-half s-plane for this thi.rd-order about the origin on the real axis. These two roots may be determined by factoring
/ 0 (s) in (5.21) to yield (s + j)(s - j). Furthcrmore, we note that fc,(s) is itself a
polynomial. We observe that the second special case must occur if any roots
appear on the jw-axis. Since the second case is not indicated in Table 5.6, we factor of the original polynomial in (b). The roots of that original polynomial are
co nclude that the remaining root is in the left-half s-plane. + j, - j, and - 2. The corrcsponding closed-loop system may be classified as
Table 5.7 shows the Routh array for the polynomial in (b). Jf we try to form marginally stable.
fl 1. t from thc two prcvious rows, we obtain The Routh array for the polynomial in (c) is shown in Table 5.8. Since thc s \
4
and s rows are identical, the s 3 row would contain ali zero cntries if the general
rules of the previous subsection were applied blindly. We place dashes after the s ~
P1 ,1 = ( - l)I~ 2 ;I =o (5 .20) row, indicating the second special case, and form an auxiliary polynomial as

Rccause therc are no nonzero entries in thc s 1 row, thc second special case is
2
fa(s) = p4_s 4 + P4 ,2S 2 + P4.3 (5 .2))
indicatcd . We forman auxiliary polynomial / 0 (s) from the s row as

fa ( S) = 2s 2 + 2 (5.21) where p4 _1 = 1, P4.2 = O, and p4 3 = 4. lts dcrivative f~(s) is 4p4 _1s 3 + 2p4 _2 I .
which yields entries of 4p4 1 = 4 and 2p4 2 = O in the s 3 row, as shown in Tahli
Taking the derivative of /"(s ) yields 5.8. Completing the table yields two sign changes in the first column. Wc norc
that the facto rs of /,,(s) are (s + l + .f), (s + 1 - j). (s - 1 + j), and (s - l - ./)
, ( s) = d/ s) = 4 s
0
( (5 .22) Thcrcfore, a quad of the corresponding four roots are formcd in the s-plane wirh
/." ds a fifth root somewhere at (s = - 1 for this particular case) in the lcft-half s-pla ,w

TABLE 5.7 The Routh Array for Example 5.3(b)

EXAMPLE 5.4
S3 1 1
52 " 2
I For the closcd-loop system of Figure 5.4, determine the range of K for asymrto r11
J 1 -- ---~ ~ - - - - -- -
51 stability, thc value of K to place closed-loop potes on the jw-axis. and rlle
s location of thosc .fw-axis polcs.
 246 Stability Analysis in the s-plane
The Root Locus Method 2 47

R(s) Y(s) Jw*, and we have

+ CE) I,. + 1) (s: 2) (s + 3)
2
6{Jw*) + 66 = Q , (5.29)

from which w* = fil= 3.32.

For K < - 6, p0 1 is negative and there is a single sign change in the first
Figure 5.4 System for Example 5.4. column of the Routh array. We again recognize the second spccial case, form an
auxiliary equation of / 0 (s) = p 1, 1s1, and sets= /w* to obtain w* = O. Thus, we
Wc form the characteristic equation from
conclude that fr K = - 6 a closed-loop pote occurs at !he origin of the s-plane
and for K = + 60 closed-loop potes are on the jw-axis at s = Jv'll. The system
1 + G(s)H(s) = l + 's + l)(s +K 2)(s + 3'
=O (5.24) is marginally stable for these values of K .

S1 111plifying yields

s 3 + 6s 2 + lls + 6 + K =O (5.25)
~3THEROOTLOCUSMETH0D
1lw Ro uth array is given in Table 5.9. To determine the range of K for
u,y111 ptotic s tability, we require that ali first column entries are positive, that is, The root locus method enables us to sketch the locations of the closed-loop poles
as sorne system parameter is varied [11-13]. We restrict ourselves in this section
60 - K and the next one to positve variations in the open-loop gain K in the negatvc
P1.1 =
6
> O =K < + 60
feedback configuration, as shown in Figure 5.5.
Po. 1 = K + 6 > O =K > - 6 (5.26)

1 h1 111tcr!)cction of the ranges of K specified in (5.26) yields the range for 5.3.1 The Root Locus Concept
,1w1 11pt o tic s tability as
Consider as a standard form for root locus construction the open-loop transfer
-6 < K < +60 (5.27) function given by

<nm1dc1 the n.:sult of setting K equal to + 60. For K > 60, p 1 1 is negative 1 K(s + z )(s + z ) (s + z,)
111d , rnm cqucntly, there are two sign changes in the first column. We note that GH(s) = ---- --2 - -- - - (5 .30)
(s + Pi)(s + P2)(s + p3) ... (s + Pp)
1 1 1 O whcn K = 60, and the second special case occurs. We may write the
.111\ 11 1.11 y cq uation as
where there are z finite zeros and p finte poles of GH(s ). We wrt thc
characteristic equation for the system of Figure 5.5 as
/ 0 (s) =6s 2 + K + 6IK=60 = 6s 2 + 66 (5.28)
1 + GH(s)::: O (5.31)
1lll' o f /,,(s) = O are also roots of the original polynomial in (5.25). Since
111111\
tl11 w , 001.s occur on the jw-axis, we denote the corresponding values of s as
R (s) + Y(s)

'I AIJI 1 5.9 The Routh Array for Example 5.4

"' 6
t 11
K+6
' - - - - - , H(s) 1-1 - - -
.,1
/1 1, 1 606~
1> 1 Po. 1 K 1 6 Figure 5.5 Feedback diagram for basic
root locus construction.
248 Stability Analysis in the s-plane The Root Locus Method 249

and search for ali points s in the s-plane satisfying (5.31). Therefore, we require utilize a digital computer, possibly with associated computer graphics, to search
for those points in the s-plane satisfying (5.33) and (5.34), pr (5.37) and (5 .38).
GH(s) = - 1 (5 .32) Instead, we emphasize here the development of sorne basic root locus construc-
tion rules derived from (5.33) and (5.34) and their application to obtain root loci
There are two conditions implied by (5.32). These conditions relate to the sketches.
magnitude and phase (or angle) of GH(s), that is,

IGH(s) I = 1 (5.33)
5.3.2 Basic Construction Rules
/ GH(s} = {2k + 1)11 (k = o, 1, 2, ... ) (5.34)
Six root locus construction rules are provided in Table 5.10. We discuss in thi.s
In words, (5.33) states that the magnitude of the open-loop transfer function must subsection the first three of thesc rules and then work sorne carefully sclected
be unity. 1 The gain K may be adjusted to satisfy this requirement, provided examples that require only the application of these rules. Rule 4 is described more
(5.34) holds. Thus, the root locus method may be characterized from (5.34) as fully in the next subsection, and Rules 5 and 6 are presented in Section 5.4.
dctcrmining ali points in the s-plane such that the phase of the open-loop transfer
function GH(s) is 11 radians or any angle obtained by 2k11 (where k =
O, l, 2, ... ) radia ns of rotation beyond 11 (in either direction) to yield this RULE 1: Starting and Ending Points
same terminal position. Once the points for which GH(s) = 11 radians in the
s-plane are found, K may be determined by using (5.30) in (5.33) to give Root loci start ( K = O) on the open-loop poles and end ( K --+ oo) on the
open-loop zeros.
Kl(s + z1 )(s + z2 ) 1
~~~~~~~~- =! (5.35)
l(s + P1)(s + Pi)(s + p3) 1
Rule 1 may be proved directly from (5.36). When s = - p 1, - p 2 - p 3 , the
from which value of K is O. Therefore, the root locus plot starts ( K = 0) at the open-loop
poles. Furthermore, since K = oo for s = -z 1, - z 2 , , we see that the root
Is + Pil Is + Pil Is + PJI
K = - - - -- - -- - - (5.36} locus plots end on the open-loop zeros. The number of separate loci is the
Is + zj Is + z 2 1 maximum of the number of finite poles ( p) and the number of finite zeros ( z ).
For most systems of interest, z is less than or equal to p.
In terms of the factors in (5.30) and the corresponding s-plane vectors, we
rcquire from (5.34) that points on the root locus for O < K < + oo satisfy
z p RULE 2: Root Locus Segments on .the Real Axis
/GH(s) = [/.r+z - [/s+p; =(2k+1)11 (5.37)
i= I i=l Root loci occur on a particular scgment of the real axis if and only if there
are an odd numher of total potes and zeros of the open-loop transfer
Moreover, (5.33) and, more precisely, (5.36) yield
function lying to the right of that segment.
Product of vector lengths from potes
K = - -- - - - -- -- - - - (5.38)
Product of vector lengths from zeros
Equation (5.37) may be used to prove Rule 2, which determines the sections of
Alternatively, we could satisfy (5.33) first and then (5.34); both conditions must the real axis where roots occur. lf a point s on the real axis is to the right of a zero
hold regardless of the order in which they are considered. In either case, we could Z; ora pole P;, the s-plane vector drawn to that point from this zero or pote has
an angle of O radians. If a point s is to the left of a zero, the associated s-plane
vector from the zero to the point s contributes 11 radians to the net angle of
1 For the ro/i locus method. we combine the transfer unctions o the (open-loop) plant G(s) and the
GH(s). If the point s is to the left of a pote on the real axis, the associated
eedback controllcr 1/(s) n Figure 5.5 to form thc open-loop transcr function Gl/(s). Henceforth. s-plane vector from the pole to the point s contributes -11 radians to the net
thc tcrms "open-loop zeros" and "open-loop polcs" will be used to reer to the z.cros and poles o angle of GH(s). Consequently, we must have an odd number of the sum of finite
Gll( s ). potes and finite zeros to the right of the point s to yield a net angle of 11 radian!i
)
 250 Stability Analysis in the s-plane The Aqot Locus Method 251

far GH(s). This point s may be anywhere along the section where Rule 2 is
TABLE 5.10 Baslc Rules for Root Locus Constructlon
satisfied. "
I
1. STARTING ANO ENOING POINTS
Root locus plots start (K = O) on the open-loop pales and end (K .... oo) on the RULE 3: Imaginary Axis lntersections
open-loop zeros.
Use the Routh-Hurwitz criterion to determine }"!,-axis crossings of the root
2. ROOT LOCUS SEGMENTS ON THE REAL AXIS locus plots. BoJh the gain K* and the value of w may be found from the
Routh table.
Root loci occur on a particular segment of the real axis if and only if there are
an odd number of total poles and zeros of the open-loop transfer function lying
to the right of that segment.
The s 2 row of the Routh table may be used to determine the value of w for
3. IMAGINARY AXIS INTERSECTIONS jw-axis crossings. Rule 3 is based on the second special case of the Routh table
where jw-axis poles are encountered. Recall that ali elements of a particular row
Use the Routh-Hurwitz criterion to determine jw-axis crossings of the root locus
are zero in this case. We wish to consider this situation when it happens for thc s 1
p lots. Both the gain K* and the value of w* may be found from the Routh table.
row. Thus, the s 2 row may be used to write the auxiliary polynomial as indica ted
4 ASYMPTOTES (FOR p *- z) in Section 5.2. The procedure is to set the entry in the s1 row to zero and sol ve for
the gain K* which yields roots on the jw-axis. This value of K* is then
Root locus plots are asymptotic to straight lines with angles given by substituted into the auxiliary polynomial / 0 (s), which is equal to zcro. U sing
s = jw*, we solve for w*. These operations were dernonstrated carlier in Example
(2k+ 1)'1T
OA = - - - - 5.4. These first three root locus construction rules are illustrated in the examples
p- z . that .follow.

d ~ s approaches infinity. These straight lines intersect ata point o1 on the real
uxis specified by

RCsl + K(s + 1) (s + 2) Y(s)

[ Poles of GH( s) - [ Zeros of GH( s) l: l 1
o,= s(s + 4)
p-z

where p is the number of finite poles and z is the number of finite zeros of
Gll( s ).
(a)

5. I\NGLES OF OEPARTURE ANO ARAIVAL

Assume a point s8 arbitrarily near the pole (far departure) or the zero (for
orrival) and then apply the fundamental angle relationship
s-plane
L Angles of zeros of GH(s) - [ Angles of pales of GH(s) = (2k + 1)'1T

K=O K= > K= > K ;= O

6. O-lEAKAWAY POINTS -4 -2 -1

Orookoway points may be determined by expressing the characteristic equation

s and then solving far the breakaway points s8
l c>r th o galn K as a function of
fr om

dK(s) =O (b )

ds s -s.,
Figure 5.6 Closed-loop system diagram and s-plane
root locus construction for Example 5.5.
252 Stability Analysis in the s-plane The Root Locus Method 253

TABLE 5.11 The Routh Array for Example 5.6

EXAMPLE 5.5 s2 K+ 1 6K- 10

s1 5K- 9 ./'
We plot the location of the potes and zeros of GH(s) from Figure 5.6a in the s 6K - 10 ____--
s-plane of Figure 5.6b. Toe root loci for the closed-loop system start on the poles
and end on the zeros by Rule 1. We determine by Rule 2 that root loci occur on
the real axis between the pole at - 4 and the zero at - 2 and between the zero at
- 1 and the pole at the origin. The complete root locus plot is the simple the poles and between the zeros. From Rule 1, the root loci begin ( K = O) on thl
construction shown in Figure 5.6b. potes at -1 and + 10. As K increases, these loci move toward each other along
the real axis and at sorne point become equal. At this point the roots break away
from the real axis to forro two complex closed-loop poles, as shown in Figure 5.7r
EXAMPLE 5.6 At yet sorne larger value of K , the roots re-enter onto the real axis somewhcn-
between the ~wo zeros al - 2 and - 3. From this re-entry point the two root,
The two poles and two zeros of GH(s) from Figure 5.?a are plotted in the s-plane separatc and move along the real axis toward the zeros as K-+ oo. We reserve thc
o f Figure 5.7b. We find from Rule 2 that root loci occur on the real axis between details of how to compute these breakaway and re-entry points (Rule 6) until thc
next section.
1 Y(s) We examine the Routh array to determine wbether any jw-axis crossings occur.
R(s)
+(t)
I + 2)
K(s
~ lf the breakaway point on the real axis is to the right of the origin, then the two
-- loci move across the jw-axis toward the re-cntry point between - 2 and - 3. On
-- the other hand, if the breakaway point occurs between the origin and -1, thcn
+3 thcre are no jw-axis crossings. The Routh array is given in Table S. l I. Th<:
+1
elemcnt P1.1 is positive for K > 9/5 = 1.8 and Po.i is positive for K > 10/6
-- (a) 1.67. Thercfore, the root locus segment starting at -1 crosses into thc right-half
s-plane al the origin. The root is at the origin for K = 1.61. A breakaway point
occurs for sorne slightly larger K, and the two loci cross the jw-axis moving into
s-plane
... ... s-plane the left-half s-plane for K* = 1.8. We form the auxiliary equation from the s 1
row as

+10 -3 -2-1 +10

(s) =(K + l)s 2 + (6K - lO)IK' =I.R
-3 - 2 - 11
.. .. / 0 (5 .39)

Thus, w satisfies
(e)
(b)
1
s-plane
(K + 1)(jw*}2 + (6K - _JO)IK'='-S = O {5.40)
~, K
w'
= J.8
= 0 .54
+10 from which w* = 0.54 radians. The complete root locus construction is shown 111
~ Figure 5.?d.
"<'--""
As a final remark on this example, we observe that root loci for secon-ordl'1
I systems always occur as straight lines, circles, or segments thereof. Thus. the o, al
1 curve in Figure 5.7d can be shown to be a circle.
(d)

Figure 5.7 Root locus construction for Example 5.6.

254 Stability Analysis in the s-plane The R~ot Locus Method 255

K Y(s)
5.3.3 Asymptotes of Root Loci as K - oo < + u e, + 2>< + 3>
I
(a) ,

RULE 4: Asymptotes (for p * z) '.'

;,..
Root locus plots are asymptotic to straight lines with angles given by ''
'' K = 16
{2k + 1)'11' '' w = l
/
{JA =
p -z
. '' /
/

'' /
/

as s approaches infinity. These straight lines intersect at a point a 1 on the ',1

real axis specified by ',
a, =
L Pales o GH(s) - }'; Zeros of GH{s) ' ',1
p-z
where p is the number o finite pales and z is the number of finite zeros of
GH(s ). (b)

Figure 5.8 Root locus construction for Exam-

ple 5.7.

Rule 4 provides an expression for the angle of the asymptotes as K-+ oo.
When the number of finite pales ( p) exceeds the number of finite zeros ( z ), one EXAMPLE 5.7
or more o f the loci tend toward zeros at oo as K ---+ oo. For example, if
p - z = 3, the denominator of GH(s) is o arder three greater than the numera- The open-loop potes from GH( s) in Figure 5.8a are plotted in the s-plane of
tor, and we have asymptotes of '1T/3, '1T, and - '1T/3. The chart in Table 5.12 Figure 5.8b. By Rules 1 and 2, the loci begin on the four poles, with the segment
should be helpful for simple cases. Rule 4 also gives an expression for the on the left formed by the loci at - 2 and - 3 approaching each other along the
intcrscction of the asymptote's with the real axis. For brevity, we omit the proof of real axis, forming a breakaway point, and moving toward the asymptotes as
Rule 4 hcre; it is based on approximating (5.30) for large K by retaining only s! K---+ oo. The segment on the right between - 1 and the origin on the real axis is
and s ! 1 terms in the numerator and sP and sP I terms in the denominator of formed similarly, as shown in Figure 5.8b.
(,'1/(s). The two examples which follow use the first four root locus construction The asymptotes by Rule 4 (and Table 5.12) are
1ulcs.

(2k + 1)'11' (2k + 1)'11'

(J = - p-- -z (5.41)
A 4
TABLE 5.12 Angles of Asymptotes for Certain Values of ( p - z)

p- z Angles of Asymptotes (OA) which yields 7T/ 4 and 3'11'/ 4 radians. The intersection ( o1 ) of these asymp-
totes with the real axis is given by
o No Asymptote
1 7T
2 w/ 2, -7T / 2 -3-2-1-0 6
3 w /3,w, - w /3 (] I = - l.5 {5.42)
p-z 4
4 '11 / 4, 3w / 4, - 7T / 4, - 3w / 4

5 w / 5, 3w / 5, 7T, - '1T / 5, - 3w / 5
Table 5.13 shows the Routh array. The cntry p 1 , 1 is zero for K* = 10, and w* is
256 Stability Analysis in the s-plane Angles of Departure and Breakaway Points 251

TABLE 5.13 The Routh Array for Example 5.7 R(sl +

l:) .. , K(s + 2) (s + 3) 1 Y(s)
s<s + 1) (s + 5)(s + 6). 1
S4

-vKK_.. ., -
1
6 11 K
' 1

s3 10 6 /
s2
60
s,
10 (n!_
s K
1
1
fo und from the auxiliary equation as 1
1
1
1
fa(s) = 10s 2 + K 1 1 ~~~
1
2 1
10(jw*) + KIK . ,o = O~ w = l (5.43) 1
1 -
1
Therefore. the sketch in Figure 5.8/J is complete except for determining the exact 1
location of the breakaway points (Rule 6 to be discussed in Section 5.4).

EXAMPLE 5.8
o
-6
t OE f
-5 u,= -3.51
,
1
1

-r-
o
-3
<>
-2
-l- '

-1
1

1
1
The root locus plot of closed-loop potes in the s-plane as K vares from O to + oo 1 -
1
is shown in Figure 5.9b for the closed-loop systcm in Figure 5.9a. There are two 1
hrcakaway points and one re-entry point for root loci on the real axis. Two of the 1
loci approach asymptotes o f 7T/2 radians with o 1 = - 3.5, and the others 1
1
approach the finite zeros at - 2 and - 3. There are no jw-ax.is crossings. The oval 1
curve on the right is approximately circular in shape, though not precisely a circle 1
1
since the system is not second order. 1
1
(b)

Figure 5.10 provides a number o root locus construction examples, indicating Figure 5.9 Root locus construction for Example 5.8.
only shapes and omitting details such as jw-ax.is crossings and exact breakaway
locations. Observe that zeros tend to attract root loci toward them and poles tcnd
to repcl them. These plots are included here to encourage the student to ponder
the many possibilities that can occur in root locus construction. A broader class
of problems having complex open-loop potes or zcros is considercd in the RULE S: Angles of Departure and Arrival
following section.
Assumc a point se arbitrarily near the pole (for departure) or the zero (for
arrival) and then apply the fundamental anglc relationship

5.4 ANG LES OF DEPARTURE ANO BREAKAWAY POINTS I: Angles of zeros of GH(s )- t Angles of poles of GH(s) = (2k + 1)77
I
This scction continues with the development and illustration of the root locus
construction rules listed in Table 5.10. We emphasize Rules 5 and 6 in the five Rule 5 is a direct application of the basic angle requirement for s-planc roots o f
examples p resented hcre. the characteristic equation. The angle of GH(s) must be 7T radians or any
 258 Stability Analysis in the s-plane
Angles of Departure anq Breakaway Points 259

j2

s-plane s-plane j2 s-plane

- 1
ji

-- t --- L - jl

jO
-2
r-1 o -4 l -3 o
-2 -1
jO
o
-ji
~ -ji

-j2
-j2
(a) (b)

(e) (/)
I
I Figure 5.1 O ( Continued)

s-plane s-plane Examples 5.9 and 5.10 illustrate angle of departure and angle of arrival calcula-
tions.

-2 -1'
''
-
-1 2
RULE 6: Breakaway Points

Breakaway points may be determined by expressing the charactcristic

equation for the gain K as a function of s and then solving for the
'' breakaway point(s) s8 from
'' dK1s) L.s = O

(e) (d) Rule 6 presents one of many acceptable ways to calculate breakaway a11d
re-entry points. We note that, when two root loci come together along the real
Figure 5.10 Root locus examples involving systems having real open-loop poles axis, there is a relative max.imum value of K at the breakaway point. Conversely,
and zeros.
when two loci have "broken onto" the real axis, that is, formed a re-entry point,
the value of K at that point is a relative mnimum. hus, by forming an algebraic
c4uiva lent angle (by rotation) that has the same terminal side. A point on the root expression for K(s) from the characteristic equation and by setting its derivativc
locus that is an infinitesimal distance away from a particular pole (or zero) is (with respect to s) to zero, we obtain this relative extremum in either case.
selected , and s-plane vectors are drawn from all poles and zeros to that point. We Examples 5.11 and 5.12 demonstrate the calculation of breakaway and re-entry
thcn w,c Rule 5 to compute the difference between the sum of the angles for the points.
zeros and the sum of the angles for the poles and equate this difference to The fifth example of this section illustrates the interdependence of ali six root
(2k + l) 1r, where k =O, 1, 2, .... The angle of departure from the particular locus construction rules. For each of the first four examples, Rules 5 and 6
pole (or angle o f arrival at the zero) is the only unknown in this equation. provide supporting details for root locus construction, but a rough sketch of 1he
260 Stability Analysis in the s-plane Angles of Departure and Breakaway Poi ni, 11111

TABLE 5.14 The Routh Array for Example 5.9

7
'/
s3 4 6
/4
,' s-plane
s2
K* = 20
.,. = 2 .45 ~/
V'
's
20; !5_

s K+4~

/ '/ frl4 radians (45) these asymptotes with the real axis given by
- 1 + j x ' ~ ~ 11

4 /~ 1 I: Poles - l: Zeros ( - 2) + (-1 + J) + ( - 1 - J) 4

u =- ,
3 '-
1r13 radtans (60) <1 =
3- 0
3 (\ 1'i
\ ljO p-z
- 2 ' .- 1 '
The rough sketch shown in Figure 5.11 requires only Rules 1, 2, and 4; the 1()11t
- 1- }
locus that starts on the ple at - 2 moves to the left along the real axis tow:u,1
infinity at an angle of '1T radins, and the root loci that start on the two compln
poles move into the right-half s plane toward asymptotes of '1T/ 3 and - 1r 1
0

radians, that is, 60 and - 60. .

Rules 3 and 5 provide supporting details for thl sketch. In applying Rule 3, wc
form the Routh arry in Table 5.14 for the characteristic. equation given by

K
1+ =O
(s + 2)(s 2 + 2s + 2)
Figure 5.11 Root locus plot for Example 5.9. (5.4(,)
s 3 + 4s 2 + 6s + K + 4 = O
root locus could be made by using only Rules 1 through 4. However, Example
S.13 rcquires the calculation of breakaway points even before the shape of the We sce that the entry in thc s 1 row is zero when K = K * = 20. Using thc
rough sketch of the root locus can be determined. auxiliary equation from the s 2 row with s = jw* gives

2
4(jw*) + (K* + 4) = O (5.4 1)
EXAMPLE 5.9
which yields w* = /6 = 2.45 radians per sccond.
Co nsider the plant transfer function given by Finally, we compute the angle of departure ( O,) from the pote at s = - 1 1 1
Referring to Rule 5, we need to cvaluate
K
Glf(s) = (s + 2)(s2 + 2s + 2)
(5.44) r:/ Zeros of GH( s) - r/ Poles of GH(s ) = (2k + 1)'11' (5.'IX)

We consider a point on thc locus leaving the pole at - 1 + j which is onlv ,111
which appears in a negative feedback configuration, as shown in Figure 5.5. We infinitesimal distance from this pole. The angle Ox is formed by the s-plane Vl'\ 1111
plo t thc open-loop pote locations at - 2, - 1 + j , and -1 - j in the s-planc of drawn from s = -1 + j to this point. The angles of s-plane vectors from ili,1
Figure 5.11 and procccd to sketch the root locus as K vares from O to + oo . other two poles are 'TT/2 radians for the pole at - 1 - j and 'TT/ 4 radians for th,1
We fitst apply Rules 1 and 2 to note that the three roots start al thc given poles pote at - 2. Since there are no open-loop zcros, (5.48) becomcs
and one roo,J. (thc one at - 2) moves along that section of the real axis that is to
the left of - 2. Next, we use Rule 4 to determine that the angles o the asymptotes
for p - z = 3 - O = 3 are +'1T/3, - 1T/ 3, and '1T radians with an intersection of -(X+ i + ) = (2k + 1)'11' ( 'i tq
 262 Stability Analysis in the s-plane Angles of Oeparture and Breakaway Points 263

which yiclds Ox = 11/4 - 2k11. We may set k equaJ to any positive or negative TABLE 5.15 The Routh Array for Example 5.1 O
1nteger in (5.49) to obtain Ox. Using k = O gives O,.= 11/4 radians. Since the root nu
i3 1 "f\ - 2
loci appear symmetrically about the real axis in the s-plane, we observe that the
~ K 2K+ 4
angle of departure from the pole at -1 - j is -11/4 radians.
lnevitably, some students are concerned with whether the root locus crosses the 2K2 - 4K- 4
s,
asymptote of 11/3 radians or always remains on the sarne side of this asymptote. K
Not only does this root locus remain above the asyrnptote, as shown in Figure s0 1 2K + 4
~. 11 , but we can also show the same result for the larger class of linear plants of
s-plane poles at 1 + j and 1 - j and move across the jw-axis to end ( K -+ oo) on
1he form given in (5.44), but with poles al . - a, - 1 + j, and -1 - j for
the zeros at -1 + j and -1 - J. The locus of points- for the third root starts at
1) a < 1 + ff. Moreover, when a > l + ff the root locus that starts at
s = - 2 and moves to the left along the reaJ axis to infinity. Rules 2 and 4 apply
1 l J remains below the asymptote of 11/3 radians. The student should verify
to designate this section of the real axis for the root locus and an angle of 11
tl11.,c.: rcsults, including the special case for a = 1 + {f where the root locus from
radians for the asymptote, since p - z = 3 - 2 = l. With regard to jw-axis
1l w polc at - 1 + j coincides with the asymptote of 11/3 radians. Although this
crossings (Rule 3), Table 5.15 shows the Routh array from which we compute
1\IH' l ).C fo r a restricted form might be interesting, we can show other cases of

il11111-. with more general pole/zero configurations for which the root locus does
K* == 2.73 and w == 1.86 radians per second.
1,1,.,, thc.: asymptote one or more times. We let Ox denote the angle of departure from the pole at 1 + j, as shown in
Figure 5.12. Using Rule 5, we obtain

1 XAMPLE 5.10
E/Zeros of GH(s) - E/Poles of GH(s) = (2k + 1)11
1" 1ll11!>lrate calculations for both angles of departure from poles and angles of
11 1 1\ ,11 at 1.eros, we consider the open-loop plant with transfer function given by
(O + ) - (Ox + i + tan - 1
(})) = (2k + 1) 11
2
K(s + 2s + 2)
GH(s) = (s + 2 )(s2 _ 2s + 2) (5.50)
..o.,,= 2 .03 radians (116.6 ) (5.51)

A ,l..dd1 o f the root locus is shown in Figure 5.12. Root loci start on the right-half
by using k = -1. We designate OY as the angle of arrival at the zero at - 1 + j.
K' = 2. 73 From Rule 5, we have
w' = 1.86
splane

y o, E/Zeros of GH( s) - I:/ Poles of GH( s) = (2k + 1) 11

-1 + j --1-
+ 1 +j

(oy+ i)-(11 + :
3
+ ) = (2k + 1)11
-2

77
-1 - j :.BY= radians (90) (5.52)
+1-j
2
~
which was obtained by using k = O. From symmetry about the real axis, the angle
of departure from the pole at 1 - j is - 2.03 radians and the angle of arrival al
the zero at - 1 - j is - 11/2 radians.
Figure 5.12 Root locus plot for Example 5.1 O.
264 Stability Analysis in the s-plane Angles of Oeparture and Breakaway Points 265

axis occurs al

E Poles - E Zeros (O) +( - 1) + ( - 2)

o, = -----=-1 (5 .54)
p -z 3- O

A rough sketch of the root locus in Figure 5.13 can be made without further
calculations.
Additional details Cor a more precise sketch require a calculation of the jw-axi\
crossings (Rule 3) and the breakaway point (Rule 6). A jw-axis crossing is
obtained for K = 6 al w = ,fi = 1.41 radians per second. In applying Rule 6,
wc first write the charactcristic equation as
-2
K
1+ =0 (5.55)
s(s + 1)(s + 2)

Then we solve (5.55) Cor K as a function of s to obtain

K( s) = -(s 3 + 3s 2 + 2s.) {5.56)

Beca use K ( s) is a relalive maximum al the brcakaway point (s = s n) somcwherc

between - 1 and O, we set dK(s)/ds equal to zero to give

dK(s)I = O = - (3s 2+6s+2) (5.57)

Figure 5.13 Root locus plot for Example 5 .11 .
ds ,-s.

Solving (5.57) gives

EXAMPLE 5.11
- 6 /36----=- 4{2)(3) = -0.42, - 1.58 (5.58)
Wc illust ratc the calculation of breakaway points by using Rule 6 in this cxample.
SB = 2(3)
Let GH(r) be givcn by
Thc value of s 8 belween - 1 and O is - 0.42. Wc note thal the other solution in
(5.58) is a valid breakaway point Cor the root locus plot obtained when K varic\
GH(s) == K (5.53) ovcr the range betwcen O and - oo. Cases involving negative K are considcred in
the following seclion.
whcre the ncgativc feedback configuration of Figure 5.5 is again applicable. Using
Rule 2, we determine that root loci occur on that section of thc real axis between
s = O and s = - 1 and to the left of s = - 2. Observe the similarities and EXAMPLE 5.12
diffcrences between th.is case and the one in Example 5.9. There are three polcs
and no eros in both cases. Both have a polc at s = - 2 and a root locus on the To illustrate both hreakaway and re-entry points in the same root locus sketch, Wl'
scction o f the real axis to the lcft of - 2. While the other two poles were complex consider the linear plant with transfer function given by
in Exarnple 5.9 and hencc required an anglc of departure calculation, the G//(s)
in (5.53)' has polcs only on the real axis. As shown in Figure 5.13, lhe root loci GH(s) = Ks(s + J) (5.'.'i'>)
that start al s = O and s = - 1 approach each other, form a breakaway poinl al
1-omc point s = sR (to be dctermined), and approach asymptotes of 1r/3 radians
and - 1r/3 raciians (Rule 4). The intersection of these asymptotes witb the real with a negativc feedback configuration. There are no asymptotes sincc we haw .111
 266 Stability Analysis in the s-plane Angles of Departure and Breakaway Points 267

- ,-plane

w/2 rad,ans (90")

-1 ss2 sa1 +2

- (a) n = 2 (b) n =3

Figure 5.14 Aoot locus plot for Example 5.12. 1r/4 rad,ans (45)

cqual number of poles and zeros (p = z = 2). According to Rule 2, root loci must
m;cur along thc real axis between the two poles at + l and + 2 and between the
two Leros at -1 and O. As shown in Figure 5.14, the two roots for this
'> ccond-order system start at the poles at + l and + 2, move toward each other (e) n = 4
along the real axis, form a breakaway point (s 8 ,), and proceed along a circular
p.11h into the Jeft-half s-plane. The loci form a re-entry point (s 8 ,) between -1 Figure 5.15 Angles of approach and departure for breaka-
an<l O on the real axis and then separate and move along the real axis toward the way and re-entry points involving n roots.
,eros at -1 and O.
The Routh table can be used to determine the jw-axis crossing for K = 3 at In Examples 5.11 and 5.12, we considered only pairs of roots in the formati on of
w = {i /2 = 0.71 radians per second. To determine the breakaway point and breakaway points or re-entry points, that is, two roots coming together from
1e-cntry point, we find K(s) from the characteristic equation as opposite directions. In such cases these two roots depart from the breakaway or
re-entry point at angles of 1r/2 radians from the directions of approach. In

K(s) = -(s 2
-
s2
3s +
+s
2) (5.60)
general, when n roots come together to form a breakaway or re-entry po1n1 s 8 ,
these II roots depart from s8 at angles of .,, /11 ra<lians from their directi,rns of
approach. Figure 5.15 illustrates cases for 11 = 2, 3, and 4 roots.
Setting dK(s)/ds equal to zero yields

dK 1
ds, - ,.
= - [{s 2 + s)(2s - 3) - (s 2 -
(s 2 +s)2
3s -f 2){2s + 1) l
ls- s
= O (5.61)
EXAMPLE 5.1 3

As a final example in this section, we consider the linear plant transfer fun ction
given by
which is satisfied when
K(s + l)
(5 .63)
4sl - 4s 8 - 2 =O (5.62) GH(s) = s2(s + a)

Therdore, the breakaway point (s 8 ) is the positive root of (5.62) and the re-entry
point (s 8 ) is thc negative root. We find that s 8 , = 1.37 and s 8 , = -0.37, as where a > l, and the negative feedback configuration of Figure 5.5 is used. We
shown in Figure 5.14. apply Rule 4 to determine that the angles of the asymptotes for p - z = 3 - 1 = 2
are .,,/2 and - 1r /2 radians. The intersection of these asymptotes with the real
268 Stability Analysis in the s-plane Angles of Departure and Breakaway Points 269

aplane
11
1
Plane
11


~
lt
-o
(o,~0)
-UJ ..
1

-24.Slr = -
-24.46

~ ~Oouble
1
-o -1
(o = 2) cr, = - o.si 1
polel
1 1
1 1
1 1
1
1
1
JI 11
l.

1
(b)
1
1
1
Figure 5.16 ( Continuecf)
1
Col contour and fo rm a re-entry point on the real axis to the left of - 1. Onc of thcse
roots then movcs to the right along the real axis toward the zero (as K -. oo ).
Figure 5.16 Root locus plots for a = 2 and a - 50
The other root moves to the Jeft along the real axis and meets the third root.
in Example 5.13.
which started a t - a, to form a breakaway point. From the breakaway point o n
axis is the real axis, these two roots move toward asymptotes of 1T/2 and -1r/2 radians
with o, = -24.5.
t Po les - [Zeros (0)+ (0)+{ - a)-( - 1) 1- a We can determine these breakaway and re-entry points for a = 50 by using
o, = = (5.64) Rule 6. We form K( s) from the characteristic equation as
p -z 3- 1 2
2
We use the Ro uth table to determine tha t thcre are no j w-axis crossings. K( s) = _ s (s + a} (5.65)
The shapc of the root locus plo ts in the s-plane depcnds on the value o a, as s +1
\ hown for thc two cases in Figure 5.16. For bo th cases the loci depa rt from the
douhle po lc .i t the origin at angles of 'IT/2 and - 1T/ 2 radians (Rule 5) and then Sctting dK ( s )/ds equal to zcro yiclds
procced into thc second and third quadrants. By Rule 2, that sectio n of the real
a,is betwccn -a and - 1 must contajn a segment of thc root locus plo t.
1lm,c,cr. thc root Jocus sketch for small values of a (c.g.. a = 2) has quite a 0
2
dK (s ) 1 = = - [ (s + 1){3s + 2as) - s 2 (s + a)( l ) l
{5.66)
diffcrcnl shape than the sketch for large values of a (e.g., a = 50). As shown in dr - sa (s + 1)2 1.,-.,,, J
Figure 5. 16a, for a = 2 the root loci move away from the double pote at the
origin and proceed smoothly toward asymptotes of 'IT/2 and -1r/2 radians with
", = - 1/ 2, while the third root moves d ircctly from - a to - 1 a long the real Therefore, thc equatio n for thc breakaway or re-entry poi nts is
axis. Howevcr, as shown in Figure 5.16b, for a = 50, thc root loci that start o n
thc polcs at thc origin movc around the zero a l - J in an approximately ci rcular sn(2sl +(a+ 3}sn + 2a) = O ( 5.<,'/)
270 Stability Analysis in the s-plane Angles of Departure and. Breakaway Points 271

We see that a breakaway point occurs at s8 = O (for K = O). Solving the 1

1
quadratic equation in (5.67) gives 1 s-plane
1
1
-(a+ 3) /(a+ 3)2 - 4(2)(2a) 1
(5.68) 1
So= 1
2(2)
1
1
For a = 50, (5.68) yields values of - 2.04 and - 24.46 for s 8 . Thus, the re-entry 1
1
point in Figure 5.16b is at - 2.04 and the breakaway point at - 24.46. From 1
(5.38), the corresponding values of K are 191.9 and 651.3, respectively. For a = 2, 1
(5.68) gives s 8 = -1.25 j/7/16, which is nota valid breakaway point because -a U= -4 1 -1 JO
the root locus does not even pass through this point. Although it is possible to (a= 9) (Double
1
1 {,/ 1 pole)
obtain breakaway points not on the real axis, such cases are rarely encountered.
1
When these complex breakaway points do occur, we use Rule 6 to determine s 0 1
and then verify by using (5.33) and (5.34) that s = s 8 is a point on the root loci. 1
1
Let us attempt to explain why large values of a (e.g., a = 50) yield a much
1
ditforent shape of root locus plots than small values of a (e.g., a = 2). For 1
" = 50, we obtain a root locus plot for relatively small values of K which 1
1
c~~cn tially neglects the etfect of the pole at - a. Because a is so large, 1he double 1
po lt! al 1he origin and 1he zero al - 1 lend to dominate the root locus plol for 1
~111all K . l lowevcr, as K becomes larger, lhe root thal started al - a moves into 1
1hc vicinity of anolher rool which is moving to the left from the re-entry poinl of
Figure 5.17 Root locus plot for a = 9 in Example 5.13.
thc rool locus plot. In fact, as K -+ oo, lhese two roots tend to dominate the root
lm:us plot. Asan approximation for K -+ oo, we may even neglecl the zero al -1
anti the rool (closed-loop pole) approaching it from the re-enlry point on its left.
two real roots for a > 9. Therefore, the shapes of all root locus plocs for
In hrid, we may think of lhe root locus plot for large a in Figure 5.16b as lhe
1 < a < 9 are similar to the plot in Figure 5.16a, and the shapes of all plo1s for
combination of two plots: one for relatively small K which neglects the pole at
a > 9 are similar to the one in Figure 5.16b. However, the root locus plot
a and another for relatively large K which neglects the zero at - 1 and the
for a = 9 is quite different from either of lhe other two plots. The breakaway
c.lo!>cd -loop pole approaching it from the re-entry point on lhe left.
point al s 8 = - 3 has three rools coming together: lhe two roots that started at
F or small values of a (e.g., a = 2), Figure 5.16a is not easily decomposable
the double pote at the origin and the third root that started at s = - 9. These
into separate parls for small K and large K. We see from (5.64) that lhe
roots approach the breakaway point from angles of 7T, 7T/3, and - 7T/3 radians.
111tcrsection of 1he asymptotes wilh the real axis (o1 ) is determined from the
Moreover, these roots depart from this breakaway point at angles o f O, 2 7T/ 3, and
location of the open-loop zero at - 1 and the open-loop pole at - a, that is,
- 27T/3 radians. These angles correspond to those shown in Figure 5.15b for
o, = (1 - a)/2. This pole-zero pair tends to attract the root loci that start from
n = 3, lhough the loci directions are opposite. As K -+ oo, the root at O radians
thc double pote al s = O because the zero ( - 1) is nearer these poles. Consc-
proceeds along the real axis toward the zero at - l and the other two roots
quently, the root loci move into the second and third quadrants, as shown in
approach asymptotes of 1r/2 and -1r/2 radians, as shown in Figure 5.17.
Figure 5.16a. lf we were to allow a to be 1, then cancellation would occur and 1he
In summary, we have considered the construction of a root locus plot 1ha1
two rool loci from the double pole at s = O would move along asymptotes
requires the calculation of breakaway points (Rule 6) even before the shape of 1he
coincidenl with the jw-axis. lf the pole at -a were to be placed between -i and
root locus plot can be determined. Without the results of Rule 6, wc would not
O, 1lu:n o, would be positive and the root loci from the double pote would move
know whether the plot is similar to the one in Figure 5.16a or that in Figure
into the first and fourth quadrants and approach the asymptotes lhere.
5.16b. In the critica! case, we might even obtain lhe root locus plot chat has a
Figure 5.17 shows the root locus plot for the case of a = 9, an intermediate
simultaneous re-entry and breakaway point involving three roots, as shown in
valut: bt:lween a = 2 and a = 50. This particular value of a was determined from
Figure 5.17.
(5.68) by setting 1he quantity under the radical to zero. We see that (5.68) yields
two complex roots for s 8 with 1 < a < 9, a double root s 8 = - 3 for a = 9, and
272 Stability Analysis in the s-plane Root Loci for Negative K or Positive Feedback 2 73

K( + z) (s + t>zl where is the same as (5.30) and (5.31) for Case ( n) whcrc A O hc ha\ic root
,,;(< + '' ($ ... >21
locus construction rules in Table 5.10 apply to both Cases (a) and ( t')
,; + ,;;,<s+ nJ - <,+PJI(~
(K <0) Modified rules are needed for Cases (b) and (d), becausc thc tlml'cl loop
(K > O)
system transfer function is different from that of Cases (n) an<l (<') 'I he
characteristic equation Cor Case ( b) is

(ni
Use basic
(b)
Use mod ofted K(s + z 1 )(s + z 2 J
rules
rules 1+ =O (5 .71)
(s + P1Hs + P2)
K(, + z) <s + zi)
K(< + zI ( .< + z7l
~ +Pi) (s + ni . -:--
for K < O. Thus, K = - IKI and (5.71) becomes
,.-:-;.~+ ;;;, -
(K ~ Ol
(K , 01 IKl(s + z 1 )(s + z2 )
1- =O (5.72)
(s + P1Hs + P2) ...
(d)
()
which is equivalen! to the characteristic equation for Case (d) when K > O. Wc
Figure 5.18 The four cases of positive and negalive gains and feedback.
may express (5.72) as

5.5 ROOT LOCI FOR NEGATIVE K OR POSITIVE FEEDBACK IKl(s + z. )(s + z1) ... = + 1
GH(s) = (s + p 1 )(s + P2) (5 .73)
Thc hasic rool locus construction rules o Table 5.10 are modified in this scction
to pr<widc s-plane plots as K is varicd rom O to - oo for negative feedback. frnm which
1 hcse modified rules are also applicahle to positive feedba ck, with K varied from
O t<' + oo. The specific changes from the rules in Table 5.10 are that the word jGl/(s)I = 1 (5 .74)
"odd" in Rule 2 is rcplaccd with "cven" and (2k + l)'IT is rcplaced by 2k'1T in
Rule\ 4 and 5. Rules l, 3, and 6 remain the same as beore.
/ GH(s) = 2k'1T (k =O, l, 2, ... ) (5.75)

Thercforc, those basic construction rules that are based on the angle requirernent
5.5.1 Development of Modified Root Locus Rules
must be changed for Case (b)- negative feedback and K < 0 - and Case
(d) positive feedback and K > O. In particular, Rule 2 has thc word "even" in
Cnn . . ider thc our cases shown in Figure 5.18. Case (a), with K > O and negative
place of the word "odd" and 2k'1T replaces (2k + 1)'1T in Rules 4 and 5. Tahle
ccdback, and Case (e), with K < O and positive feedback, have the same
5.16 gives the angles of the asymptotes ( OA) for severa! valucs of ( p - z ), and
characteristic cquation. To show this equivalence, we exprcss the characteristic
Table 5.17 shows the modified root locus rules.
cquat1on for Case (e) as

K(s r z,)(s + z2 ) TABLE 5.1 6 Angtes of Asymptotes for Modlfled Rules

1- = O {5.69)
(s + P1)(s + P2)
p-z Angle of Asymptotes (OA)
S111cc K < O. wc may set K = -1 K . Thcrefore, (5.69) becomes o No asymptote
1 o
- IK l(s + z1 )(s + z2 ) )
l - = o 2 0,7T
I
( (s + P1)(s +Pi) 3
4

2'11/3. - 21r/ 3
tr/2, 7T. - 11/2
!Kl(s + z1 )(s-+ z2 ) 5 O, 2w/ 5, 411 /5, - 2'1T/ 5, 41r/5
1+ =O (5.70)
(s + p,)(.~ + P2)
 Root Loci for Negative K or Positive Feedback 275
274 Stability Analysis in the s-plane

TABLE 5.17 Modlfled Rules for Root Locus Constructlon

s-plane I

1. STARTING ANO ENOING POINTS

Root locus plots start (K = O) on the open-loop poles and end (IKI--+ oo or
K --+ - oo) on the open-loop zeros.

ROOT LOCUS SEGMENTS ON THE REAL AXIS

-3
- 1-
-2 - 1 o
2.
'ss=-1.5
Root loci occur on a particular segment of the real axis if and only if there are
an even number of total poles and zeros of the open-loop transfer function
lying to the right of that segment.
3. IMAGINARY AXIS INTERSECTIONS
Use the Routh-Hurwitz criterion to determine jw-axis crossings of the root locus (a) Example 5.7
plots. Both the gain K" and the value of w" may be found from the Routh table. splane

4. ASYMPTOTES (FOR p * z)
Root locus plots are asymptotic to straight lines with angles g iven by -6 -5 - 4 - 3 -2 - 1 o

2k11
8 = - -
A p- z

as s approaches infinity. These straight lines intersect ata point o, on the real (b) Example 5.8

axis specified by

E Poles of GH( s) - E Zeros of GH( s) splane

O =
p- z
j2

where p is the number of finite poles and z is the number of fin ite zeros of
GH(s) .
ji

S. ANGLES OF OEPARTURE ANO ARRIVAL

Assume a point s 8 arbitrarily near the pole (for departure) or the zero (for jO

arrival} and then apply the fundamental angle relationship -2 -1 o

E Angles of zeros of GH(s} - E Angles of poles of GH(s) = 2k11

-1 1

6. BREAKAWA Y POINTS
Breakaway points may be determined by expressing the characteristic - j2
equation for the gain K as a function of s and then solving for the breakaway
points s 8 from

dK(s} I =0
(d Example 5.9
ds s - s9
Figure 5.19 Root locus plots for negative K.
 s-plane Root Loci for Negative K or Positive Feedback 277
- l+j jl -
Figure 5.19 shows sketches of the root locus plots for Examples 5.7 through
5.11 and Example 5.13 for negative K- that is, as K in Figure 5.18b is varied
,\ from O to - oo for each example. Example 5.7 in Figure 5.19a has asymptotes at
angles of O, 'TT/2, 'TT, and - .,,2 radians. Moreover, Rule 6 can he used to
i
- 2 ~ - 111 -~8 = --0.94 jOO --- +l -
lsn = l.23
compute s 8 = -1.5. Therefore, since o, = -1.5 also, the two loci at - 1 and - 2
move together along the real axis, then break away at - 1.5, and are coincident
with the asymptotes thereafter. Example {8 in Figure 5.19b may be sketched by
using oply Rules 1 and 2. On the other hand, Example 5.9 in Figure 5. l9c
t requires the use of the first five rules. From Table,5.16, O,. is O, 211/3, and - 21r/3
radians. We compute o, as

( -1 + J) + ( - 1 - J) + ( - 2) 4
- l -J -jl o,= (5.76)
p-z 3

The root at - 2 moves along the real axis mto the right-half s-plane. Wc
(d) Examplc 5 . 10
determine the value of K for which the closed-loop pole is at the origin either by
setting Po. 1 = K + 4 = O in Table 5.14 or by u_sing s-plane vectors, that is,
~
~
~ s -plane Product of s-plane vecto/lengths
~ IK I = from polcs to origin
,\ \
\ = 1- 1 + JI 1- 1 - JI 1- 21 = 4 (5.77)
\
\
\ Thereforc, we conclude that K = - 4, since only the negative range of K is heing
n = - 1.6 \
\ considercd here. The angle of departure rom the pole at - 1 + J is - 31r/ 4
\ radians. The difference in this angle and that obtained by using the basic root
-2 1 /- 1 o locus construction rules is .,, radians.
I Figure 5.19d shows the negative-K root locus for Example 5.10. Loci leave the
I
I two complex potes and form a re-entry point on the real axis. One closed-loop
I pole moves to the right along the real axis toward the asymptote of O radians, ami
I
I the other moves to the left to meet the locus coming from the open-loop pole at
I - 2. They break away from the real axis and move toward the two complex
I
open-loop zeros. From Rule 5, the angle of departure from the pole at + l + j is
'
/ - 1.11 radians ( - 63.4), and the angle of arrival at the zero at - 1 - J is - "'/2
' radians ( - 90). Both of these angles are 1r radians (180 ) away from thc
'
respective angles obtained from the bsic rules for K >O.Figures 5.19e ami 5.19/
(r) faample 5. 11
are straightforward to interpret and as such require no further explanation.
,-plane

EXAMPLE 5.14

- a - 1 o Sketch the locus of closed-loop poles in the s-plane as K varies from - oo to + rYJ
(Double for the system of Figure 5.20. Provide ali details, including exact breakaway
I pote) points.
We will use the basic root locus rules in Table 5.10 to construct the sketch frn
(/) Example 5 . 13 O < K < + oo and the modified rules in Table 5.17 for - oo < K < O. In hoth
Figure 5.19 (Continued) Root locus plots for negative K.
275
278 Stability Analysis in the s-plane Root Loci for Negativa K or Positive Feedback 279

1
R(s) + K(s + 1) Y(s)
1
l:) 1 ---;r- 1
1 splane
.' I
1
1
1
Figure 5.20 System for Example 5.14. 1
- 1
1
cases, we have the open-loop transfer function given by ( Triple 1
pole)
~
1
-1 \cr1 = o.~
GH(s) = K(s + 1) (5.78) 1
s 1
1
An odd number of total poles and zeros les to the right of the real axis section 1
hctween - 1 and the origin. Therefore, according to the basic rules (K > O), one 1
1
of the triple poles at the origin moves to the left along the real axis toward the (K > O) 1
open-loop zero al - 1. The other two break away from the axis immediately al the 1
m igin, as shown in Figure 5.21a. We wish to determine the angles of departure 1
1
from this triple pole. Using Rule 5, we have 1
(o)

E/ZerosofGH(s) -E/PolesofGH{s} = {2k + l)w

O - 3/J, = (2k + l)w

O=
x
(2k + l)w
- 3
= -
w w
w --
3' ' 3

T hus, the departure angle of w radians already found by Rule 2 has now been
(5 .79}
- s-plane

1cconfirmed. The other two poles depart at angles of +w/3 and -w/3 radians.
By Rule 4, these two loci approach asymptotes of +w/2 and - w/2 radians with sn=-l.5 -1 o
thc intersection of the asymptotes, with the real axis given by (Triple
pole )

o =
1
O+ O + O - ( - 1)
p- z
=-
1
- =+ -2
3- 1
1
{5.80) - <K < O)

Thc complete sketch for K > O is shown in Figure 5.21a.

(b)
For K < O, we note that the modified Rule 2 requires that an even number of
total potes and zeros must be to the right of a permissible section of the real axis.
Figure 5.21 Root locus plots for Example 5.14.
Allowable sections le to the left of -1 and to the right of the origin. Thus, we
expect the triple pole at the origin to break apart, with one pole moving along the
positive real axis, and the others to move off the real axis and form a re-entry
point onto the real axis to the left of -1 . We compute the angle of departure
280 Stability Analysis in the s-plane Root Loci for Negative K or Positive Feedback 281

1 The re-entry point s to the left of -1 may be found by Rule 6 as

1

ji
1
1
1 .~-plane -dK(s)'
- = -d [ - -s 11
3
= - (s + 1)3s
2
- s1
(1) Ji _
- O (5.84)
(s + 1)2

- 1 ds s= .n ds s +l . s .ra . J=.fn

1
1 from which s 8 = - 1.5. We note from Taole 5.16 that the angles of the asymp-
1
1 totes are O and '1T radians. Therefore, after re-~ntering onto the real axis at
1 Sn = -1.5, one closed-loop pole moves toward infinity along the ncgative real
axis and the other moves toward the open-loop zero at - l. The third closed-loop
-2 -1.5 -1 0.5 pole, as we have already noted, moves along the positive real axis toward infinity.
1 The sketch for K < O is shown in Figure 5.21b.
1

- 1
A composite of the two sketches for K > O and K < O is shown in Figure 5.21c.
1 We can trace the movemcnt of the closcd-loop poles as K is varied from - oo to
1 + oo by reversing the previous dircction for the ncgative K portion of thc plot.
1 that is, by starting at K = - oo and proceeding toward K = O. On reaching
-ji 1 K = O, wc arrive at thc open-loop poles. The remainder of the movement utilizcs
1
1
thc basic root locus plot for K > O, which may be traced in thc ordinary direction
1 used in its construction. Observe that we started on the lcft of the open-loop 7.ero
1 at - I for K = - oo and arrived back at its right side for K = + oo with an angle
(el

Figure 5.21 ( Continued) Root locus plots for Example 5.14. s- plane

fro m thc triple polc using the modificd Rule 5 as

" rad,ans (60)

r,/Zeros of GH(s) - r,/ Potes of GH(s) = 2br (n = 3)
3
(Triple
pole)

O - 30X = 2br

2k1r 2,r 2,r

0 = -=0 - -- (5.81)
x -3 ' 3 ' 3 (al

We write the characteristic equation as s-plane

K(s + l) T radians (90' )

1+ 3 =o (5 .82)
s Sf1: - J.5 - 1 o

(n = 2)
Solving for K(s) yiclds

(b)
-sJ
(5.83)
K(s )=(s+l) Figure 5.22 Breakaway points for Example 5.14.
282 Stability Analysis in the s-plane Root Loci for Negative K or Positive Feedback 283

difference o 1r radians. This characteristic o an angle dilference o 1r radians Ks Y(s)

also distinguishes the modified rules o Table 5.17 from the basic rules o Table (s - 2)(s + 1O)(s + 20)
5.10.
Observe that the tracing of the closed-loop pole movement as K varies from
- oo to + oo makes apparent the triple re-entry and immediate breakaway point
al the origin as K passes from negative to positive values. For the triple pote, the (a)

direction of approach toward the origin for negalive K is O, 21r/ 3, and -21r/ 3
radians, as shown in Figure 5.22a. The angles o breakaway (departure) from the
1
origin are 1r /3, 1r, and - 1r /3, ditfering by '11/3 radia ns (for n = 3), as shown in 1
Figure 5.22a. On the other hand, the breakaway point at - 1.5 involves only two 1 (K > O) s-plane
closed-loop poles and the angular difference in direction of approach and 1
departure in '11 /2 radians, as shown in Figure 5.22b. 1
1

t 1
EXAMPLE 5.15

T he closed-loop system in Figure 5.23a represents a simplified model of a large

rockel in its vertical position just prior to launch. The supports that held the
-20
sa =
----
14.9

i
---
1 u = - 14

- 10
' +2

rocket in place earlier have been removed. The resulting unstablc mO<lel behaves
much as an invcrted pendulum, ready to crash to the ground with the slightest
disturbance. We sketch the root locus for K > O (basic rules) in Figure 5.23b and
for K < O (modified rules) in Figure 5.23c. We note that the anglcs o the
a:,ymptotes (OA) are +1r/2 and -1r/ 2 radians for K > O and O and 1T radians for
K < O. In Figure 5.23b, o1 is -14. Observe that one closed-loop polc is in the (b)
nght-half s-plane for ali K between - oo and + oo. Thus, we cannot stablize this
un:.table syslem by simple gain adjustment.
The power of the root locus method for design is exhibitcd by an analysis o the ~-planc
modificd feedback strategy o Figure 5.24a. We propose to feed back the output (K < OJ
Y(s) through the transfer function (s + 1)/ (s - 1) to combine negatively with
R(s). The corresponding GH(s) is given by -20 - 10 OI +2

Ks(s + 1)
(5 .85) (e)
GH(s) = (s _ l)(s _ 2)(s + lO)(s + 20)
Figure 5.23 Closed-loop system diagram for a rocket prior to
launch and associated root locus plots.
Thc basic root locus rules are applied for K > O in Figure 5.24b. The importan!
difference from Figure 5.23b is that the loci starting at the two right-half s-plane
open-loop potes at + 1 and + 2 come together, break away from the real axis
hdwccn the two open-loop zeros, and proceed along the real axis toward those solution corresponds to the balancing of an inverted broom on the palm o your
two zeros. The closed-loop system is asymplotically stable for K > K* , where K hand by moving the extended arm in a direction to restore any imbalance.
and thc associated w* can be obtained from the Routh array (Rule 3). The root Jmplementing this algorithm can prove to be an interesting project for the control
locus plot for K < O is given in Figure 5.24c. As in Figure 5.23c, the closed-loop engineering student. A "broom" or pote, resting on a small cart, can be
system is unstable for ali K < O. constrained by wires so as to allow movement in only one plane. As thc broom
Wc have stabilized the original system in Figure 5.23a by placing a second begins to lean toward one side, the angle and motion are senscd and the cart can
open-loop pote in the right-half s-plane, as shown in Figure 5.24a, and adjusting be made to accelerate suddenly in the same direction until balance is restored.
the gain K to yield ali four closed-loop potes in the left-half s-plane. This
284 Stability Analysis in the s-plane Root Loci for Negative K or Positive Feedback 18!\

-
R(.,) +
(s - 2) (s
Ks
+ 1O) (s + 20)
Y(s)
R(s) + K(s + 1)
s -;. 1
Y(sl

s+ 1 (al
s -
-(a)
~,-plane

1 1 s-plane ... +l
...
1 <K = - 1)
- 1 (K = -1)
1
1
1
1 (b)
1
t
S

-?O - -
= - 14.8 1
1"1 =
1
- 13

- 10
Figure 5.25 An illustration of asymptotes for
flnite K.

5.5.2 Asymptotes for Flnite K

t 1 A specal situation can occur for negative-K root loc when p equals z. To
\ 1
1
<K , O)
illustrate this situation, consider the simple case in Figure 5.25a. We may express
the closed-loop transfer function as

K(s + 1)
(h )
Y( s) G ( s) s_ 1
R(s ) = l+G(s) = K(s+ l)
1 + -- -
s- 1
s-plane

K(s + 1)
(5 .86)
- (K+})s+(K-1)
o
-20 - 10 - 1 I 1 2
(K < Ol Therefore, we see that for K = -1, Y(s)/R(s ) becomes

Y( s) - ( s + 1) l (5 .ln)
(e)
R(s) = _2 = 2(s+ l)
Figure 5.24 Modified design for Example 5.15.
The system order is 1 if K 1' -1 but is reduced to zero for K = - 1. From (5 .86) ,
the characteristic equation is given by

(K + l)s + (K - 1) = O (5 !IX)
 286 Stability Analysis in the s-plane Root Contours 287

s-plane
s-plane
zeros, and finally move along the axis toward the two zeros. Another possibility is
for the locus from the left pole to move around to the right-half s-plane, from a
breakaway point there, and so forth. The decision between thes~ cases Q1ay be
determined by computing the locatio11 of the breakaway point s8 as
o---<!> rf )( H -.
= -1)
). 0----0 e: ., 1 "
= - 1)

)-1 (K )1 ..
(K +I +2 =
= !!_( -s(s + 4)
<K = - lJ -3 -2 - 1 + 10 (K -IJ
dK(s) )1 =0
ds , - ,a ds s 2 + 3s.;t- 2 s - s8

(a) Example 5. 12 (b) Example 5. 6 .'. - s + 4s 8 + 8 = O (5.90)

- s plane
from which s 8 = 2 -lff = - l.46 and + 5.46. Figure 5.26c shows the desired
root locus plot. Observe that both breakaway and re-entry points are given by
(5.90).
sa = - 1.46 58 = + 5.46

:( H ~ ~ 1) 1 1 1 r: 1
(K = - 1) - 4 - 2 \ -1 IO (K = -1)
5.6 ROOT CONTOURS

The parameter to be varied in root locus construction may not necessarily be the
open-loop gain K as in ali previous examples of this chapter. We show in this

--
{e) Example 5. 5
section how to vary other parameters that appear linearly in the numerator or
denominator of the open-loop transfer function Gf/(s ).
Consider the system of Figure 5.27a with the parameters K and a to be varied.
We set a to sorne desired (nominal) value and proceed to vary K from - oo to
Figure 5.26 Negative-K root loci for examples considered previously.
+ oo according to the procedures already established in Tables 5.10 (basic rules
which yields foc K > O) and 5.17 (modified rules for K < 0). Now suppose we want to set K a t
sorne nominal value and vary a from - oo to + oo. We can apply 1he prcvious
1- K rules if we are able to determine another closed-loop system that has the same
s =- - (5.89)
I+K

Thc root locus for negative K is shown in Figure 5.25b. Rule 2 of the modified
, u les yields roots on the real axis to the right of the open-loop pote at + 1 and to
thc le ft of the open-loop zero at - 1. We see from (5.89) that the closed-loop pole
--
R<s) + KN 1CsJ
D CsJ
Y( s)

actually moves along the real axis to the right of + 1 toward infinily and then N 2(s_) _ _
moves from infinity on the negative real axis toward the zero at - l. One might Di(s) + aD3(s) .
have prcviously asked whether closed-loop potes can transfer from the right half (u)
to thc left half of the s-plane (or vice versa) without crossing the jw-axis. The
system in Figure 5.25a provides a case where this transfer is possible without a YnewCs)
0D 1(sl DJ <sJ
li ni te crossing of the jw-axis.
D 1(slD2(sl + KN (slN2<sl
Figure 5.26 shows the negative-K root loci for the systems of Examples 5.12,
5.6, and 5.5. The first two of these examples are straightforward applications of IGl-l11e,..(s)I

Rules 1 and 2. In both cases, K = -1 is the value of K for which an infinite

closed-loop polc is obtained for finite K. Example 5.5 in Figure 5.26c presents (b )
two possible cases. In one case, the locus may move from the right-half s-plane at
inlinity on the real axis around to the left-half s-plane, then form a breakaway Figure 5.27 Two systems having the same character-
point, move in a circular are to form a re-entry point between the two open-loop istic equation.
 288 Stability Analyss in the s-plane Root Contours 1RU

charactcristic equation as thc given system. Moreover, thjs second system must
have thc parametcr a appearing in the open-loop gain position. Thc characteristic (tt~ Y(s)
s(s + 2{..,n)
equation for the system of Figure 5.27a is

1
KN1 (s))( N2 (s) )
(5.91)
+ ( D 1(s) D2 (s) + aD3 (s) =O (a) -

which may be simplificd to yield ((2w.) YnewCs )

52 + Id~

D1 (s)D2 (s) + aD 1(s)Dis) + KN1 (s)N2 (s) = O (5.92)

To place a in thc open-loop gain position, we divide (5. 92) throughout by the sum
of ali terms not containing a, to obtain (b)

Figure 5.28 Two second-order systems hav-

aD 1(s )D3 (s) ing the same characteristic e!'.luation.
l+ =O (5.93)
D1 (s )D2 (s) + KN1(s )N2 (s) . .~
Dividing (5.96) throughout by the sum o ali tcrms not containing r gives
Thus, wc have formed a second systcm that has an equivalen! charactcristic 2wns
equation. Thc new open-loop transfer function is given by 1 + (r) 2 2 = o {5.97)
s + w.
aD1 (s)D3 (s)
(5 .94)
Thercfore, thc new open-loop transfer function with r in the forward gain
GHntw(s) = D(s )D2 (s) + KN1(s )Ni(s) position is

A closed-loop diagram for the second system is shown in Figure 5.27b.

GHncw(s) = f(2w.s) (5.98)
si+ w2
The two output responses of the systems are generally dilferent, since the "
vcctors b and e in their state representation are not the same. The eigenvalues of
This GHncw(s) is used in Figure 5.28h to yield a new closed-loop system, which
the matrix A, and hence the stability properties of the systems, are identical. has the same stability properties (characteristic equation) as the one in Figure
5.28a.
Root locus plots for f > O (basic rules) and f < O (modified rules) are shown
EXAMPLE 5.16
in Figures 5.29a and 5.29h, respectively, and the combined plot is shown in
Figure 5.29c. For the combined plot, the olosed-loop poles for f = - oo are at thc
As a first cxample, we want to plot the root contour as the damping coefficient r open-loop zero at thc origin and at + oo on the positive real axis. As f increascs,
for thc sccond-order system in Figure 5.28a varies from - oo to + oo. As in
these roots move together to forro a breakaway point at s 11 = w,, for f = - 1
(5. 91 ). wc first form the characteristic equation as
(Rule 6). Thc Ioci then move in a perfect circtearound the open-loop zero at thc
origin, produce a jw-axis crossing (Rule 3) for f* = O and w* = w,,, and form a
w; =O (5.95)
re-entry point on the real axis at s 8 = - w,, for f = + 1 (Rule 6). One root thcn
1 + s(s + 2fwn) proceeds along the real axis toward the zero at thc origin and the othcr moves to
the left along thc real axis to the asymptote of 11 radians (Rules 2 and 4 of basi<
'
wh ich simpl!fies to yield
rules). The closed-loop system is unstable for f < O, marginally stable for f = O.
and asymptotically stable for f > O. Rccall thal in Section 4.2 we furlh('I
classified the asymptotically stable time response as underdamped for O < < 1, r
s2 + 2fwns + w~ = O (5.96) critically damped for f = 1, and overdamped for f > l.
 ,?

290 Stability Analysis in the s-plane Root Contours 291

~ (=O

JWn
s-plane
jw,,~
'= o splane

((<O)

., J
.,e
((> O) '= -1
..
a.
(= 1
~ ~
-~ o
- <-
~
""-"-* -jwn -Jwn

~
- 1
1
~
(a) (b)
N
,-..
e= o s-plane
1
r-
1()
w = wn i:i: Q)
~
~ ~
1
a.
E
~ ~ '=
<O
><
w
-1 ...
\O ~
'= y
N

ll'l' ~
1
.. s s
::,
o
eo
~ o
(e) o
~
u
e
Figure 5.29 Root contour plots tor Example 5.16. <O
E
...
<O

...,
e
O>
.~
u
EXAMPLE 5.17
~li I;~ 1~
.
a. E
Q)
;
>,
A~ a preview of series compcnsation design by root locus procedures, consider the >
-. ~
.>
(J)

m:gativc unity-feedback system shown in Figure 5.30a. A phase-lead series -,

o
---+1 o
-
- -
1 _
M
rnmpcnsation network has been inserted into the loop with the linear plant to
y1cl d a combined transfer function given by
:s t 1 ~ lli
o f::,
,3.,
o
"'11 t ..... "'11 O)
.. ,:, N
1
- ::.:

=es: )L(s: 2)) +

., +., ::.:
c., 1
11
Gpp(s) :z (5.99) ~

whcrc thc compensation zero is placed at - z and the pole at - az for a > l .
Whilc a tlcsign procedure described fully in Chapter 7 determines values of a and
z to :.atisfy prespccilied design criteria, our analysis here focuses on the construc-
1,on of root contours for K = 40 as a vares from 1 to 6 and z from 2 to 6.
Figure 5.30b shows the root locus plot without compensation-that is, G.(s) =
1 as K varic.::s from O to + oo. The points on the loci where K = 40 are
c,ktcrminetl as s = - l Jf'S9. Root con tours for K = 40 are determined in
Figure 5.30c as a varies and in Figure 5.30d as z vares. These contours are
292 Stability Analysis in the s-plane Referencui,

obtained by rearranging the characteristic equation for Figure 5.30a to oht.1111

zs(s +-2) ]
.,e: l+a [ =O
"' s 2 (s + 2) + 40(s + z)
a.

11
~
-.
+
~
''
11 II
o
\D<D
11 11
~
-.
,-
'
and

1 + z[ 1
2
a(s + 2s +:40/a)
s (s + 2) + 40s ,
l =O
(j .... tj ~\() t\!) "'d
(j Q)
a. The basic root locus construction rules in Table 5.10 are then applied to(~ 1(~11
E
(O and (5.101) to obtain the desired plots. The combined root contour cu,vn 11
)(
~ w Figure 5.30e identify the region in the s-plane corresponding to these van a111111,
... in both a and z .
.E
...
en
:,
B
e:
o
N<O NlO o
11 11 1111 oo
"(j ... (j
... SUMMARY
'O
e:
<11

E The stability of linear control systems has been ex.amined in this chapH, l,1
...
<11
considering s-plane and time-domain techniqucs. We began by stating thc d('h111
.~ tons of stability in the time domain and then relating these definition, 1,,
., 'O
procedures in the s-plane. We showe<l that the Routh-Hurwitz. criterion prov1cl,

li ij
e:

t
"' E
~ ~
Q)
1i
a necessary and sufficient condition to determine whether a gven linear. tinu- 111
~

~ variant feedback system is stable. Ali closed-loop potes must lie in the lcft 11111
.> -.1
>
l~ -.
-1
+
- ~-.,_ s-plane for asymptotic stability, but sorne may appear as single poles 0 11 th,
-. 1 ~ jw-ax.is Cor stability (but not asymptotic stability). Construction rules fo r ,,,.,
-+
-1
>.<
:::,
~
locus were presented in tables to aid in determining closed-loop pole Iocati<111. ,n
e: the s-plane as the loop gain K varics from - oo to + oo. Root contours 1111,
c3
.._, sketched for the variation of other parameters by first deriving ncw closcd '"'
o systems having the same characteristic equations and thcn applying thc~r ' "'"
~
~ locus construction rules.
"'
!::, ,...:
o,,....
: 1()
REFERENCES

t.R. W. Brocketl. Finite-Dimensional Linear Systems. New York: Wilq. Pl/rt

2.J. E. Gibson. Nonlinear Automatic Control. New York: McGraw-1lill . 1% 1
Chapter 8.
3. J. C. Hsu and A. U . Meyer. Modern Control Principies and Applu1111,;11

New York: McGraw-Hill, 1968, Chapter 11.
4. R. E. Kalman and J. E. Bertram. "Control Systcm Analysis and Jk,1p11 \ , 1
the Sccond Method of Liapunov." Journal o/ Basic Engineeri11n, , ,,,.,
actions o/ the ASME (D), Vol. 82, (1960), pp. 371 - 400.
 294 Stability Analysis in the s-plane Problems 295

5. J. LaSalle and S. Lefschetz. Stability by Liapunov's Direct Method. New (5.1) 5.2 Let the impulse respnse function g(t) for a linear time-invariant
York: Academic Press, 1961.
system be given by '
6. K . S. Narendra and J. H. Taylor. Frequency Domain Criteria for Absolute ~- t
Stability. New York: Academic Press, 1973. g(t) = e - 1 - e - 2,
7. Z. V. Rekasius. "Lagrange Stability of Nonlinear Feedback Systems." IEEE
Transactions on Automatic Control, Vol. AC-8, No. 2 (April 1963), pp.
160-163. a. Prove that the linear system satisfies Equation (5.3).
8. J. L. Willems. Stability Theory o/ Dynamical Systems. New York: Wiley, b. Determine the system transfer unction, locate the poles in the
1970. s-plane, and verify asymptotic stability.
9. A. Hurwitz. "ber die Bedingungen, unter welchen eine Gleichung nur c. Express the system in direct phase variable form, determine the
Wurzeln mit negativen reellen Teilen besitzt." Mathematische Annalen, eigenvalues of the matrix A, and prove that the system is
Vol. 46 (1895), pp. 273-284. asymptotically stable from inspection of these eigenvalues.
10. E. J. Routh. A Treatise on the Dynamics o/ a System o/ Rigid Bodies. (5.2) 5.3 For the following characteristic polynomials for closed-loop sys-
London: Macmillan, 1877. tems, construct the Routh table and then apply the Routh-Hurwitz
11 . J. J. D'Azzo and C. H. Houpis. Linear Control System Analysis and Design: criterion to determine the number of closed-loop poles in the
Conventional and Modern, 2nd ed. New York: McGraw-Hill, 1981. right-half s-plane, the number in the left-half s-plane, and the 1111
12. W. R . Evans. Control System Dynamics. New York: McGraw-Hill, 1954. number on the jw-axis. In each case, specify whether the closed-loop
13. B. C. Kuo. Automatic Control Systems , 4th ed. Englewood Cliffs, N.J.: system is asymptotically stable, marginally stable, or unstable.
Prentice-Hall, 1982.
a. /(s) = s 4 + 3s 3 + 2s 2 + s + 5.
b. /(s) = ss + 2s 4 + s 3 + 2s 2 + s + l.
c. /(s) = ss + 2s 4 + 2s 3 + 4s 2 + 2s + 4.
PROBLEMS d. /(s) = ss + 2s 4 + 2s 3 + 4s 2 + s + 2.
(5.2) 5.4 For the system given in Figure 5.32, how many open-loop poles are
(5.1) 5.1 The poles and zeros of four transfer functions are plotted in the
in the right-half plane? Determine the range of K such that the
s-plane in Figure 5.31. Which of the systems are asymptotically
closed-loop system is asymptotically stable.
stable, which are marginally stable, and which are unstable?
1111!
s-plane
1 s-plane ~ + 1 K (s + 1) Y (s)
X 1 X (i) )1 ~ +"2

X
- - x --x --x--o-
l -
X 1 1
X

l Figure 5.32 Systern diagrarn for Problern 5.4.

(a)
(b)

s-plane (5.2) 5.5 Consider the system in Figure 5.33.

1 s-ptane
X
o a. Is the open-loop plant stable? Given your reason.
' Doubte

X
pote b. For K = 4, is the closed-loop system stable?
- - x -- x
l
X- -

...-Doubte ll(s) + (i) j K (s + 6) Y(s)

j pote o )1 l! )lo s3 + 4s2 + s - 6

(e)
(d )

Figure 5.31 Four pole/ zero configurations in the s-plane for Problern
5. 1.
Figure 5.33 Systern diagrarn for Problern 5.5.
 296 Stability Anlysis in the s-plane Problems ,{ti

c. Determjne the entire range of K for which the closed-loop (5.3) 5.10 Provide rough sketches of the root Jocus plots for K > O for w,,li'tit
system is asymptotically stable. having the followng open-loop transfer functions. Assumc w~
(5.2) 5.6 tive-uruty feedback in each case. Do not solv~ explicitly for :m~.11
Use the Routh-Hurwitz criterion to determine the range of K for
of departure or arrival or for breakaway points (Rules 5 a111I r,1
asymptotic stability for the following open-loop systems when
However, you are asked to indicate jw-axis crossings and thc an~I
placed in a negative unity-feedback configuration.
and real-axs intersections of t~ asymptotes (Rules 3 and 4).
2
K(s + 1) K
GH(s) = s(s - 2)(s + 3) a. GH(s) =
s(s + l}(s + 5)
K(s + 2) K(s + 1)
b. GH(s) = .
(s + 1) b. GH(s) = s(s _ 2)
(5.2) 5.7 Find the range of positive K for which the system of Figure 5.34 is K(s - 1)
stable. c. GH(s) = ( s+l )( s+2 )( s-6)
R(s) .,.+{i:) ,1 , K<s + 10)2 1 Y(s) , (5.3) 5.11 Sketch the root locus for the system of Figure 5.37, and specify 1111
sZ(s + 1) (s + 100) 1 number of closed-loop poles in thc right-half s-plane as K va11,
from O to oo. (Do not find exact locations of breakaway poin1., 11
any.)

Figure 5.34 System diagram for Problem 5.7. R(s) + '.i:) )1 s + 5 1 Y{s) .,.
:& (s -ll(s-3) 1

(5.2) 5.8 Somctimes it is possible to stahilize an unstablc open-loop plant by

using negativc feedback.
a. Find the range of K for which the closed-loop system of Figure K(s + 1)
5.35 is stable. s - 10

b. Determine that value of K Cor which closcd-loop poles are

located on the jw-axis in the s-plane, and specify their exact Figure 5.37 System diagram for Problem 5.11.
location on the jw-axis.
(5.3) 5.12 Give a rough sketch of the root locus, that is, the locations of 11t,
R(.~) + r.:.... 1 1 Y(s) closed-loop potes, in the s-plane as K varies from O to + oo. Yn11
----,,)'"lll:J .,. (s + 5) Cs - jz 1 )
are not to find breakaway points, angles of departure or arrival, 111
jw-axis crossings. Only the general shape is required for each 111111
locus plol. Assumc ncgative feedback Cor each systcm.
K(s + 2)(s + 3)
a. = (s + 4)(:S,2 - 2s + 2)
G(s)H(s)
Figure 5.35 System d iagram for Problem 5.8. Ks
h. G(s)H(s) = (s + 3}(s2 + 2s + 2)
(5.2) 5.9 In lhe first quadrant only of the K versus T plane (K > O, T > O),
4K
indicate the region of stability for the system given n Figure 5.36. c. G(s)//(s) =
2 2
s (s + I) (s + 2)
R(s) + G:) 1 Ks
2
- 1 Y (s)
K(s 2 + 2s + 2)
) "" lo s3 +s + (1 - 2T) 1 ) d. G(s)H(s) = - - - - -
s
(5.3) 5.13 Sketch thc root locus of the closed-loop polcs in the s-planc frn 11t,
system of Figure 5.38. Determine ali jw-axis crossings. but d 11 11,"
specify exact breakaway points. K varies from O to + oo.

Figure 5.36 System diagram for Problem 5.9.

J
 298 Stability Analysis in the s-plane
Problems 299
R(s) + KI - 21 Y (s)
.-;-+I) b - 10) R(s} + K(s + 12 Y (1)
l:) .. 1 --;2~ -=--i"'

Figure 5.38 System diagram tor Problem 5. 13. Figure 5.41 System diagram for Problem 5.17.
(5.3, 5.14
For the open-loop transfer functions given below, sketch the locus
'i.4) provide these and ali other pertinent details associated with your
of the closed-loop potes as K vares from O to + oo. Assume
root locus plot.
negative-unity feedback. Give ali pertinent information, except the
exact location of breakaway points, if any. (5.3, 5.17 Investigate the stability of the closed-loop system in Figure 5.41 for
K(s + I)(s + 2)(s + 3) 5.4) positive K by a detailed application of the root locus method. You 11
may omit the calculation of any breakaway locations.
G(s) = (s + 4)(s + 5)(s + 6)(s + 7)
K(s + I) (5.3, 5.18 Give a detailed plot of the root locus for the systems shown in
b. G(s) =s_(_s___l_) _(s_+_3_) Figure 5.42, and specify the number of closed-loop poles in the
5.4)
C!:i~'U, 5. 15 right-half plane as K vares from O to + oo. Find ali breakaways,
'l.4)
a. Sketch the root locus for the system in Figure 5.39 for K > O.
b. Determine the angle of arrivaJ of the locus at the zero at angles of departure or arrival, ar i jw-ax.is crossings.
s = +j.
c. Show that the root locus passes through the poin t s = + 1.0 +
JO, and calculate the value of K for which it does. RM + K Y(s)
,2+i; + 2

R(s) + K(,2 -t 11
l:) ),,1 - ;;2-::- 4)
Y(s)

s- 1
< + 3)(, + 4)

---
(al

Figure 5.39 System diagram for Problem 5.15.

R(al _+F-\ ~+l) 1 Y(s)

(5.3, 5. 16 -;-:-j"jl
Consider the feedback control system in Figure 5.40. Determine the
5.4)
root locus plot in the s-plane of the closed-loop potes as the gain K
vares from O to + oo. Determine the location of ali breakaway
points (if any) and any angles of departure or arrival (if ap-
propriate). Determine all Jw-axis crossings (if any), and specify the
(b)
value(s) .of K and w for which these occur. In other words,

R(,) + r:) R(sl Y(s)

l: ,. j KI + l) Y(sl
l~-IJ Is- 51

- 1
;:: 2
Figure 5.40 System diagram tor Problem 5.16. (e)

Figure 5.42 Closed-loop system d109roms for

Problem 518
 Problems :iQt
300 Stability Analysis in the s-plane
R(,~) + (t) 1 K (s + 3) 1 Y (s )
R(s) KC2 - s) Y(sJ Jo l: _., s(s2 + ? s + 2) 1 Jo
(s + 3) Cs + 4)
..

s + 5
1 - 3s Figure 5.45 System diagram-for Problem 5.21 .
-
(a )
(5.3, 5.21 Consider the closed-loop system in Figure 5.45.
5.4,
R (s) +
- l:
+
- K!IO - s)2
s(s - 1)2 (3 - ,j
1
1
Y<s>,
11
1
5.5) a. Sketch the root locus in the s-plane as K varies from O to I t
Do not solve for breakaway points, if any occur.
b. Specify those values of positive K for which closed-loop poi,
are on the jw-axis.
( + 3)2 c. What is the frequency of sustained oscillation ( w*) whcn A ,
,2 + 3 + 1
adjustcd to that value found in Part (b)?
-
(b) d. Determine the angle at which the root locus dcparts from 1111
pote at - 1 + ). . .
e. Sketch the locus of closcd-loop' poles in the s-plane as K vam
R(sl + K(2 - sl Y(s)
'<'
s3(s + 1)
from O to - oo (modificd rules). Do not solve for breakawa~
,1,: points, if any occur.
1

(e)

Figure 5.43 Closed-loop system diagrams for

,. (5.3, 5.22 Sketch the locus of the roots of the characteristic equation a; A
5.4,
5.5)
vares from O to + oo for the six GH(s ) functions that follow 111
each case, determine the range of positive K for stability. f'ind :ill
breakaway points, angles of departure or arrival, and ali o th r 1
pertinent details. Assume negative feedback, as in Figure 5.5.
Problem 5.19.
K(s + 1)
(5. 3, 5.19 Express GH(s) for the systems of Figure 5.43 in "standard" form a. GH(s) = s2(s + 2)(s + 4)

-~
5.4, so that the root loci may be plotted. In each case, determine
5.5) whether the basic rules or the modified rules apply when the given
K(s + 1)
b. GH(s ) = s i(s - 2) ..
gain K varies from O to + oo . Do not plot the root loc. 1-J
rr
(5.3, 5.20 For the system of Figure 5.44, K(s + 10) 2 ..
~o, = ...,~
~.::;.....
(~A~,it,::';'\)
...
<:. . -~,
,.
5.4, c. GH(s) = s2(s + l}(s +.100} -~a,,~/1.~f'J.;.,!1
a. Plot the root loci as K vares from O to + oo . Specify jw-axis
5.5) :;~
~~~nr;
crossings and ranges of K Cor stability, but do not solve
explicitly for breakaway points.
- K
d. GH(s) = s(s 2 _ 2s + 2)
t~ ~;~~i.1':K'i
., F))
b. Repcat these instructions for the case where K varies from O to "b~1~
-OO . K(s + 1) 2 t-
V)
c. Gll(s) = (s - 1)2(s + 4) ~f
R(s) + Ci)
l: ,

K(s + 2) (s + 3) 1
(s + 1) (s - ))r I
Y(s)
) Ks (s + 1)2
r. Gll(s) =
(s + 2)(s - l)(s - 2)
(5.3, 5.23 Sketch the root locus plot for each system in Figure 5.46. Sp1, 11
5.4, ranges of K > O for stability, but do not find exact loca1to11 ,,1

Figure 5.44 System diagram for Problem 5.20.

5.5) breakaways, if any. t.
 302 Stability Analysis in the s-plane Problems 303

Y(s)
~
K
' - 4
Y(s)
..
'

$ - 1
' - 2
(a)

Figure 5.49 System diagram for Problem

5.26.

~ K(a+ll
.i1, + 10) (, + 100)
I 1
Y(s)Jo
(5.3, 5.26 Consider the closed-loop control system shown in Figure 5.49.
5.4, Sketch the s-plane locations of the closed-toop poles (i.e., root locus)
5.5) as K vares from - oo to + oo. Note that both basic and modified
rules are needed. Determine ali j<.>-axis crossings and breakaway
points.
(b)
(5.3, 5.27 For the system shown in Figure 5.50, provide a root locus sketch of
5.4, the closed-loop pote locations in the s-ptane as
Figure 5.46 System diagrams for Problem 5.23.
5.5) a. K is vared from O to + oo.
b. K is vared from O to - oo.
(~l:j'l .3, S.24 Determine the range of positive K for stability for the system of In both parts, determine ali jw-ax.is crossings, but do not specify
', 4, Figure 5.47. Sketch the root tocus ptot for O < K < oo. ex.act breakaway points.
,'i ,5)

+ K Y(s)
R(sl + (i) 1 KO - > 1 Y(s) ~ l:~
s +2
l: 11> (, + 2) < + 3) I 11>

, 1 -(11 +- , ')2 ~,~~~

Figure 5.47 System diagram for Problem 5.24. Figure 5.50 System diagram for Problem
5.27.

(5.3, S.25 For the system in Figure 5.48, give a rough sketch (shape only; no (5.3, 5.28 Plot the root locus of the closed-loop potes in the s-plane as K
5.4, detaits required) of the tocations of the poles of Y(s)/R(s) as the 5.4, vares from - oo to + oo for the GH(s) that follows. Provide ali
5.5) parameter A is varied from O to + oo. 5.5) details, such as jw-ax.is crossings, angles of departure, and break
aways.

R(s) + {i)
Jo l: )o
1
(2 - l
A
(1 + ,2,
I 1
Y(s)
Jo Ks(s + l)
+
Gll(s) = (s _ l){s _ 2){s + 3){s +4)(s + 5)
(~)2 1-
(5.3, 5.29 For the system of Figure 5.51, use the root tocus mcthoJ to plot thc
5.4, closed-loop pote locations in the s-plane as K varics frorn oo 11,
Figure 5.48 System diagram for Problem 5.25. 5.5) + oo. Supply ali pertinent informauon .
~ ~ - ~ - ~ - - - - - - - -- - . ,- .,, ,,., r tJ11nn m ff ., plnrui

1/(.I Cl: ) 1
K( + 10)?
.'ii. + 1)
Y(l

6
-
.

Figure 5.51 System diagram for Problem 5.29. Stability Analysis

(5 .3. 5.30 Consider the closed-loop control system shown in Figure 5.52.
5.4.
5.5)
Sketch the s-plane locations of the closcd -loop poles (i.e., root locus)
as K varies from - oo to + oo. Note that both basic and modified
in the Frequency Domain
rules are necded . Determine ali j w-axis crossings and breakaway
points.
A sinusoida l steady-state output is obtaincd whcn a n asyrnptotically stahlc lt11,.1i
K( + 112 Y (~ system is subj cctcd to a sinusoidal input. Thc linear systcm gcnerally modif, th<
~- amplitudc and phase of the cntcring sinusoid ; thc frcqucncy w is uncha ngcd W,
invcstiga te tite stability o f closed -loop linear systems in this chapter by 11 , 11,1
frcquency resp onse mcthods. Jnitially, we define frequency response performan, ,
spcc,fica tio ns and then apply thesc to underdamped second-o rder systems. W,
idcnt1y four kinds o f frequcncy response plo ts and p resent contour map ping a,.,
Figure 5.52 System diagram for Problem 5.30. prcliminary step 111 developing the Nyq uist eriterion. Severa) cxarnplcs o f th"
criterion are prescntcd , togethcr with detailed comparisons with the root h :u,
(5.6) 5.3 1 Sketch thc root contour in thc s-plane for the system o f Figure 5.53 mcthod. In fact, we combine our knowlcdgc of thcse two powerful procedurc~ to
as T varics from O to 1- oo, with K = + 3. determine 1w-axis crossings (root locus) and phase crossovcr frcq11encie!I (Nyq111, 1
cri terion) in thc simplest way. Wc complete the chapter by considering rclat1H
K( + 1) Y() stability and the closed-loop frequency response.
~ O +T,1<> +?;i

Figure 5.53 System diagram for Problem 5.31 .

(5.6) ~.32 C'onsider the system in Figure 5.54. Plot the location of the closed-
6.1 FREQUENCY RESPONSE PERFORMANCE
loop p oles as b va ries from - oo to ~ oo. Provide ali pertinent SPECIFICATIONS
in formation on your plot.
Perfo rmance spccifications relative to frequency response curves include band
width, peak magnitude, gain margin, and phase margin [ 1- 4). The closed-loop
R(.l + Y<l system bandwidth and the closed-loop system peak magnitude are usually of
intercst in feedback analysis and design. Thesc specications are indicatcd on thc
closed -loop magnitudc and phase curves o f Figure 6.1a. G ain margi n a nd phasc
margin specifications pertain to the open-loop plant frequency response curves.
+ 21> as shown in Figure 6.1b. We require that this open-loop p lant be placed in a
I + b
negative unity-feedback configuration. Each o f these four frequency response
specificatio ns is dcfined more prccisely below.
Figure 5.54 System d iagram for Problem
5.32.
305
 ,r,

306 Stability Analysis in the Frequency Domain Frequency Response Performance Specitications 307

2
1.5

1 MP
t 10
t

:i
~
0.5 Q.

wp w~

... -
t -90

~ t
~
-180

-270'
(a) Closed loop
-270'~-------------

Figure &:1 Frequency response performance i k:, (b) Open loop

specifications.
Figure 6.1 ( Continued)

6.1.1 Bandwidth
6.1.2 Peak Magnitude
1he dosed-loop system bandwidth (BW) is defined as that value of frequency at
w hu.:h the magnitude of the closed-loop system frequency response is reduced to The closed-loop system peak magnitude (Mp) is a measure of the relative stahility
1/ ./2 of its low frequency value. Recall that we calculated this closed-loop of the system. Values of 1.2 to 1.5 are usually not considered excessive for most
1

:-.y,11:m bandwidth in the gain-versus-bandwidth discussion of Section 3.2. Whether control system applications, and these "rule-of-thumb" values are often used for
a lurgc o r small bandwidth is required depends on the particular application. In design purposes.
:-.rn 11c cases, a large bandwidth is needed to enable the system to respond properly
fo1 a widc frequency range of input signals and control commands. In other cases,
a 1,rna ll bandwidth is needed to provide attenuation of high-frequency noise on
t h c inpu t signal. 6.1.3 Gain Margln

1 The gain margin is a performance specification that is based on the open-loop

A ll crn.Hc c.lcfinit iou~ of bandwidth are sometimcs used in other applications. For example, the
response but that provides analysis and design information about the closed-loop
handw,cJth o f tuncd circuir, i; oftcn delincd as the frequency rangc bct wcen thosc points cm the
n, agn itudc rnrvc that are located at 1/ fi of thc peak value. system response. It is assumed that the open-loop plant is placed in a negative
 308 Stability Analysis in the Frequency Domain Underdamped Second-Order Systems 30CJ

unity-fccdback configuration. The gain margin (GM) is defined as 6.2 UNDERDAMPED SECOND-ORDER SYSTEMS

Maximum allowable open-loop gain for closed-loop system stability This section applies the performance specifications defined in Section 6.1 to th1
GM = - - - -- -- - -- -- - - -- -- - -- - -- underdamped system (O < f < 1) of Figure 6.2. Typical open-loop a nd closed-loop
Actual open-loop gain frequcncy response curves for this system are shown in Figure 6.3. Closcd-loop
{6.1) system bandwidth and peak magnitude are calculated by using the closed-loop
frequency response curves, whilc gain and phas margin computations are pr ,
formed by rcferring to the open-loop frequency respon~e curves.
For cxample, if the open-loop gain K is 10 and the closed-loop system would
becomc marginaJJy stable for K = K* = 60, then the gain margin is 60/10 = 6. 2
"'n Y(. )~
A gain margin of 4 or greater is an arbitrary "rule-of-thumb" in many design ,l:. /
-~

8(5 + 2 ( ,en)
applications, although lower values are sometimes used. '
Observe that the definition of gain margin in (6.1) is not restricted to a
frequency response interpretation. In fact, we could apply the Routh-Hurwitz
criterion (Rule 3 of the root locus construction rules) to determine K*. Howevcr,
we show later in this chapter that, for a stable open-loop plant, the closed-loop Figure 6.2 A~ underdamped second-order system.
system is marginally stable if and only if the open-loop frequency response
magnitudc curve is unity at the same frequency where the phase is 180. We 3
define a phase crossover frcqucncy (wrc ) as that w at which the phase is 180.
Figure 6.11> shows open-loop plant response curves with a magnitude of a < 1 at
w = <rc- Since the actual open-loop gain could be increased by a factor of 1/a
to reach unity, we say that the gain margin is 1/a for this case. Careful reflection
o n thcse procedures should revea! that ths result is consistent with (6.1). t 2

~
s5
6.1.4 Phase Margin

. In accordance with the above discussion, we identify the gain crossover frequency
( Wc;c ) as that w at which the open-loop frequency response magnitude is unity.
f igure 6.lb shows a typical case. We define the phase margin (PM) as

PM = 180 -1/ G{jwGc) 1 (6.2)

~----+-----------
l"\",C
1
1
--
O" o 1
.,_.
whcrc /G(Jwc;c) is expressed as a negative angle, usually in the third quadrant.
As ano thcr measure of relativc stability, the phase margin indicatcs the amount of t
additio nal phase allowablc in a stable open-loop plant frequency response before
thc point of marginal stability for the closed-loop systcm is reached.
~ -90

Fo ur performance spccifications have been defincd in this section for cvaluating

system bchavior in the frequency domain. This overview of analysis and design
PM
ohjectives is intended to structure our thinking toward first guaranteeing closed- -1801------_L _____ _____ __ _
loo p system stahility and then ensuring that a desired degree of relative stability is
(al Open loop
maintained. These performance measures are applied to second-order under-
da mpcd systems in Scction 6.2, with a more extensive explanation of their Figure 6.3 Open-loop ~nd closed-loop frequency re-
significance in general dcsign cases provided in Scction 6.6. sponse curves for the system of Figure 6.2.
 ih 1r11!iliiy A11 1ly1,11 111 tho Frequency Domain Underdamped Second-Order Systems 311

1 {\ Solving for the closed-loop transfer function gives

M, Y(s) w; J (6.5)
R(s) = s 2 + 2{w,,s + w;
t I I_!
Setting s = jw, .the closed-loop magnitude and phase expressions are
~ J.IW '1
Y(Jw) 1=
p 1~ M(w) = R(Jw) 2
1
( "'2)
1- -i + ,_ (2fw)2
w,, w,,

'
1)
"'P ... - 4>( w) =
Y( w)
J
R(Jw)
= -tan - l
2[w
w,,
-
w
2 (6 .6)

11
1- 2
w,,

t 6.2.1.1 Bandwidth
'
ii
We determine the closed-loop system bandwidth (BW) as that frequency al
which M( w) in (6.6) is equal to 1/ fi. Performing the indicated operation yields
1110
-- - - - - - - - - - - --
(bJ Closed loop
BW = w,,/i - 2( 2
+ /4( 4 - 4( 2 + 2 (6 .7)

Figure 6.3 ( Continued)

A curve of the normalized closed-loop bandwidth (BW/ w,,) versus f is shown in
Figure 6.4.

e:.~.'I Opm-Loop and Closed-Loop Frequency Responses

,i: 11 i11; 1 ... w lor thc open-loop transfer function G(s) gives

1.2
2
wn w /2(
( 1' ( }W
. ) _
- -_ _ n
(jw )(jw + 2[w,,} (jw )(l + Jw/2{w,,)
(6.3) e
~
:,:: 0.8
l:Q

n,,: cn 11 r, pomltng open-loop magnitude and phase relationships are

0 .4

w,,/ 2{
jG(Jw ) I= w/ l + (w/ 2[w,,) 2 '-----'------1----'------'------'-
o 0.2 0.4 0 .6 0 .8 10

/ G(jw) = - 90 - tan - 1 (i;w,,) (6.4) Figure 6.4 Normalized closed-loop bandwidth (BW
versus f for the system of Figure 6.2.
/ w0 )
312 Stability Analysis in the Frequency Domain Underdamped Second-Order Systems 313

6
6.2.1.2 Peak Magnitude

Fo r 1he closed-Joop curve in Figure 6.3b, we wish to find MP and the frequency
5,
(w,,) al which the peak occurs. Using M(w) from (6.6), we set dM(w)/dw equal
to zero for w = wP and simplify to obtain
WP = wn/~1 -- 2!- 2 (6.8) 4

whcrc O < ! < 1/ fi. Substituting (6.8) into M( w) in (6.6) yields

MP 3
M
"
=----
2!~
1
(6.9)

Plots o f M" and (wp/wn) versus! for the closed-loopsystem are givcn in Figure 2

6.5.

6.2. 1.J Gain Margin

Both thc gain margin (GM) and the phase margin (PM) for the second-ordcr 01 1 1 1 1 1' 1
... 1
O O 1 0.2 0.3 0.4 0.5 0 .6 0 .7
systcm in Figure 6.2 are measured with respect to the open-loop frcqucncy
response in Figure 6.3a. Since the open-loop phase / G (jw) in (6.4) vares o nly
ovcr thc range fro m -90 to - 180, thc gain margin is infinitc. As w -+ oo.
'
(o)

/e; (w) approachcs - 180, and thus the phase crossovcr point occurs at
w = co with a magnitude of 1,ero. Consequently. GM is infinitc for this case.

6.2.1.4 Phase Margin

0.8
Thc phase margin (PM) is measured at the gain crossover frequency wGc defincd
as that frequency at which the magnitude of the open-loop frequency response
function is unity. Sctting IG(JwGc)I = 1 from (6.4), we obtain
0.6
4 2 ic
WGC = Wn,//4! + 1 - 2f (6.JQ) -;....
From (6.4), the corresponding open-loop phase at the gain crossover frequency is 0.4
given hy

/ G(jwGc) = - 90 - tan -
1
[ ;r ,/4r 4
+ 1- 2!
2
] (6 .11) 0.2

Thc phase margi n may then be expresscd from (6.2) as

'O 1 1 1 1 1 1
o.7
PM = 180 -1/ G(Jwqc) 1 0. 1 0.2 0 .3 0.4 0.5 0.6

1 '
(b)

= 90 - tan - 1
[ 2! //4! 4
+ l - 2! 2
] (6.12) Figure 6.5 Curves of M and ( "'P / "'n)
r
versus for the closed-foop frequency
Curves o phase margin and ( Wcc/wn) versus! are shown in Figure 6.6. response in Figure 6.3b.
 314 Stability Analysis in the Frequency Domain Frequency Response Plots 315

90-I
5.0, and this peak value occurs at a frequency ( wP) of 9.9 radians/second. The
phase margin (PM) in (6.12) is 11.4, occurring at the gain crossover frequency
(wcd of 9.9 radians/second, and the gain margin is infinite. J
- .'

PM

6.3 FREQUENCY RESPONSE PLOTS

Four types of open-loop frequency response plots are described and compared in
this section. The first of these is the linear plot used in the previous two seclions.
We next introduce Bode diagrams and log-log plots, which are both based on the
0.2 0.4 0 .6 0.8 1.0 properties of logarithms. These curves are useful for quick frequency response
( sketches or for more detailed analysis and design. The fourth type of frequency
(a) response representation is the polar plot, wh.ich is used in applying the Nyquist
criterion described later in the chapter.
Frequency response curves of magnitude and phase versus frequency may be
plotted directly on linear scales. Although an advantage is the directness of this
procedure, a major disadvantage usually identified for linear plots is the cumbcr-
0.8 some computations. However, the use of calculators or personal computers has
tended to reduce this objection to sorne extent for individual point calculations.
Nevertheless, the shapes of curves for linear plots are not as easily determined as
0.6
l those for Bode diagrams or log-log plots.
} Bode diagrams are plotted on sem.i-log paper with linear vertical scales and
0 .4 logarithmic horizontal scales. The logarithm to the base 10 of the open-loop
frequency response function magnitude is computed as a first step. Magntude
values in decibels, defined as 20 log 10 IG(Jw )1, are then entered on the vertical
0.2 scale. The logarithm.ic property allows us to form the resultant Bode magntude
curve by summing the contributions due to individual factors. The Bode phase
curve has a linear vertical scale in degrees and a logarithm.ic horizontal curve for
o~_ _...,,___....,.,..__-:-",:---...,,...,,---...,...,,- frequency. The phase curve is usually sketched beneath the magnitude curve with
0 0.2 0.4 0.6 0.8 1.0
frequency values aligned. Magnitude and phase curves for four types of factors
'
(b) (constant gains, linear factors, poles or zeros at the origin, and quadratic factors)
are sketched on sem.i-log paper in Figure 6.7.
Figure 6.6 Curves of phase margin (PM) and We begin the construction of Bode rnagnitude and phase curves by expressing
normalizad gain crossover frequency ( "'ac / wn)
versus f for the system of Figure 6.2. the open-loop frequency response function in the "standard" form of (3.16) with
s = jw, as shown in Figure 6.7a. For a constant gain factor of the form K in
Figure 6.7b, the Bode magnitude is 20 log10 !Kl, and the phase is 0 for K > O
and 180 for K < O. The Bode magnitude curve in Figure 6.7c, for a linear
EXAMPLE 6.1 factor of the form (1 jwT), has a low-frequency asymptote of zero decibels and
a straight-line high-frequency asymptote that intersects the horizontal axis at che
Determine the frequency response specifications identified above if w,. = 10 comer frequency w = 1/T. Alternatively, we may express the linear factor as
radians/second and f = 0.1 for the system of Figure 6.2. (1 Jw/wn), where wn is the comer frequency. The actual magnitude curve
By direct evaluation from (6.7), the closed-loop bandwidth (B W) is 15.5 differs from the asymptotes by approximately three decibels at the comer
radians/second. From (6.8) and (6.9), the closed-loop peak magnitude (MP) is frequency, and this ditference is smaller for ali other values of w. When a linear
316 St.,b,hty Analysis in the Frequency Domain Frequency Response Ploh :11 ,

~ 20
20 tog1oll + jw'71

t , '
-
KO + J.,T) K(I + j.,t,.,.)
G!jw) = G<sl,- ., = - - - -- ~ 2 "' 2 2
-J (jm)'"(l :t 2((j,,,)/o, - OJ'i) (jo,)'"(! :t 2((jru)/OJ - ., /.,)

(ll) Standard form

.. .1
~
t,,l

:1:
:::::::: f'..._ 1 / ...
201-
/20 log1olKJ (K > 1)
-
-20

t
f'1l -- - - - -
.
201og10 IKJ (K < ll
-
-20 t- -

180" 1 -
t
:ll
"'
.e
., _
"
o..
L.!!_ (K < 0)

901 - -90"!- - - or/1 -1 jwT -

(el Linear factors

t
~ /f..!S. (K > O)
Figure 6.7 ( Continued)
"'
.r:
o..
0.1
,.,_
Bo<lc magnitude curves in Figure 6.7d for potes or zcros at the origin. that 1,.
(jw )"', are straight lines with slopes of - 20m db/ decade, whcre 111 is the numbc1
-90" 1- - o potes (for positive m) or the number of zeros (for ncgative m) at thc o rigin
The corresponding phase is a constant o ( - 90)m for ali w. Finally. quadrati<.
factors of the form (1 2f(jw)/wn - w2/w;) are considercd in Figure 6.7r.
Whilc different curves are obtained for different valucs o the damping coefficient
-180"1 ,,.,--::L!i (/( < 0) - r. observe that low-frequency asymptotes of zero decibcls and high-frcqucncy
asymptotcs o 40 db/ decade are obtained for all Bode magnitude curves
(b) Conslant ga,ns
Moreovcr, Bode phase curves are given by tan - [(2fw/w"}/(1 - w2/w;)J and
Figure 6.7 Bode magnitude and phase curves for four types of vary bctween O and 180 or between O and - JSO, depending on whethcr tht
factors. term appears in the numerator or denominator and whether a plus or minus sign
r.
appears before the term involving The distinction can be discerned in a mannc1
fac tor appcars in the numerator of the open-loop frequency response function, we analogous to the linear factor cases. For convenience, only a denominator tcrm
form a straight-line high-frequency asymptote that has a slope o + 20 decibels with a plus sign is considercd in Figure 6.7e. Observe that a quadratic factor can
pcr deqadc. This asymptote has a slope of - 20 db/ decade when a linear factor
appears in the dcnominator. The phase is + tan- ( wT) both for (1 + jwT) in the
be expressed as the product of two linear actors for > l. r
Log-log plots, as the name implies, require logarithmic scales o n both vertical
numerator a'nd for (1 - jwT) in the denominator; it is - tan - ( wT) for (1 - jwT) and horizontal axes. The advantage in plotting magnitude curves is idcntical to
in thc numerator and for (1 + jwT) in the denominator. that dcscribed for Bode diagrams- that is, contributions of individual factors are
 fi H,l111i1 v .l\t1 tW11h lo lho rroquency Domain Frequency Response Plots 319

20 1og10 j(jw)2j 20
(m = -2)

1 '= 0.4 \. /'\/<' =0.1 . 1 J

20 logJOliwf
t '
(m = -1)

1
w ....- .....
JI
.o

::i:
o
1
,t ~ " " ' -

wn
w-
20 logJOI (Jwl
-20
1
(m = +I)
1
20 log 1ol ~>
(Jw 2
(m = +2)

1111 ,: -
/uw9
/ IJW/"
_I

t
IC ' .
L!:::_
- o l-.::::: ---.....
Wn
1
w- 1

t
~ " w- 5:
"'
,:
o.
-90
,,

,,e !~ - - - 1801------- - - -1-

1 - - - - - -
1
1
~~

'
(e) Quadratic factors

1/l(J
/j;~
- Figure 6.7 (Continued)

(el) Poles or zeros al the origin

phase plots. Both of these plots are preferred over direct linear plots that can be
Figure 6.7 (Contnued) determined point-by-point by using electronic calculators but that are more
difficult to sketch without a large number of these computations. Easily sketched
111i111 wd 111 1'0 1111 thc resultant curve. A disadvantage is that the phase curve from Bode dagrams, polar plots are needed to apply the Nyquist criterion
11 .11111 , ., l111ar vertical scale. Consequently, sketching the phase curve on the described in Section 6.5.
lio, loH p,1pc1 hcncath the magnitude curve presents a minor problem usually
, , .11lw cl hy 111od1fying the vertical log scale to yield a linear scale.
EXAMPLE 6.2
1' 111.11 plo h for open-loop plants are necessary for understanding and applying
11 11 Nyq11 1\I c.: ,itcrion for stability analysis in the frequency domain. Figure 6.8
Sketch linear plots, Bode diagrams, log-log plots, and polar plots for each of the
, l111w~ ,, t yp,c ll polar plot. The gain crossover frequency ( "'cd occurs at a phase
open-loop transfer functions given below for s = jw.
111 , 1pp1 w,1111utcly 110, and the phase crossover frequency ( wcc) occurs at a
111.ig111l11d1 of " Therefore, we may determine the phase margin as 70 and the lOOO(s + 1)
/'.11 1,1.1, g111 a~ 1/ a . The polar plot is plotted more easily by first sketching the a G ( s) = -(s-+-2)-(s_+
_ l_O)-( s- +
- 20-)
1 ,11, n po11d111g Uodc diagrams.

l 11 \ 11111111111 y, Bode plots are recommended over log-log plots for magnitude lOO(s - 1)
b. G(s)= - --
1111 v1 , hn-.111., c of the prcsence of a linear vertical scale, which is also needed for s(s + 2) 2
3?.0 Stability Analysis in the Frequency Domain Frequency Response Plots 321

G(jwl - plane .'

j0.5 4

-1.01 0.5 1.0 1.5 2.0

., = o
t 3

\ !
s;,
\ 2
\ -j0.5

',', '
'
wc;c -ji.O

o
o 10 20 30 40 50 60
(d~

90-
Figure 6.8 A typical polar plot showing gain and phase margins.

To ohtain the frequency response function for (a), we substitute s = jw to form

IOOO(jw + l)
t
~
o

-90"
"'-1
20 30 40 50 .,_ 60

(6 .13)
G (Jw) = (jw + 2)( jw + lO)(Jw + 20)
-180- , - - - - - - - - - - - - - - - - - - - - -
(a)

The magnitude for the linear plot is given by 11 ) Figure 6.9 Frequency response plots for Example 6.2(a).

IG(Jw)I= IOOOW+T (6.14) To obtain the Bode diagrams a nd log-log plots, we first express (6.13) in thc
./w2 + 22 ./w2 + 102 yw
/ 2+--
202- "standard" form

and the phase by

G(J~) = ( 2(::0) )(1 +J~)
1 + -10 1 + ~~ )
1+ jw ) ( .
/G(Jw) = tan- 1 (w)- tan- 1 (i} (
)W ) (
2

2.5(1 + jw)
- tan -
1
~) tan -
1
(6.15) (6.16)
( 1 + ; )( 1 + ~~ )( l + ~~ )
{ - { : )
1
I

Thesc curves Me plotted in Figure 6.9a. We discuss procedures for detcrmining

thc peak magnitude, gain margin, and phase margin 'in Section 6.6. The Bode and Iog-log magnitude plots differ only by a scale factor and hencc 111.1'"
lll!l ,t1l>11ity Analysis in the Frequency Domain Frequency Response Plots 323

- - -20 . - - . - - - , - - - , - - - - - , - - - , - - - , - - - . , . . - - - ,
40
16
Asymptotes
!
t 12 L..1.---4"
t
t - ~ -8
!
30

!~-~~:~,
~ 20
~
o 4
ff
1
51 1
10
,~,
10
0.6

90
1 1
0 2 3 4
w-
o L - - - - 1 ' - - - - - - J - - - L . - - - - ' - - -.......- -
5

9~~---------------------
tt::1: 1

t. ~r
j
~ -90
0

n"' 1
2
~ 15
710 / 20
1
40
.,_:__
1

~
1, 2 3 4 5
w-
-90

- 180
-
Resultan!
phase - - -180 1----------------------
(b) (a)

t
!
GUw) - plane Figure 6.10 Frequency response plots for Example
6.2(b).
i:., w =
be plotted as the same curve with two different scales, as shown in Figure 6.9b.

The corresponding phase diagram is identical for Bode and log-log plots.

-
- 1 / 10 The polar plot for (a) in Figure 6.9c starts (w = O) ata magnitude of 2.5 and
5
w = 40 phase of O. We use the Bode plots of Figure 6.9b to help us sketch this polar
-1
Re G(jw) plot. As w increases above zero, the phase goes positive briefly and the magnitude
increases, resulting in a plot that is in the first quadrant of the G(jw )-plane. For
w = 20 larger values of w, the phase retums to 0 at a magnitude somewhat larger than
2.5 and then continues to decrease toward -180 as w -+ oo. The Bode magni-
tude plot decreases steadily, thus indicating a zero magnitude as w --+ oo. This
frequency response is typical for an open-loop plant that passes signals having
-3 w = 10 frequencies below 20 radians/second and attenuales those with higher frequcn-
cies.
The required frequency response plots for (b) are given in Figure 6.10. We say
(e)
that a stable plant transfer is "mnimum phase" if its zeros are located in the
left-half s-plane. The transfer function for (a) is mnimum phasc. On the other
Figure 6.9 ( Continued) Frequency response plots for Example 6.2(a). hand, we classify a stable plant transfer function as "nonminimum phase" if one
 Contour Mappings 3l5
324 Stability Analysis in the Frequency Domain
1
40 : I"' = o+
50 G(jo,) - plane. 1
32 1
20 1
1

t 24 1 40
1
1
1

t
3
:::,
--!,--
J
"
16
-

30
1

o
2 20

3
8
2 10

--- - 01 1 1 1 \ ~ 1
-10 10 20 40 50
... "'5 ReG<J>-
.. = 4

90
___ l __ J____ l __
Resulting + 90 dueto 1ao for
., = 2 "'= 1.5

negative gain (on standard form) -20

{

t o
' ........
0 .5
and - 90" due to 1/jw
(e)

~
10 Figure 6.10 (Continued)
,v_
1

to the number o f zeros (Z) minus the number of potes (P) contained within thc
-90'
s-planc closed contour. This principie, based on the nel change of angle o F( r).
is central to the development of the major frequency-domain stahlity critcrion.
which is prescnted in the following section.
-180' L - - - J L - - - - ' - - - - ' - - - - - ' - - - _ . . __ _, Consider the functon F(s) given by
(b)
K(s + z )(s + z ) ... (s + zm)
1
Figure 6.10 (Continued) Frequency response plots for Example 6.2(b). F(s) =--- -- 2- -- - (6.17)
(s + P1)(.r + P2) ... (s + Pn)
or more o its 1.eros are located in the right-half s-plane. The case in (b) is an The locations of the zeros and potes of F(s) are shown in Figure 6.11. Let U\
cxample of a nonminimum phase transfer function. arbitrarily select a closed contour in the s-plane that encircles only sorne of thc
zeros and poles of F(s) and does not pass through any. We begin al Point P, an<I
evaluate F(s)l,-r, by using s-plane vectors. This value of F(P 1 ) is then plottcd in
the F(s}-plane. We continue by calculating F(P,) for other points P, along a
closed contour in the s-plane and by plotting the results in the F(s)-planc.
Let us examine the propcrties of the net angle change for the F( s)-planc
6.4 CONTOUR MAPPINGS
I
contour. We see from Figure 6.12a that if we have encircled only a single 1.cro
Mnppings o( a function F(s) from the s-plane to thc F(s}-plane are considcred in (and no poles) in proceeding around a closed contour in the s-plane. then thc mt
this scction as the point s movcs around a closed contour in the s-plane. We show change in angle of F(s) is 2.,,. radians. 11s net angle change of 2.,,. radian~ fnr
that thc net numbcr of encirclements (N) of the origin in the F(s)-plane is equal F(s) corresponds to a single encirclemcnt of thc origin in thc F(s)-planc
 .116 Stability Analysis in the Frequency Domain Contour Mappings 327
,-plane
F(s)-plane through any zero (or pole) of F(s). If there are Z zeros inside the closed contour
l in the s-plane, then the net contribution to the angle change in 'F(s) is 211'2
radians. Similarly, if there are P potes inside the closed s-plane contour1 the net
contribution to the angle change in F(s) is - '2.?TP radians. Nte that the negative
sigo is obtained because the net angle change dueto poles-that is, /(s + p;)-is
1r/,11 d , ')( n ==> subtracted in (6.18). Thus, (6.18) becomes

,.;
~ 1Ductocio-1 = 2'1TZ - 2'1TP {6.19)
s-plane contour

Since the net change in the angle of F(s) may itself be written as 2'1T tnes the net
l luure 6.11 Mapping from the 9-plane into the F(s)-plane for the function in number of encirclements (N) of the origin in the F(s)-plane, we have 2?TN =
(h ' /) .
2?TZ - 2'1TP. Dividing by 2'1T yields

1\ l11nover, the direction of tls encirclement is the same as the direction of

N=Z-P (6 .20)
11111wment around the closed contour in the s-plane. For example, if the s-plane
, 1111 lo u r is counterclockwise about a single zero, then the corresponding contour
i11 tllc F(s )-plane is also counterclockwise once about the origin. Observe from wlch is the main result of this section. ln words, (6.20) states that the net number
1 11(111 e 6. 12b that zeros of F( s) lying outside the given con tour in the s~plane do of encirclements of the origin by the conformal mapping of this closed s-plane
111,1 1 o ntribute to the net angle change for F(s) as the con tour is traversed.
contour into the F(s)-plane is equal to the number of zeros minus the number of
\'fr may express the angle of F(s) in (6.17) as
poles of F(s) inside the closed s-plane contour.

~=
m "
[/(s+z;)- [/(s+p;) (6.18)
1-1 1- 1 EXAMPLE 6.3
\ , wt o bscrved in Figure 6.12a, the net change in /(s + z;) as a closed contour Sketch the F(s )-plane contour obtained by mapping the function
111 1lt1 .1- pla ne is traversed 'is 211' radians if z; is inside the closed contour and O
1.1111111, if z, is outside. We refrain from constructing a contour that passes
2(s2 + 4)
{6.21)
F(s) = 5 2( 5 2 - 4)
s-plane s-plane

)e 1 / r,t X
1 for the closed s-plane contour in Figure 6.13a. We use the s-plane vector concept
to calculate F(P;) for the points P1 through P9 The results are plotted in the
F(s )-plane in Figure 6.13b. Because of the symmetry involved both in F( s ) and
,. '(I _( \ 1 --<>-,.. in the given s-plane contour, the mapping obtained from points P5 through P9 is
-z
identical to that obtained from points P1 through P5 Since F( P5 ) = F( P1 ) and
X

(a)
-- ~ X 1

(b )
{ P9 = P 1, the F(s)-plane contour for points P5 . through P9 is superimposed over
the same contour for points P1 through P5 The F(s )-plane contour encircles the
origin twice in the opposite direction from the direction of traversing the s-plane
contour. Whereas the s-plane contour is counterclockwise, the F(s )-plane contour
encircles the origin twice in the clockwise direction. We note that there are no
zeros ( Z = O) and two poles ( P = 2) inside the s-plane con to ur, which yields a
1 h,111 .. 6 12 Determining net angle changes of F(s) dueto inside and outside zero value of N = - 2 from (6.20). This result is consistent with Figure 6.13b.
"''" 111110111:.,
328 Stability Analysis in the Frequency Domain The Nyquist Criterion 329

}2
s-plane

,-plane
..., F{s)-plane

1/\
2
P3 ji o

r
P4 P2
JO

-2 -1 o
1 1 2
,
Ps JO P1 = P9 (a)

~
(P=y GH(s)p1ane
-ji
p6 Ps
P7
-1

-j2 (-1 +jOJ

(al

F(s)-plane
F(P 2) = F(P6 )
ji (e)

~
Figure 6.14 Definition of planes for contour mapping.
O 1~ F(/'3) = F(P7)

-'1 -2 -1 JO the system is unstable. Contour modifications in the s-plane are discussed for
open-loop poles on the jw-axis, and several examples are worked to illustratc
(N = -2) ~ -jl
~ these details in a variety of situations.
To apply the results of Section 6.4 for the development of the Nyquist criterion
F(P) = FCP5) = FCPgl FCP4) = F(Pgl
as summarized above, we select the entire right-half s-plane as the closed contour
(/)) of interest. Moreover, we investigate the conforma) mapping from this s-planc
contour to the F(s)-plane defined by choosing
Figure 6.13 Contour mapping for Example 6.3.
F(s) = 1 + GH(s) {6.22)
6.5 THE NYQUIST CRITERION
The right-half s-plane contour is indicated in Figure 6.14a, the F(s)-plane in
We use thc result in (6.20) in this section to develop the primary frequency 6.14b, and the corresponding GH(s)-plane in 6.14c. The mapping for a typical
response stability criterion for linear, time-invarianf feedback systems (5-8]. In stable second-order plant transfer function.is shown as an example. Observe from
particular, we select thc closed contour in the s-plane as the entire right-half plane (6.22) that F(s) = O implies GH(s) = -1 . Therefore, encircling the origin in thc
and !et the function F(s) be the characteristic polynomial for a closed-Joop F(s)-plane is equivalent to encircling the -1 (or - 1 + jO) point in the G/f(s)
system. We show that encircling the - 1 + jO point in the GH(s)-plane is plane. Hereafter, we will consider only the mapping from the s-plane in Figure
equivalcnt to encircling thc origin in the F(s)-plane, since F(s) = 1 + GH(s). 6.14a to the GH(s)-plane in Figure 6.14c. Note that there are no encrclemcnt~
Thcrcforc. a polar plot in the GH(s )-plane with s = jw is constructcd first, and (N = O) of the -1 point in the GH(s)-plane for the example. As shown in Fig,m
thc nel number of encirclements ( N) of the -1 + jO point is determined. We 6.14a, the s-plane contour in general consists o a portian from the origin along
show that P is the number of open-loop poles in the right-half s-plane. Thus, Z the jw-axis toward +joo, an infinite semicircle from + 90 to - 90 containing
may bl determined from (6.20) as the sum of N and P, where Z is the number of the first and fourth quadrants, anda portian along the -jw-axis from - joo to tlw
zcros (or roots) of the characteristic equation, or equivalently the number of origin. The contour mapping into the GH(s)-plane corresponding to these par1~
closed-Joop polcs, in the right-half s-plane. If Z is zero, there are no closed-loop of the s-p]ane contour is composed of the polar plot GH(jw) for the first portion
polcs in the right-half s-plne, and the system is stable. If Z is greater than zero, and a mirror reflection of this polar plot about the real axis in the Gll(s )-plarw
 111 mnr,11i1v Ar1uly:11:. 111 the Frequency Domain The Nyquist Criterion 331

1111 11,r ;i:H 1i,j ,1 l1eh\1;c11 - j,~ and the origin along the -jw-axis. The contribution
i iht1 11111,ii,J ~c111i, H k " sunply a redirection of the GH(s)-plane contour, TABLE 6.2 Steps In Applylng the Nyqulst Crlterlon
!Id,, 111 lhl 11umerator of GH(s) is not greater than the order of the
1. Determine the number of open-loop poles in the right-half s-plane (/,)
U/11 S') l,ns lhl form by inspecting the factored open-loop transfer function GH(s).
2. Sketch Bode magnitude and phase diagrams of the open-loop
K(s + a1 )(s + a2) . (s + a,) transfer function GH(jw ).
(,//( 1} 3. Sketch the polar plot in the GH(jw )-plane and complete the
(s + /J 1)(s + /J2 ) (s + /JP) (6.23)
GH(s)-plane contour closure resulting from the selection of the
entire right-half s-plane as the contour to be mapped into the
ll!i! ~. /( ,, i lltfiV lllJ CllfH l~.\\Cd as GH(s}-plane.
4. Determine the net number of encirclements (N) of the -1 + jO point,
/ 1 4 1 ... 1 1 ','// ( 1)
where N is positiva if the encirclements are in the same direction as
the s-plane contour and negativa if the direction ,s opposite.
A(1 1 a 1 ) (s + a,) 5. Compute Z - N + P, where Z is the number of right-half s-plane
1 1
( I' 1 /3.) ... ( s + /Jp) closed-loop poles. The closed-loop system is stable if and only if Z is
zero. lf Z is negativa, an error in finding P or N (probably N) has
been made!
( ., 1 /l 1 }(s 1 /J ) . . . (s + /JP) + K(s + a 1 )
-2 ---'--..:....;..__ _ _ _
(s + a.)
_ _ _._ (6.24)
(s + /1 1)(s + /12) ... (s + /JP)
diagrams for s = jw as w goes from O to + ao. As noted above, the corresponding
polar p lot for negative w, that is, for s = jw as w varies from - ao 10 O, is
lli ,1 111,J I'"", of F(~) are the poles of the open-loop transfcr function
obtained by constructing the mirror reflection about the real axis in the GJ/(s)-
i ho 1n11i1I, 1 111 thc,c poles inside the right-half s-plane closed contour is
plane.
11 ll}!IH l i\~ I ' M111rnvc1 , the zeros of F(s) in (6.24), that is, the roots of the
The net number of encirclements (N) of the - 1 + jO point may vary as K
lt1I .H.!'ii ij1ic fq11 ll11111, 111c the closed-loop poles. We denote the number of these
becomes larger or smaller. We determine the ranges of K by first calcula ting thc
imilhJ tltiJ 11r h1 1t.11 I p laneas Z, where Z is to be determincd by applying
phase crossover frequency ( w pe) from
hij\il~, Uitriii111 Wc further note that the open-loop zeros, that is, the zeros
t ll f), [11t'c,:i1 1ft, , hupe of the GH(s)-plane p lot but do not otherwise enter
lmGH(jw}l ... - ...,c = O (6.25)
ltili- 1hr, t i!lr, ic,11 l'11t.111011, These observations are summarized in Table 6.1.
I!! l1;itf, wu 1i11 1li-trn11111c P directly from the open-loop GII(s), then find N At w = w Pe we compute the value of gain K* for which the Gl/(jw )-plane plot
liHH !IHJ (1//(J,,1 ),1l.1 11c plot, and finally determine Z from Z = N + P. The passes through the - 1 + jO point by setting
1 irn1,p 1,yh 1e111 1 ,1.,hlc 1f and only if Z = O. The steps of this algorithm are
11 lt d In 1lll!l ,; 2, l h~ vuluc of P May be found quite easily since GH(s) is Re GH(jwpc}IK-K =- l (6.26)
1
1 1J111i1td IO !o 111 1111 hlll'd form and the open-loop poles can be Jocated in the
,1,ilil' llitc, 1ly Nc,:,1 Wl' cxpress Gl/(jw) in standard forro and sketch the Bode Ranges of K versus N are then tabulated , and Z is determined from Z ~ N + P,
ifiil 1,11,1 1,1; d iag, .1111s. The polar plot is thcn sketched from these Bode where P was determined in the first step of the algori thm.

EXAMPLE 6.4
Relation to Criterion Result
Apply the Nyquist criterion steps identified in Table 6.2 to determine as a
) tn ri ght hall s-plane Open-loop poles, that is, poles of function of positive K the stability of the closed-loop system formed by feeding
GH( s) in right-half s-plane back negatively the output of a combined plantjcontrollcr transfcr function givcn
) ,n r19ht hall s-plane Closed-loop poles in right-half by
s-plane
K
Affect shape of GH( s}-plane plot (6.27)
GH(s} = (s + l)(s + 2)(s + 3)
 The Nyquist Criterion 333
332 Stab1hty Analysis in the Frequency Domain
j0.2 results in a redirection of the GH(s)-plane contour with an infinitesimal magn1
GH(j..,)plane GHC,)-plane lude. The traversing of the negative imaginary axis in the s-plane, that is, s - ,,,
for - oo < w < O, completes the closure of .the s-plane contour. This section o
jO.I
.,..--
;0.1 - ....~
the s-plane con tour yields a GH( s )-plane con tour that is the mirror refl ect1on
about the real axis of the corresponding section for s = jw with O < w < + oo.
1 I.N = 0) 1 '\ The result of traversing the entire s-planc closed contour is shown in thr
6 - 60
'\ / 6 GH(s)-plane plot o Figure 6.15b for K = 1. _
-1 - 0.1 O1 02 -1 -O 1 O1 We have arbitrarily chosen to procced around the right-half s-plane in a
clockwise direction beginning at the origin. This choice coincides with traversing
~ w the positive imaginary axis from w = O to w = + oo, the convention used 111
constructing first the Bode diagrams and then the polar plot o G H(jw ). Ho\.\.
ever, once the polar plot and its mirror reftection ab out thc real axis in thc
GH(s)-plane has been constructed, the direction of traversa) in the .t-planc is
-j0.2 -j0.2
arbitrary. 2 If a clockwise direction in the s-plane is sclected and the corrcspond-
Ca) K = 1 Cb)K = 1
ing mapping in the GH(s )-planc encircles the - 1 point in a clockwise direction 11

(N = +2)
112

,.,
.,.. --- ------- ...... ............
'
', G//(s)plane
times, then N is positive. Jf the corresponding encirclcmcnts of the - 1 p oint are
in a counterclockwise direction, then N is negative. Reversing thc dircction
selected in the s-plane (counterclockwise) simply reverses the direction o encir
' clements o the - 1 point in the GH(s)-plane, yieIJing a positive value of N for
'\\
\
counterclockwise encirclements and a negativc value for clockwisc encirclements
Whichcver direction is selected, N is positive if the encirclements in the Gil( f)
\ plane are in the same direction as the s-plane contour and ncgativc if thl'
\ 100 direction is opposite. This simple statement, appearing in Step 4 of Table 6.2.
jO \/ T circumvents continua! references to clockwisc and counterclockwise argumcnt!>.
o 4 8 12 16 20 In applying the Nyquist criterion to the closed-loop systcm defined hy thc
GH(s) in (6.27), we first determine the numbcr of open-loop poles in thr
right-half s-plane ( P) as zero. This is Step 1 of the procedurc itemized in Tahlc
-j4 6.2. From Bode diagrarns (Step 2), we havc already sketched the polar plot in thc
GH(jw )-plane and have completed the closure (Stcp 3) in Figure 6.1Sh for
K = l. We sce that there are no encirclements of the - 1 point ( N = 0) for thi!>
case. Therefore, Z = N + P = O for K = l , and thc closed-loop system is stahlc
for this value of K .
Figure 6.l Sc shows the same GH(s)-plane sketch as in Figure 6.15h, except it is
-)12 or a value of K = 100 and, consequently, is 100 times larger. The scale has at,o
(clK = 100 been rcduced in Figure 6.15c for conveniencc. Tracing along thc GH(s)-planc
contour corrcspo nding to thc clockwise direction arbitrarily sclected for thr
Figure 6.15 Polar plots for the application of the Nyquist crilerion to Example 6.4. s-plane contour around the right-hal plane yields two net encirclements of thl
- l point: N = 2 for K = 100 (Step 4). Extending a ray from the - 1 poinl
Figure 6.1 So shows a polar plot for GH (jw) for K = 1. This plot corresponds outward in any direction toward infinity, as shown in Figure 6.1 Se, allows us to
to thc con formal mapping of Gll(s) Cor that scction of the s-planc contour along determine the value o N easily. Thercfore, we can now use Step S in Table 6 2 to
thc positivc imaginary axis-that is. for s = jw with O < w < + oo. Thcreforc, we
arhitrarily begin traversing the s-planc contour at the origin of the s-planc and
con~idcr a li points along thc positive imaginary axis. Next, we continue by
8 2 Mathcmaticians gcncrally in~ist on tr:wcrsing lhc splanc conlour in a coun1crclock" 1~c d1m 11nn
traversfng the s-plane contour clockwise along the infinite semicircle Re' , where
8 !>(cause a rcgion is said 10 be "encloscd" hy a closcd conlour if 1ha1 rcgion lics on thc k ft a, 1h,
R _. oo and O vares from +90 to -90. Since R-+ oo, then G/l(Re' ) oontour is lravcrsed. Many cont rol 1cx1books conform 10 lhis convcnlion and rcqui rc countcrdo\\ " i
approaches zcro, regardlcss of the value o f O. Consequcntly, the conforma! cncirclcmcnls in the Gl/(s)plane for positivc N and clockwisc encirclcmenls for nega1i,c .V
mapping of the infinitc semicircular are from thc s-planc into the GH(s )-plane

.-
 f:\j IYGl:1 111 lho r roquency Domain The Nyquist Criterion 335

.,.. N 1 /' "' J I O 2 for K = 100. Observe that the value of P

11 @!In~ rnr -i!l h,,111gl111ut the apphCtion of the criterion; only N (and hence TABLE 6.3 AppllcaUon of Nyqulst Crtterton to Example 6.4 ,1
1 rnnv , l,cwc M K v111H,
Wri !!fiWJ mlW 1h1)w11 th,11 /, O for K "" 1 and Z = 2 for K = 100. Reviewing Range of K p N z St~bility
1l)j
!111 1Jltjl' fo1 c.:111;'1 1 r,l thl .1,: two values of K reveals that Z = O for ali K Cor which O<K<60 o o o Stable
!lu, p11in, pl_1 of f~ //( 1w) uosses the negative real axis in the Gll(jw )-plane 60<K <+oo o 2 2 Unstable
1l 1t,111n ( 1,/ 1111il ,, l '1,) at values to the right of the - 1 point. This crossing of
llw i1tR1iiih1 , i;tl uj, onurs at the phase crossover frequency wpc, which may be 11

h !nmliil f1,_,111 (ti l'I) l'o determine the imaginary part o GH(Jw ), we first
P,HHJ lHI 1'flli1111,1l11i the denominator of that function, that is,
EXAMPLE 6.5
K K
iill( jJ)
(j,., f l )(Jw + 2)(jw + 3) - (jw) 3 + 6(jw) 2 + ll{jw) + 6 Apply the Nyquist criterion to determine the stability of a closed-loop system

(<i 6w 2
)
K
+ Jw( ll - w 2
)
[(6 - 6w )
(6 - 6w 2 )
2
-
-
jw(ll - w2 )
jw(ll - w2 )
l having GH(s) given by

GH(s) = . K(s - 1) (6.31)

A j(t, 6w 2) - jw(ll - w2)] where the output of GH(s) is fed back negatively to form the closed-loop systcm.
' 2 2 (6.28) We begin by observing that there are two (open-loop) poles of GH(s) in the
(e, - 6w 2 ) + w 2
(11 - w2 )
right-half s-plane, that is, P = 2. Next we set s = jw and express GH(jw) in the
! lliHl\ 1111, 1111,1g111My part of GH(jw) in (6.28) equal to zero yiclds standard form Cor sketching Bode diagrams as
K [ - w{ll - w2 )] ( - K/8)(1 - jw)
lmGJJ(jw) = =O {6 .29) ( 6.32)
{6 - 6w 2 ) 2 + w 2 (11 - w2 ) 2 GH(jw) = (l _ jw/2)(1 - jw/4)
! 1ti111 \O. l11d1 w cquals either zero or /iT
radians per second. The value of w = O The Bode magnitude and phase diagrams for GH(jw) are sketched in Figure 6. 16
ill t: 1111111 , ,1 pos111ve real axis crossing; the vaJue o w we seek is the pha~e Cor K = 1 with the corresponding polar plot shown in Figure 6.17. The conforma!
ltJ ,11\1 1 11 1.:tfucncy "'"e =,v'iT.
Observe that adjusting the value of K does not mapping o the closed contour around the entire right-half s-plane is obtained by
111,, 1111, licqucncy a t which the angJe of GH(jw) is - 180. In fact, we see from perforrning a rnirror reflection of the polar plot for O < w < oo about the real axis
1111~ ll11dl' da.1grams that adjusting K has no elfect at ali on /GH(jw ), only on in the GH(s)-plane; ths resulting plot is shown in Figure 6.17a also. We se.: that
1111,: 111,11'.,utudc o GH(jw). there are no encirclements of the - 1 point Cor K = 1, that is, N = O, and
l 11 dt:tlt nune tha t vaJue of K, denoted as K*, for which the GH(jw )-plane plot consequently Z = N + P = O + 2 = 2. In other words, the unstable open-loop
p.1 ~.n thwugh the - 1 point, we use (6.26) and (6.28) to obtain plant remains unstable when connected in a closed-loop system with the gain K
set equal to l. However, we observe that one or more encirclements do occur
K*(6 - 6w 2 ) when K is adjusted to sufficiently larger values.
Rc(iJ/( w
/ PC
)1 K-K
= - - - -- - - -
(6 - 6w2)2 + w2(11 - w2)2 To determine the number of closed-loop poles in the right-half s-plane (Z) for
lw-/u larger values of K , we first express Gl/(jw) in rectangular form as
6(/IT)2)
K*(6 - . K (jw - 1) K ( - 1 + jw)
GH ( 1w)= = -- - -- -
= (6 - 6(/"IT)2)2 + (/IT) 2[11 - (/IT) 2] 2 (jw - 2}(jw - 4) (Jw ) 2 - 6(jw) + 8
- K*
= - -- = - 1
60 (6 .30) K(-1 +jw) ((8 -w 2
) + j6w)
= (8 - w2 } -j6w (8 - 2
w ) + j6w
1111.: 1do, e, wc find that K* = 60. These results for Example 6.4 are summarized
111 l .1hlc 6 1 Observe that results from the Routh-Hurwitz criterion for the same K [( -8 - 5w 2 ) + jw(2 - w 2 )]
,y., tl'lll ,n Fxample 5.4 are in agreement with this solution. (6.33)
(8 - w 2 ) + {6w) 2
2
336 Stability Analysis in the Frequency Domain The Nyquist Criterion J ;J /

(N = O) (;1/(s) plane

0 .5 2 4 -- jO. I
/.,..-~,,
1
1c.:,
o
~ . hAli- . .
~ JO

ff
oN
-4011 1 1 1 1 1 1
- - - - jO. I

1'. (a)K =1
9 0-1-r- -1---1 - - -r- - -c.::-: :E: l .1 :
(N = -2) \ .,..- --- ....... ...... 1 C.H(<I rane

' '
'
w-
;O 5
()"
2 4 10 ' "\
\
1 I \

~ -90
_____ ___ ~
,1 I
\
~7
1 - 1 \
\
_7~,,;:
~
\,
6 8 o
-ISO"LL __L __ J ~ -- L-- ..::_,___ __ 1
- 1. 5
' .....~
-1 / - o5 .iO

-270"

-
Figure 6.16 Bode d iagrams for Example 6.5.

U~ing (6.25), we can solve for "'re from

( b) K = 7
2
K[w(2 - w ))
=O (6.34) Figure 6.17 GH(s)-plane plots for Example 6.5.
lmGH(jw)l..,- ...1'(" = (8 _ w2 )2 + (6w) 2 , w- wr1
for N in both cases because encirclements about the - 1 point occur in a
direction opposite to that of thc direction ihat thc contour was travcrsed in thc
to o btain wrc = O and "'re = .fi.. Substituting these values of "'re nto (6.26) for s-plane. Thcrcfore, Z = N + P = - 2 + 2 = O for 6 < K < 8, and Z = - 1 + 2
the GH(jw) in (6.33) yiclds = 1 for 8 < K < + oo. These results are summarized in Table 6.4.
K(-8 - 5wic)
' = -1 (6.35)
Re GJl(Jwrc)IK- K = ( 8 _ wic) 2 + (6wpc) 2
6.5.1 Open-Loop Potes on the / (,.)-Axis
fro m 1.Vhich K = 8 for wre = O and K = 6 for "'re= .fi.. Therefore, we see
tha t the v~Jue of N changes for K = 6 and for K = 8. Figures 6.17b and 6.17c Thc s-planc contour must be modified when poles of GH(s) occur o n the jw-axis.
sho w GH(s)-plane plots for K = 1 and K = 10, respectively, from which we can Consider the case of a single open-loop pole at the orgin, as shown in Figun
de te rmine N as - 2 for K = 1 and N = - 1 for K = 1 O. The minus sign is correct 6.18a. Lct the s-plane contour pass around this pole in an infinitesimal scmicirt k
 Hl 111 In lho I roquency Domain The Nyquist Criterion 339

-- --------..., '
ji

GH(s)-plane
- .......
s-plane '-
,
Gll(s)-plane

( \ '' \
.t
', \ \
\ 1
\
\ j0.5 I I
\ / /
\ / /
1.1 I
\
\ .
_..,, ,,,
I
wt L ..- /
./
/

\I
~

')\. o (a) (b)

/ 10 - 0 .5
/ u Figure 6.18 Contour mapping for an open-loop pole at the origin .

where e is arbitrarily small (e> O) and O varies from 0 to 90. Let the GH(s)
function under consideration have the form
\'
K{l + s/a 1 ) . . . (1 + s/a, )
(6 .37)
GH(s) = s(I + s/P1) ... (1 + s/Pp)

,,
Id Thus, for s = ui8 , GH(s) becomes
-)1
w=6
(e) K = 10 8
Kl+ - ee1 ) {
... 1+ eei )
fl1111J li 1/ ( Continued) GH( s)-plane plots for Example 6.5. ( 1 a,
GH(s) I, -,.; = _( eei) ( ui8 )
ee 11 l + - .. 1 + p
----
Appllu ,Uon of Nyqulst Crlterlon to Example 6.5
P2 P

p N z Stability
K K
2 o 2 Unstable = - -8 = -e-J' (6 .38)
- eei E
2 -2 o Stable
2 - 1 1 Unstable
19
Since e is arbitrarily small, the magnitude of GH(s) for s = ee in (6.38)
becomes infinite. Moreover, since O vares between O and 90, the angle of
1li1,1 ,,1,~, l11tl1 \ th e polc from the interior of the closed contour. The remainder of GH(s) in (6.38), that is, - O, vares between O and -90. Figure 6.18b shows the
11,a ,_,i1tl 11111 1, 11mlia11gc<l from the previous discussion. We examine only that GH(s )-plane mapping for the part of the s-plane contour in (6.38) when K is
1!.,, 1 , ,I Ili,J o, 11 111 11 l'k in thc first quadrant and obtain the result for the fourth positive. We show in Example 6.6 below that the real part of Gl(jw) is sorne
,p, 11d1 111,1 1,v 11 111111m 1cllcctton about the real axis in the GH(s)-plane. In the first finite nonzero value as w -+ O. Consequently, the small finite offset in Figure
tjll(1lf111tl , WC fi IVC' 6.18b from the negative imaginary axis is indeed correct. Note that the angle of
GH(jw) as w -+ O is - 90 even with this small offset, since the magnitudc is
s = eei {6.36) infinite.
 1oq, 11>111~v -Of111,r111 1 'l h o Nyq ui:,I C, ilo t i1111 ~M 1
"
pl~nr - .......
',
(;11(~) pl,, ne 40

t
Ooub le
pole
1/
( \' \
'\ \

1 \
- "'

,
3
:::,
oo
,;;
E
o
N
20

o
10
I \ I
/
I \ / '
/

_ .,,,/
/
/
' ' , ..... _ _ ,, /
/ - 20

(b)
(11)
1 1 1 1 1
Figure 6.19 Contour mapping for a double open-loop pole at the origin in the 1 0. 1 0,4 1 4 10
o
s-plane.
~
t 1 1 / .- -1 r--___ 1 "'-
For double open-loop polcs at the origin in the s-plane, wc havc ;;;:,
-90

ui9) ... (l 9)
r.ej
Kl + -
( +
-18~~~-- - ~~~~ ,-
GH(s) -
(uie>2 1+ - -
! ',_,,)... (i +,.T.. (al

/J1 p
Figure 6.20 Bode diagram (a), the polar plot (b), and contour
mapping ( e) for Example 6.6.
:::: K K (6.39)
----;-
(
1:
)2 = - e - 126
E2 for K = 1. To determine the real part o f GH(jw ) as w - O, we write

As shown in Figure 6.19, the G/f(s )-planc contour moves atan infinitc magni-
K ( l - jw ) ]
tude from thc positive real axis through - 180 (clockwise through 180), since Rc(GH(jw)} = Re [ jw( t +jw ) l -1w
/ GH ( s) = - 28 and 8 vares, as bcfore, from 0 to + 90 (countcrclockwisc)
around thc polc al the origin in the s-plane. In general, if there are m potes at the
- j K(1 - jw )]- ~
origin in thc s-plane, then thc samc rules apply, except that thc GH(s)-plane (6 .41)
= Re [ w( l + w2) - l + w2
co n tour varies through m quadrants, that is, m times - 90, a t an infinite
clockwise d ircction from its starting angle.
Thercfore, the real part of Gll(jw) as w - O is - K, as indicated in Figure 6.20h
Cor K = l.
EXAMPLE 6.6 We continue by mapping thc infini tesimal s-plane eontour s = ,e 19 for 0 .$ O
,s; 90<> into thc GH(s )-plane, as shown in Figure 6..20c. Then wc use the po lar
Considcr a ncgative feed back system with open-loop transfer function given by plot o f Figure 6.20b to map the s-planc contour along the positive imaginary axi,
from w = + E to w = + oo. The infi nite circula r are in the first q uadrant o f lhc
K s-pla nc yiclds only a redrection o f the GH( s )-plane mappi ng, since the magni
(6 .40)
' GH(s) = s(s + 1) tude o f GH(s) for that are is zcro. We reflect the mappingjust dcscribed for all
parts o f the s-plane contour in the first quadrant to yield thc rcmaindcr of the
1 he Bode diagrams and polar plo t for E < w < + oo are sketched in Figure 6.20a s-plane mapping for the fourth quad ran t. Applying the Nyquist criterion y1d1h
!I! !ill JIIIIY /Ht llvoiti 111 tho rrequency Domain The Nyquist Criterion 343

G/J(jw)plane
------ ........
, ...._ G/ICs)-plane
J
t
w
J3r '
~ ,,

'
2
\
2L \
- ji J ' \

,,, o J I

() 4 1
1
-2 \\. \
... () lit1
- j3
-1
2
\
1 I
1
(b)
1
//
- .......
s-plane w = 0.5.
t '~ \
\
."itt -;, ~//
/
/

w -j3
w = 0 .3 ,
~ \ ______ _.,.,,, / /
\ 1
,1
01--)
IP=OI'I 1
(e) (Continued)

t ~/ Figure 6.20 ( Continued)

le)
- //

Closed-loop systems are formed by feeding back the outputs of these GH(s)
functions negatively.
Figure 6.20 ( Continued) Polar plots in the GH(s)-plane are shown for these two cases in Figure 6.21.
Both GH(s) functions contain a double pole at the origin in the s-plane and,
N 1 / O I O O. since no net encirclements occur ( N = O) for any consequently, an infinitesimal sencircular are (s = u 18 ) is selected to bypass
Hit ilt V\J 11 ,uu l lhc open loop plant is stable (P = O). these poles, excluding them from the interior of the closed contour in the s-plane.
Therefore, P = O for both cases. Considering only that part of the infinitesimal
are in the first quadrant, where O ~ 8 ~ 90, we obtain a GH( s )-plane rnapping
/\Ml'l ,1: 1, / asan infinite are that begins at 0 and vares through -180-that is, a 180
clockwise rotation. Figure 6.21a shows the resulting plot in the GH(s)-plane for
A~ 1 ~'. ,.i 1111 l ,:,a 111ph: n this subsection, consider two open-loop transfer functions (a), which has the (s + 1) factor appearing in the denonnator of GH(s), thus
gi \lc11 1,v contributing a phase componen! that vares as an-arc tangent curve from 0 to
-90. The resultant phase curve for GH(jw) vares from -180 to - 270 as
K shown. For all values of w (and especially for w --+ O), the polar plot les in the
a. Gfl(s) = 5 2(s + 1) second quadrant of the GH(jw )-plane. On the other hand, the polar plot for the
GH(s) function in (b) lies entirely in the third quadrant, Since the (s + 1) factor
appears in the numerator for (b), we have a corresponding phase componen! that
h. GII( s ) = K(s + 1) vares asan are tangent curve frorn 0 to +90. Figure 6.21b shows the resulting
s polar plot for the combined GH(jw) function.
344 Stability Analysis in the Frequency Domain The Nyquist Criterion 345

GH(s)-plane Gll(~)-plane
.
.,
-1 + JO ,.

1\
\
-1 + JO
I
I
1
\\ -- ... //
J
\
,:/ '' ' //

'~ ...........
' --
__ ,,,,.,,,,/
-- --
...... _,,/

(Ne +2) --- , ......

......
(N = O)
-- ...................
'\.

GH(.~)-plane
~\
\
\
\
GH(8)-plane

"' \
\

I
\
1
,, -l+JO/ 1 ~

,---~- -
l.. / \ /
' '~......__ __......
\
\ ~/

',~
',...... __ __.,,.. //
~/ //

(/,)

(al Figure 6.21 ( Continued)

Figure 6.21 GH(s)-plane plots for Example a finite crossing of the negative real axis occurs at sorne phase crossover frequency
6.7. "'re, we obtain different values of N (and hence Z) by varying K over its positive
Figure 6.21 also shows complete G//(s)-plane plots consisting of the polar ranges.
plo ts just described and their mirror reflections about the positive real axes. We We now extend these results to negative K ranges by first rotating the
see that N = + 2 for (a) and N = O for (b). Therefore, for all positive K the G//(s)-plane plots 180 to account for the additional negative sign on K and
values of Z determined from N + P, where P = O in both cases, are 2 and O, then applying the Nyquist criterion as before. Ranges of negative K yiclding
rcspectivcly. different values of N (and Z) are possible when the rotated GH(s)-plane plots
have finite negative real axis crossings. Alternatively, we could also perform the
same analysis on the original GH(s)-plane plots before thc 180 rotation by
6.5.2 Applications lnvolving Negative K identifying N as the net number of encirclements of the + 1 point in thc
Gfl(s )-plane. Ths altemate viewpoint corresponds to positive K and positive
Thc Ny,quist criterion has bcen applied to cases thus far having variations in K
eedback that, as indicated in the early part of Section 5.5 on negative K root
frorn O to + oo. Bode diagrams and the resulting polar plots have becn con-
locus, is equivalcnt to negativc K and negative feedback.
structcd for 'K = 1, and the complete GH(s)-plane mappings have been formed
Figures 6.22a through 6.22e show the GH(s)-plane plots Cor ncgativc K
hy cornbining polar plots and infinite contours for the first quadrant of the
(normalized to K = -1) for the GH(s) transfer functions considered in Exam
s-plane with their mirror reflections about the real axis in the GH(s)-plane. When
 1a1 !!1!\V /\iililY!ilO 111 lhu r roquency Domain The Nyquist Criterion 347

1
j0.2 1
1

\'"
Gll(s)-plane 1
1 J
/
GH(s)-plane
/ 1
1
1
60 /
1' 1
1
1
- :r -0.l
- O("'"
O1
r
I 1
I 1

'\. __
,....,,.__
....... -jO. l I
I
'
1
1

-1 '\ +11
\ \ 1
- j0.2 \ \ 1
(a) Example 6.4
\
\
\
1'
'
\ 11
1,
~--,/~
U) GIi(, J-pla" \\ 11 :
\N jO. l ..k-

\" t1 1

"
' '
',
.......
"
_J,1,
1
o
---- 1

-o u-
- -- - ' '
(e) Example 6 .6

(b) Example 6.5

(N = + 1) ......
IN= +ll - ......

-
1 l111uu 6.22 GH( s)-plane plots for Example 6.4
1111111111h (l / wlth negativa K. GH(s)-pla~'\ ~'
,..----n \
GH(s)-plane
' \

1 !1111111/'h 11 / . For Figure 6.22b, we see that N = O for ali K < O. However,
( , t. ,1 1,,uuplc 6.5, we obtain Z = N + P = 2. For the GH(s) function
l 'i111111l'I, 1, h (Fi.i,mc 6.22c), we have N = + l and Z = N + P = l + O for
W, ,d.11 11htain N '"' + 1 and Z = 1 for K < O for both GH(s) functions
! IJ@iipl,) 11 / 111 Figures 6.22d and 6.22e. Example 6.4 in Figure 6.22a yields a
\
\
,, - ~

~/
"
'I
I '\ \~
....... ...._ _ _ --y

-
/
I

' , ....... _
/
,_,,,~ii 1111, ol thc ncgative real axis after the GH(s)-plane plot has been
111111 1t'd I HII'' 111 11n.:oun t for the negative sign on K. This finite crossing occurs at
11 1lirn lort, for Ex.ample 6.4 we obtain N = O (Z = O) for - 6 < K < O
1 1(/ 1) for K < - 6.
'' '
--
(d) Example 6.7(a)
- //

Figure 6.22 ( Continued)

..,,,.-
_. /
(e) Example 6. 7(b)
/

f1)(i\Ml'I I fl u
Bode diagrams and the polar plot of GH(jw) for K = l are shown in Figure
1_L " 1tl1 1 t h, 1wv,at ve feedback system having the open-loop transfer function 6.23a, with the complete GH(s)-plane mapping of the right-half s-plane shown in
iiv, 11 l, y Figure 6.23b. Toe two phase crossover frequencies are obtained from (6.25) as 2
and 3 radians per second. Adjusting the gain K in (6.42) to values of 64 and
GH(s) = K(s + l)2 121.5 for these two frequencies, respectively, places the GH(s)-plane plot on the
( 6.42) -1 point. Therefore, we identify N = + 2 both for O < K < 64 and for 121.5 <
s 3 (s + 6)2
348 Stability Analysis in the Frequency Domain Relative Stability 349

.'
-
/
o 10 20 / 1 1 Gh(s)-plane

...
1
1 "'\
1 I

~
~
::,
-20
,I *'PC = 2. \ 1/'PC =. 3

'
1

~
o -1
121 5
tii
.2 - 40
\ I
o
\\ /

-
N Asymptote
h/
' ' ' ......
-,
-- --
/
-60 1-1 1 1 1 1 1 ~ 1 /

(a) (Contintted)

90
JI///
///
/ 1/
/,,,--
11
- -
..... Gll(slplane

',~
/
/
//
/
t
-- -.C-,
I ''
,':t
GH<s)-plane

''"
',~
'\\ \
\
/ I T /
I I
1,
,_
\
\ I '
\ \
\

,,, '
'
11 - 1 ' \ - 1 ' 1
1 \ \
\ \,., /
I \ I I
\.....,,
~ ,'/'/1
\ / / /
///
(Two coinc,dent curves) ,,,'" , .... ......,_'
\\ ' 1
_/.,, ' ', ...... _____ ..,,,,.,,
//
//

',-1
(/,)
(el

Figure 6.23 ( Continued)

TABLE 6.5 AppllcaUon of Nyqulst Crlterlon to Example 6.8

(11)
Rangeof K p N z Stability
Figure 6.23 Bode diagrams and GH( s)-plane plots for
-
O<K<64 o +2 2 Unstable
Example 6.8. o
64 < K< 121 .5 o o Stable
121.5 < K < +oo o +2 2 Unstable
- oo<K<O o +1 1 Unstable

K < + oo and N = O for 64 < K < 121.5. Moreover, we see from Figure 6.23c
that N F 1 for ali K < O. In ali of these cases, P = O and Z = N. Thesc findings 6.6 RELATIVE STABILITY
are summarized in Table 6.5. We say that the closed-loop system is "conditionally
stable" because the system is stable for 64 < K < 121.5, but adjusting the gain This section applies three of the frequency response specifications defincd in
bclow or above this range results in an unstable system. Section 6.1 to linear time-invariant systems of order n. These performance
specifications are peak magnitude, phase margin, and gain margin. We illustratcd
 Relativa Stability 351
lf:ill 111i}ili1y A1111ly1if~ In the Frequency Domain

G(iw)-plane
_, G(jw)-plane

-1

K < K2 < K3

l luu,~ G.24 Polar plots for the system in (6.43).

Figure 6.25 A polar plot illustrating an acceptable
tlil'li 1111p1_11t11111 111 Scc.:tion 6.2 for underdamped second-order systems. We
phase margin and unacceptable gain margin.
H! 111v1,11Kak these performance specifications here as measures of
11,f,illt,v l111t 111lly, we show that feedback systems having acceptable gain specifications to help ensure an acceptable system performance in the frequency
11!! 111111gins, determined from the open-loop frequency response,
1_1 li.1 .. t'
domain.
IIH'r Yll yrlil iJ. ' n~ive values of peak magnitude for the closed-loop system
Let us reappraise the meaning of these two performance specifications. To
f, 1j[ililV 11.;i , .......,. ('ontours of constant values of closed-loop frequency magni-
satisfy gain margin and phase margin requirements, we specify only that the
ltHl ! \f 1 [!I, drnVl'd and plotted as circles in the G(jw )-plane. Finally, we
open-loop plant frequency response characteristics obey preset conditions at
u1 trn1 11w ,111l1111~ huscd on Newton's method Cor calculating quite accurately
exactly two frequencies- that is, at the phase crossover (w 1,c) and gain crossover
,!i!i~f- ,,wit'. " .11111 ,,11111 murgjn.
(wcd While it is true because of the construction of the G(jw) polar plot that
the magnitude and phase at neighboring frequencies are also constrained to sorne
t Monouros of Relatlve Stability extent, we observe that it is possible to satisfy both gain margin and phasc margin
specifi.cations and yet have an unacceptable closed-loop frequency response. If the
'
{ 11ititlu 11,, , ; ( ,,.,) plane plots shown in Figure 6.24 for three values of positive
G(jw )-plane plot passes through the -1 + jO point, the closed-loop system has
s-plane poles on the jw-axis. Consequently, it follows that if the G(jw )-plane plot
A [111 11,_ ~'-' 11111l 111dc1 system with open-loop transfer function given by
approaches very near to the -1 point, the closed-loop system becomes less stable.
A third measure of relative stability is the peak magnitude of the closed-loop
K
frequency response (Mp) . We show in the following subsection that MP ap-
G(s) = (
l+ -
s )( .!..)
l+/)
{6.43)
proaches infi.nity as the G(jw )-plane plot approaches the - 1 point. For com-
/3 1 P2
pleteness, we illustrate in Figure 6.26 a G(jw )-plane plot for which the gain
margin and phase marginare both acceptab~e and yet MP is excessive. Additional
restrictions beyond these three performance specifications may need to be placed
w1i rr11 1h, 1,111p11t ol G(s) is fed back negatively to form the closed-loop system.
on the frequency response to achieve a desired closed-loop system bandwidth, for
Nwt ll,111 1111 ,,111 11 111argin is infinite, regardless of the value of K, because the
example, or to obtain a sufficiently sharp cutoff. .
ii l' ij! t11 v, 11 ti 11\i\ 111 Figure 6.24 is intersected by the G(jw)-plane polar plot only
H 1li,J ,_,1111 11 (11-. "' oo) . T herefore, on the basis of the gain margin specification
1h!H.l, 111, ,v~li 111 111 (6.43) appears to behave in an acceptable manner. Further-
~ tlrnt, .,, K increases, the phase margjn becomes smaller. For a
6.6.2 Contours of Constant M
11ll i~e11tl v l.1q1,1 A. , ;i prcsct phase margjn specification-for example, PM :; 30
Let the frequency response for an open-loop plant be expressed as
wuu ld 111 , ,lutnl <>11 thc other hand, Figure 6.25 shows a different G(jw )-plane
pl1 ''-'! \\ 1111 h 1111' phu~c margin specification is satisfied but the gain margin is
11!1 n111.d l 11111\, wc upparcntly need both phase margin and gain margin
G(jw) = X(w) + jY(w) (6.44)
352 Slabllity Analysls In the Frequency Oomain Relativo Stau1111y :tn

where .X( w) = Re G(jw) and Y( w) = lm G(jw ). For a negative unity-ec<lh.11 ~

G(iw)-plane system with the plant frequency response function G(jw ), we have

-1 + jO
M(w)/ cp(w) = G(jw) (6 '1 ~)

where M( w) is the magnitude and cp( w) is the phase or the closed -loop frcqucncy
response. Therefore, we may express tbe closed-loop magnitude M( w) as

IX+JYI .xi+ y2
(6 .46)
Figure 6.26 A polar plot for which MP is excessive and
M(w) =ll+X+ JYI /(1 + X)2 + y2
gain and phase margins are acceptable.
We wish to derive curves or constant M and plot these contours in the G(jw )-
plane. Sctting M( w) M (a constant), squaring both sides o f (6.46), and
1 rearranging yields
~
L .
M = I.O 1 14

M = 1. 1 G(iw)-plane
M2 )2 ( M )2
(X+ M 2 - 1 + y 2 = M 2 - 1 (6.47)

M = 12 M= 0 . 8 ~
Equation (6.47) indicates that constant-M contours are circles in the Y versus X
)Jane, that is, G(jw)-plane, with centers at X =- - M 1/( M 2 - 1), Y= O, and
radii r oC IM/(M 1 - 1)1- These curves are shown in Figure 6.27.

EXAMPLE 6.9

-6 -5 2 3
Use the constant-M circlcs in the G(jw )-plane to determine a linear magnitude
plot of the closcd -loop frequency response from the open-loop polar plot o

5
(6.48)
G (jw) = ( Jw ) ( jw )
(1+Jw)l+2 1+ 3

Figure 6.28a shows the desired polar plot and its intersections with severa!
constant-M circles. These values o M are plotted versus thc specific intersection
frequencies ( w) in Figure 6.28b. The derivation of the curve or constant closed-loop
I phase in the G(jw )-plane is postponed until Section 6.8, which also describes a
procedure for using a Nichols chart to transform directly from open-loop to
Figure 6.27 Constant-M contours in the G(/w )-plane. closed-loop Bode diagrams for unity feedback systems.
lll!llilV f\,,i.v1m1111 11111 1,oquoncy Domain Relative Stability 355

1 t-

'
j4
M = 1.0 G(jw)-plane
M 11
2.5
'
2.0

3
t 1.5

i
1.0

2 3
1
0.5

OL--L--L---1..--.1.---~-
o 2

(b)
3 4

w- 5

Figure 6.28 ( Continued)

G(jw), we may write

-j4

(a)
' tan- 1 ( ~
O(wk} = L wk) - LP tan- 1 ( -p
wk)
. {6 .50)
,f ,](111 ,10111 M ctrcles to obtain closed-loop frequency response ,-1 a, ,- 1 ,

where - a; are open-loop zeros and -/J; are open-loop poles. In this case, the
derivative d8/dwk becomes
W!_11111 1111 thod to calculate the phase crossover frequency (wpc)
1'r.i.111'~iv, 1111 mulas 1
1 -
- p /1;
d8 z a,. E~~ - ( 6.51}

O( w) k - [ - 180] ](_.!!._) dwk E

111+ ("'-k)2 - 11 + (wk)
.p, _
1,l4 a,
dO 1 180
dw ... -,.,.
Although more complicated terms are obtained for quadratic factors, such factors
occur far less often in practice than the linear factors to which (6.50) and (6.51)
io!A 1 1 '"' IJ A I t:i.wA (6.49}
apply.
Convergence is assured, in general, only for initial guesses sufficiently near thl!
i i- 11,0 ii 11 11 l1111 p phm,c as a function of w. For simple linear factors in final solution. Once "'Pe has been determined by the recursive application of
356 Stability Analysis in the Frequency Domain Relativo Stahlllly tll!1

where -a; are open-loop zeros and - /3; are open-loop poles as in (6.SO) 11 ..
TABLE 6.6 Numerfcal Resulta for Example 6.10 corresponding derivative for /( w,,) in (6.55) is
lteration 9( w k) dtl(w,,) ! d"'k z w p w
"'" /'( w,,) = .E 2 +" a,2 - .E 2 +* /3~;
1 3.00 - 172.9 -0.42 ,- 1 wk ;-1 w,,
2 3.30 - 179.6 -0.37
3 3.32 - 180.0 -0.37 Thus, we have relatively simple expression.s for / ( w") and /'( w") Cor the C:l\t'
1

when only linear factors are present.

(6.49), the magnitude of G(jw) is computed and the gain margin is given by
EXAMPLE 6.11
1
(6.52}
GM = IG(Jwpc) I Use Newton's method to calculate "'ce and PM for the negative unity-fecdback
system with the open-loop frequency response unction given by
as indicated earlier in Section 6.1. 50
G(jw} = (jw + 1)(jw + 2)(jw + 3)
EXAMPLE 6.1 0
Using (6.55) and (6.56), we form
Determine the phase crossover frequcncy and gain margin for the open-loop 1
2 {ln{w + 1) + In (w + 2
2
requcncy response in (6.48) of Example 6.9. /(w1c)"" ln50 - )

Table 6.6 shows the results of those calculations. The Roulh table may be used
tO Verify that Wpc = m==< 3.32. The fCSUJting gain margin is 2.0. + In { w + 32 )}

/'(w,,) = -wk[-wr- ~-1 + w: 22 + -w--:-3~2 ]

6.6.4 Calculatlon of Phase Margin
Applying (6.58) recursively yields wcc = 3.05 radians per seconi Using this
In calculating the gain crossover frequency ( wcc), we musl solve the equation value o w in (6.57) gives G(j3.05) = 1.0/- 174.01 ?. Thus, the phase margin
IGUwcc)I = l. lt is somewhat more convenient instead to solve the equivalent (PM) is
cquation
PM = 180 - 10( wad 1= 5.99
J(w) = In IG(Jw) I = o (6.53)

We may use Newton's formula rccursively as

/(w,,) 6.6.5 Calculation of Peak Magnltude

tlw,, = - /'( w1,.)
The closed-loop magnitude Cor a negative unity-feedback system with a frcquency
response function G(jw) is given by
wk+ l = wk + tlw,, (6.54)
G(jw) 1
For shnple linear factors, we have M(w) = 1+ G(jw)
1
1 z 1 P
/(w1c) = - L ln{w + a) - - L ln(w + /3/) ( 6.55) We want to determine the peak value of M( w) by solving for the w such thal
2, _, 2,=l dM(w)/ dw = O. To avoid the square root associated with the magnitude opcr
 llllllv l\ltfily!il!i 111 tho 1:roquoncy Domain Combinad Root Locus and Nyquist Approaches 359

!HHfili l i11 1., t lhl tlrnvativc o f M 2 (w) equal to zero. Let G(jw) gives
111 111! w j/ !}(j;,, ), 1111tl lct thcse numerator and denominator func- '
d d
l lm n.,ol!Ed 1,1 t,1111\ of thcir real and imaginary parts as (100 - w2 ) -d (100 - w2 ) +(O+ 4~) -d (O+ 4w) =O t (6 .65)
w w
N( ,., ) XN(w ) + JYN(w)
which simplifies to yield
/1( /1J) Xl)( w) + JY0 (w) (6.61)
(100 - w2 )(-2w) + 16~ = O (6 .66)
11 !o o i u,l H lt111g tlM 2(w)/dw = O yields

,/
/, ( \ N,
Xt I Y,J
1 X/))2 + (YN + Yv)2 = O
l (6.62)
Therefore, (6.66) gives the frequency ( wp) for peak magnitude as wP
We may then form MP from
= ../92. s; 9.59.

. M =1 100 1= 100 ~ 2.55 (6.67)

h ti 1lf 1iv,1 \.,J i11 (/, fl 2) i!> cqual to zero if 2
P 100 - wP + j4wP /(100 - 92)2 + 16(92)
. d
1i .,~ 1 ,\',, i , ( }'N ' Yv)2] dw [X~ + YJ) Since the given system is a second-order one, we may compute w,, and then use
(6.8) and (6.9) to yield wP and MP as
( ,\ J
N 1 >NJ) d- ( (XN+ X0 ) 2+ (YN + Y0 ) 2] (6 .63) = 1100 = 10
dw Wn

11111:11 N( w) 1s a constant, that is, not a fun ction of w, (6.63) f = 4/(2w,,) = 4/ (2(10)] = 0 .2

,/ ' d Wp = w)1 - 2f 2 = 10/0.92 s; 9.59

1\ 1 \',,) l ,\ N I X0 ] + (YN + Y0 ) - [YN + Y0 ] =O (6.64)
,l,J dw 1 1
(6.68)
!'' 11, 1, 1), 111 of (6.64) for the special case, yields the value of MP = 2!~ = 2(0.2)/o.% s; l.5S
i 111 1d,1o h tlic dosed -loop peak magnitude (Mp) occurs.
which agrees with the results in (6.66) and (6.67). However, it is impo rtant to
reiterate that (6.64) is applicable for unity-feedback systems of any order 11 , whik
(6.8) and (6.9) are valid only for second-order underdamped systems o f the forro
shown in Figure 6.2.
li!lil1 1lii) 111]1,~ v11 l11t of the closed -loop magnitude for the system of Figure
itv 11iji11~ t {,. ,11111 l'ompan.: with the solution obtained by (6.9) of Section

itk111il'y 11 d ,111d 1111aginary parts of the numerator and denominator as

i ' lO. 1 . (,,1) O, Xl>(w) = - w2, and Y0 (w) = 4w. Therefore, (6.64)

,-------. ...
6.7 COMBINED ROOT LOCUS ANO NYQUIST APPROACHES
1
...( i ) ~ l ~ Y (s)
We view the root locus method of Chapter 5 and the Nyquist criterion of tls
. " chapter in a combined context in this section. lt is evident that the jw-axis
crossing (Rule 3) of the root locus method ( w*) corresponds to the phase
crossover frequency (wpc) of the Nyquist criterion. We itemze four procedures
for finding w* (or wpc) based on the Routh table, the polar plo t, graphical
methods with Bode diagrams or s-plane vectors, and Newton's method. T hus, we
1li11110 6.29 The c losed-loop sys tem of Example may select the easiest of these possibilities in a particular stability analysis
j' t ~
 360 Stability Analysis in the Frequency Domain Combined Root Locus and Nyquist Approaches 361

J"' (Nyquist)
TABLE 6.7 Methods for Calculatlng the Phase Crossover Frequency
.,.plane t ___,... K (root locus)
Method Procedure

1. Routh table Use the s2 row of the table to

col]lpute w.
2. Set lm [ GHUw ))1... - ..l'C O Solve for "'Pe directly from the
polar plot.
3. Graphical Use the Bode diagrams or s-plane
vectors to determine the frequency
at which the phase is 180.
4. Newton's method Use the phase expression and its
derivative in Newton's recursive
formulas.
Figure 6.30 Nyquist and root locus interrelations.

prohlcm a nd thcn use the result in applying both root locus and Nyquist
tcchniqucs. ..,
EXAMPLE 6.13
Both root locus and Nyquist approaches Cor stability analysis utilize the s-planc
extensively. The root locus method plots the olosed-loop pole locations diretly in
Use the methods of Table 6.7 to determine the phase crossover frcquency (wl'<)
thc s-plane as K varies, and the Nyquist criterion relies on a conforma) mapping
for the system of Figure 6.31 , and apply the root locus and Nyquist techni4ues :i,
from the s-plane to the GH(s)-plane. Figure 6.30 shows a typical jw-axis crossing
K vares from - oo to + oo.
in the s-plane for root locus. The angle of GH(s) is 180 at ali points on the root
We forrn the characteristic equation as
locus as it movcs from the lct-half s-plane to the jw-axis at s = jw and into the
right-half plane. We may determine w by the Routh table, which is Rule 3 of the 2
root locus construction rules. Altematively, we may investigate stability by K( s + 6)
mapping a closed contour around the entire right-half s-plane,-ioto the GH(s)- 1+ , -o (6.69)
s(s + 1)
planc. A key aspect of this Nyquist approach is the GH(s}plane plot Cor ali
s = jw. Consequently, we note that the angle of GH(jw) is 180 at that frequency
We simplify (6.69) to yield
w where the s-plane contour intersects the root locus plot. In terms of the
Nyquist criterion, this frequency is interpreted as the phase crossover frequency
(wrc> In brief, the angle is always 180 on the root locus plot but thc vlue of s s 3 + (K + 2)s 2 + (12K + l) s + 36K = O (6.70)
is purcly imaginary only on the jw-axis, whereas this section of the s-plane
con tour is always purely imaginary for the Nyquist criterion but the angle is 180 which is placed in the Routh array of Table 6.8. The entry P1. i is given by
only fo r w = wrc Thus, we see the correspondence betwcen the two procedures
and the usefulncss of applying both approaches simultaneously. (K+ 2)(12K+ 1) - 36K
We have describcd four procedures Cor finding the phase crossover frequency P1.1 = (6.71)
K+2
( w re). These methods are listed in Table 6.7 as a convenicnt reference. Genera11y,
wc would seek to apply one of the first two methods initially. The graphical
mcthods listed third in the table yield only approximate solutions .and, hence, are RCsl + KC + 6)2 Y(.l
1o(1:) I < + 1)2
uscful primarily in establishing a starting guess for Newton's method. We
rccommend this fourth method only when both of the first two prove to be too
cumber1ome. For example, if Methods 1 and 2 result in a high-order polynomial
which is to be solved by Newton's method, it is more straightforward lo use
Newton's method directly on the phase expression.
Figure 6.31 The system of Example 6.13.
 jlilV A11ftlyoi@ 111 1110 Froquency Domain C lo:.u<I l oop I wq110111 y lh",p1111,1 Jt,J

th .A11il-; '' Humple 6.13

40
I( 1
1
,i 12K 1
36K
1

''i 1
I/
20

t
1
!JI
A+ n unl to 1/4 and 2/3. By using the s 2 row of the Routh 1:r:
(.!)
o
i
w = 36}(* (6.72)
o
VK*+i N
-20

li,~ 1tao 1w11 valucs of 2 and 3 from the Routh table.

l\k1 l1e 111 2 i11 Table 6.7 to yield
- 40

li11f,//( w)L-w,. ... fm K(36 - w 2 + j12w)

2
JI =O {6.73)
90 t-t- -,-- -t--- 1
- - ----+-->----+-
[ jw(l - w + j2w)
w- w,c ! ~
(Two coonc ,dent c urves)

llt !1 1, o -- ~- T 20
w~c - 13wic + 36 = O
lrn11~((l /41) yid,h thc same two values of w obtained earlier by using (6.72).
(6.74)
t -90
11 1
11,J---k.tI I w+
l lwo co,nc,dent curves)

~
l h,1g Mc1l11xl I uf Table 6.7 yields the Bode diagrams in Figure 6.32. The
t m.~11vc1' lmucncies can be determined approximately as 2 and 3. :r:
!jJjil) 1111:1 l\h lliod 4, ~e use s-plane vectors to form -180

U( j,.1/'t ) 2/(jwpc + 6) - 2/(jwpc + 1) - 90 = 180

2tan '(wpc/6)- 2tan - 1 (wpc) - 90 = 180 (6.75) - 210 L.L---L---'---.,__-..L._ __.1...._...,____...___.

Wu 111:1y 1,.:111,111gl' (6.75) and use Newton's method with appropriate starting Figure 6.32 Bode diagrarns for the open-loop plant in Figure
6.31.
!liiC\ 10 ht 1111 w,.< 2 and 3. However, (6.75) may aJso be solved directly by
/J) as
R oot locus and Nyquist plots are shown in Figure 6.33 for both positive and
tan a - tan /3 negative ranges of K. Table 6.9 summarizes these stability results for the Nyquist
11111 ( ,~ criterion .
/3) = J + tan a tan /3

(wpc/6) - (wpc ) = tanI35 = - 1

(6.76)
1 + ( w,,c/6)( wpc)
6.8 CLOSED-LOOP FREQUENCY RESPONSE
wl11d1 ~\;s
We examine the closed-loop frequency response in this section by three me1hods:
w~c - 5wpc + 6 = O (6.77) (1) contours of constant magnitude and constant phase in the G(jw)-plane, (2)
l lte 11,l111i,,11 ~ 111 ((1 11) are "'P< = 2 and 3 as bcfore. the Nichols chart, and (3) direct formulas for the magnitudc and phase curves as

1111
u1 'Y. J.\r1UfV&.b... _h I L tlu..i ,_ l! ... - - . - -
._

Closed-Loop Frequency Response 3fl',

364 Stability Analysis in the Frequency Domain

-K >O
,-plane jlO

j8 (-
\
-- .,,,.- --- --- - j40- _

j30
....._

"' '
' '
GH(s)-plane

j6
~
\ ' j':\

j4

j2
\
\.

' ' ' ,,"'=..... 2

w = 3 ,1 -
-
~) \ \
_)
ji \
\
88

-16- = - 16 .6

-14 -12 - 10
~
zero

-
jO
-30 ,
I
-jlO I
\ -j2

/
I
-j20
-j4 K>O I
/
-j6 -j30
/

""---..... ____ _ ///

- (a)
-j8

-jlO (/>)
-j40--/

Figure 6.33 ( Continued)

Figure 6.33 Root locus and Nyquist plots for Example 6.13.
Using (6.44) in (6.45) yielded an expression for M( w) as
(unctions of w. We also determine the sensitivities of these closed-Joop system
magnitude and phase curves to plant parameter variations.
IX + jYI /xi+ y2
( 6.46)
M(w) = 11 +X+ jYI = ,f(l + X) 2 + Y2
6.8.1 Constant M and N Contours
Setting M( w) as a constant M and simplifying (6.46) according to (6.47) yicldcd
Con tours of constant magnitude (M) for the closed-loop frequency response were the equation of circles in lhc G(jw)-plane with centers at X= -M 2 /(M 2 - 1),
determined in Section 6.6 as cirdes in the G(jw )-plane. To review the develop-
Y= O, and radii r of IM/(M 2 - 1)1. Figure 6.27 shows sorne typical constant-M
ment there, we recall that the plant frequency response function G(jw) was first
circles. These contours were useful in Section 6.6 for determining from the polar
cxpressed in terms of its real part X( w) and its imaginary part Y( w) as plot of a given G(jw) the peak value of M( w ), that is, MP , as an importan!
measure of relative stability.
G(jw) = X(w) + jY(w) (6.44) Contours of constant phase 4>( w) can be determined by a similar procedure.
Substituting (6.44) into (6.45) and solving for 4>( w) yields
For a negative unity-feedback system, the closed-Joop system magnitude M( w)
and phasc cf>( w) could then be written as 4>(w) = tan - 1 (Y/X) - tan - 1 [Y/(1 + X)] (6.78)

M(w)/4>(w) = G(jw) (6.45) Taking the tangent of both sides of (6.78), letting a = tan 1
(Y/ X) and fJ =
um 1111,il!!VAlil!lynl!J 111 lho r,oquoncy Domain Closed-Loop Frequency Response 367

---1
GH(s)-plane
--40
--------- G(j..,)-~lane J

"" 130

!N
j20
I
I
I
i
10 - 110 ........
........

\ - )10 ''
\
\ "<o
' '\
\~
\ - )20

',' , , ",
-)30
\

_.)
\
\
-4 3 4

',........
--- - )40
-----
(el
1,.,..,.,. ti u (Contlnued) Aoot locus and Nyquist plots for
u.-Coilpl1, 11 ' 1

ltm 11 ) / ( 1 1 .\')L 111111 using the two-angle tangent formula gives

tan a - tan P
11111 l,Ji(,J) I tan(a - /1) = - -- - -
1 + tan a tan /1
(Y/X ) - [Y/(l+X)] Y
(6.79)
1 1- (Y/X)[Y/(1 + X)) - X + X+ y 2
2
4> = - 10

,r, 1.,.1 "' )1111 (6.79) cqual to sorne constant N and simplifying yields

(x ~r + (r- 2~r = ~(1 + ;2)

1 (6.80)
Figu re 6.34 Constant-N contours in the G(jw)-plane.
Af111llcntlon of the Nyqulst Crlterlon to Example 6.13

p N z Stability which enables us to identify constant-N contours as circles with ccnters at

-- 2
X= - 1/2, Y= 1/2N, and radii r of (1/2)/1 + (1/N ). Figure 6.34 shows
o o o Stable
o +2 2 Unstable sorne typical constant-N circles in the G(jw )-plane. Therefore, curves of M( w)
o o o Stable and 4>( w) versus w can be obtained by observing intcrsections of the G ( jw )-planc
o +1 1 Unstable plot with each of several constant-M and constant-N circles at particular frequen-
cies.
368 S tab1hty Analysis in the Frequency Domain Closed-Loop Frequency Responso Jfict

40 obtaincd by noting the intersections of the plot with these contours at parltrnl.11
frequcncies. A minor difference from the con stant M and N contours is tha( thc
361 1 1 1 1 1 1 1 1 1 1 1 1 closed-loop magnitude plot obtained from the Nichols chart is expresscd 111
32 I 1 1 k: 1 1 1 1 1 11 )~. ' 1 decibels. In brief, we avoid the tcdious construction of a polar plot in using 11t1
Nichols chart to determine closed-loop frequency response curves by transernng
28 1 :J-><f 1 information directly from Bode diagrams of the open-loop frequency response.

241 1 1 ~I 1 1 ".J 1 11 1 1 1
20 ' 1 ,z4 ~ 1 1 :.;;::> -,--s:-, 1 1 \ 1 , 1 6.8.3 Dlrect Formulas

We can obtain exact expressions from (6.45) for M( w) and </>( w) in terms of thc
;> magnitude and phase of the open-loop frequency response as

1 o
8
R(w)
"
3
:::>
4 M(w) = ,[1+ R2(w) + 2R(w)cos8(w)
e
o
.
- 4 1
R(w)sinO(w) )
4> ( w) -= O( w) - tan - ( 1 + R ( w) cos 8 ( w) (6.81)
-8

- 12 whcre R(w) = IG(Jw)I and O(w) = /G(jw ). The availability o calculato rs,
personal computers, and larger computers makes thcse direct formulas especially
useful for todays control engineer. These formulas yield highly accurate result'-.
-20 depending on the preciseness with which both R( w) and 8( w) are known.
-l4i.....~..&L~-LJL.......-JL..L~-'--''---'--'---'L...J.~-'-..1.----L--'-~..__,___.___.._~.......~......... Altcrnatively, we may calculate the closed-loop frequency response fun ction ami
- 180' - 170 - 160 - 150' - 140" - 130 -1 20 -11 0' - 100- - 90 - 80" -70" - 60' then determine its magnitude M( w) and phase ti>( w) directly. The sensilivities o
/ G(J<.,) - M( w) and </>( w) to plant parameter variations are determined later in this section.

Figure 6.35 The Nichols chart.

EXAMPLE 6.14
6.8.2 The Nichols Chart
Determine the closed-loop frequcncy response magnitude and phase Cor thc
Whtle general shapes o polar plots o G(jw) can be determined casily. accurate negative unity-feedback system with the plant requency response function G( w)
rlots rcquire somewhat more careul calculations. Consequcntly. the use of given by
con~tant-M and constant-N circles to determine M( w) and </>( w) often yiclds only
arproximate results. The Nichols chart shown in Figure 6.35 provides a means o ]
ohtnining closcd-loop rcqucncy response curves by using information from Bode ( 6.82)
G(jw) = jw(jw + I)
diagrams directly. T hese Bode diagrams presumably can be constructed fairly
accuratcly. Thc Bode magnitude plot yields values for the vertical axis o the
Nichols chart and the Bode phase curve gives corresponding values for Figure 6.36a shows the G(jw )-plane plot for (6.82) and its intersec tions w11h
thc hcyi1ontal axis. Thercforc, a plot on the Nichols chart has frequency as constant-M and constant-N circles. Curves o M( w) and </>( w) are shown 1n
thc varying parameter along the curve. The codstant-M and constant-N circlcs in Figure 6.36b for these interscction frequencies. Figure 6. 37a givcs tlw llo<k
thc Ci(jw ),plane are transferred to the modified contours shown on the Nichols diagrams for G(jw ), Figure 6.37b plots this information o n thc Nicho !~ dt:11 I
chart of Figure 6.35. As before, thc closcd-loop frequency response curves can be and Figure 6.37c provides Bode diagrams for the closed-loop frcqucnl'Y n,po11 ,c.
 1 11tlllly Analyss in the Frequency Domain Closed-Loop Frequency Response 371
III

1 1- ;4.

""M = 1. 1
M = I.O ,
G(jwl plane
J

M f:I
M=0.8 ~

2 3

-4 3 4

/=03 -j3

-j4

t
w = 0.2

(al

t luuru 6.36 Determining M(w) and ,,(w) by using constant-M and constant-N
e II e l11~,

111 l"1ng thc direct formulas of (6.81), we first calculate

1
R ( w) = 1G(jw) 1 = w/w2 + 1

(al (Co,ili11uedl

O(w)=/G(jw) - 90 - tan - i ( w) ( 6.83)

Figure 6.36 ( Continued)

1J"11g thc,c rc!>ults in (6.81) yields the M(w) and q,(w) plots wilh equally spaced
l111111rnl 1c\ givt:n in Figure 6.38. Alternately, we can determine M(w) and q,(w)
a, 1h~ 11111g111tuc.Jc and phase of the closed-Joop frequency response function given
 368 Stability Analvss in th"' i=ro,... .. --- '"'

Closed-Loop Frequency Responso '. 1/ 1

372 Stab ility An alysis in the Frequency Domain
20
l.5

t 1.0 t.o
ol
0.2
1 1 ~.. /_ 1 1
,, , ~
1

~
...:9

!
0 .5 -
- 20

2
61--
3

~ ~ 1

)
? .,_
3'
o I
0 .2
1
0 .3
1
0 .5
1 1
2
1
5
w
- 1

t
- 9o
:o t
. .. .

~
- 1so - - ----------------
( b)
- 180' ~ --1- - - .... _ _ _ - - - - - - - .... - --- - -- ---
Figure 6.36 ( Continued) Determining M( w)
and </> ( w) by using constant-M and ((I)
constant-N circles.
Figure 6.37 Determining M(w) and </>(w) by using the Nichols
by (6.45). These calculations yield chart.

1 1 contours or the Nichols chart in converting from open-loop to closed-loop

M(w) = -;== = requency responses. However, ali three methods of this section are subjccl to
V(l - w2)2 + w2 /i - w 2
+ w
4
errors due to inaccuracies in G(jw ). We now examine the scnsitivities of the
closed-loop frequency response magnitu4e and phase to plant parameter varia-
<t>(w) = -tan- (-w-
l - w
) 1
2
(6.84) tions.
As discussed in Section 3.4 in introducing the sensitivity concept, we may
approximate incremental errors in M( w) and </>( w) resulting from small varia-
Thc student is asked in Problem 6.32 to verify that using (6.83) in (6.81) yields
tions in a single plant parameter a as
(6.84) for ali w.
aM(w , a)l 6.a
6.M(w) = aa N

6.8.4 Sensitlvities to Plant Parameter Variations - a4>(w, a) 1 6.a (6.85)

I 0.<J>(w)- aa N
Two kinds of errors sometimes associated with closed-loop frequency response
curves are: (1) graphical errors, and (2) errors dueto an imprecise knowledge of
where 6.a is the incremental variatio n about sorne no minal valuc nN a nd thc
G ( jw ). Thc first error can occur when using cither the constant-M and constant-N
 Closod Loop I ruquoncy Hu~.po11tHJ 3/!i
16 tiilOy l\t1 i1ly11lo 111 tho I requency Domain
2or-----,~~-r--~~.--~~-,--~~~....--~~~

t
.Q
..31
3
~
0.5
~...... 1 1
5
w- 1

-20

O 1 1 1 J l l 1

t
3
"& -90

-180 ~ __ ._ - _ .... - - --+-- - - _,_ -- -- _ ... - - - -

(e)

1~O" - 140 - 130 -120 -110 - 100 - 90 - 80 - 70

Figure 6.37 ( Continued)

/G(jw) - a4>(w,a) = a4>(w, a) ( aR) + a4>(w, a) ( ao)

(b)
JR aa ao aa
1,1,~.. ._. 6.:JI ( < m1t1nued} Determining M(w) and 4>(w) by using Nichols chart.
ii Ui11 1i111 <., >I N 1ml1c..a t~ that the derivatives in (6.85) are evaluated at the nominal
U)11d11i,~11 , lly m 111_; (6.81) to evaluate the partial derivatives in (6.85), we obtain
-sinO(w) (R)
= 1 + R 2 (w) + 2R(w)cosO(w) -;;
.IM(w , ") aM(w ,a) (R) + aM(w,a) ()
aR aa ao aa
d,w
1 + R{w)cosO(w) ( ) (6 .86)
1 + R(w)cosO(w) ( aR) + 1 + R2 (w) + 2R(w)cos0(w) aa
lt + R ( w) + 2R(w)cosO(w)}3
2 12
aa
Alternatively, we may either perform the operations in (6.85) directly on M(w,
o:)
R
2
(w) sinO(w) ( ao) and <f>(w, a) in (6.45) or determine numerical values for these parta! derivatives
1
[1 + R 2 (w) + 2R{w)cosO(w)] 312 aa
 Closed-Loop Ffequency Responso 'JI I
376 Stability Analysis in the Frequency Oomain

TABLE 6.10 Numerlcal Results for Example 6.15

1.5
iJM(w, a) 1
w M(w) <f>(w)
iJa N

1 1.0 o.o
0.5
1.00
1.11
0.00
- 0.34 -
. O.O
-'33.7
- 1.00 - 9.0
...J 1.0
1.5
1.00
0.51 - 0.30 ,- 129.8 -1 "'
0.5 2.0 0.28 - 0.09 - 146.3
2.5 0.17 -0.03 - 154.5
3.0 0. 12 - 0.02 - 159.4
These values are yet to be multiplied by 57.3 to convert them from radians to degrees betor1,
2 3
,~- using in (6.85).

1 ~
2
L-
3
..,- 1.5
Exactb,,= -0.2
.

t lncrementallt.~= -0.2
3
~ -90.
t 1.0

$.
- 1so 1- - - - - _ _ _ _ ______ _ __ _ 0 5

Figure 6.38 Curves of M(w) and <f>(w) from (6.83).

at a particular frcquency by using the incremental relationships

aM(w, a)
=
M(w , a + Aa) - M(w , a)
o
o 1 2 3
w-
Aa o~

a<f,(w,a) <p ( w, a + 6a) - <f, ( w, a)
(6.87)
2 3
,,,~
-
6a
t -90
where 6a is chosen to be sufficiently small.
:o

-- -- -- -
EXAMPLE 6.1 5

Determine the partial dcrivatives in (6.86) and (6.87) for the plant frequency
resf>onse function given by Figure 6.39 Comparisons of M(w) and <f>(w) ob-
tained by incremental and exact methods for .a =
l 0.2.
(6.88}
G(jw) = jw(jw + a)
 Prol>h.1m -:. 3'/9
1111 1n 1111, 1 ruquoncy Oomain
S. H. Nyquist. "Regeneration Theory." Be// System TecJmical Joumal, Vol. 11
11,v,;: th.11 1111J1cate the extent of incremental variations
(January 1932), pp. 126- 147. ,
6. l. M. Horowitz. Synthesis o/ Feedback Systems. New York: Academip Press,
tli fl 1,!111111\ o the rcquired calculations, and Figure 6.39
0Kli 11111c v11111111ons with exact calculations for the closed-loop 1963.
7. H. M. James, N. B. Nichols, and R. S. Phillips. Theory o/ Servomechanisms.
lh\''il, Wc w, that the incremental method of calculating
yic li.l , vahan that daffer from the exact values by less than New York: McGraw-Hill, 1947.
8. B. C. Kuo. Automatic Control Systems, 4th ed. Englewood Cliffs, N.J. :
111t 11t ,,l 111ul ,xact results for <f,(w) are indistinguishable from
Prentice-Hall, 1982.

PROBLEMS
~m
(6.1) 6.1 Determine the closed-loop system bandwdth (BW) and peak mag-
!11rd t) i1 "'''\ p11'\ln1ed in this chapter for the stability analysis of nitude for the closed-loop frequency response curves (magnilude
111\'lli iilil 111,lh u 1,; .. ystems in the frequency domain. Mapping the only) shown in Figure 6.40.
111\H \\ l,i;I, e 111 lm,, the en tire right-half plane into the GH(s )-plane
11lfl!iCLl 1,y 1111,1 ., kctdung Bode magnitude and phase diagrams for the
1 f1n1\[W y 1cs111111w f1111c1ion GH(jw ). The required polar plol can then
tri! (111 K @ ji.> 11ul lhc Gl/(s)-plane contour completed by performing
11 !!n 111111 {lito111 1111 ,cal ax,s, corresponding tos= -Jw. Cases involving t
jttl Md w1H u>nMdered by selt:cting an infinitesimal semicircle
lid ,,!,), e" huhng the pote from the interior of the s-plane contour

i
1 iillO 1h,) t,//(i) plane. Comparisons were made between Nyquist 0.5
111il 1h11,c obtamed by using the root locus technique o

1l1!!ili1y ,,,111 qll, 1n the frequency domain involved the closed-loop

\, nrn~1111t1(lc! 111 thc requency response ( M,) and margins Cor open-loop
Hi! (ti ,\ () 111111 pbu~e ( PM). Numerical methods were presented for
lliij\ M,. , U/11 111111 l'M The closed-loop system frequency response was
OL----'---.-L-----'-----'-----
0 10 1~
5 20

(a)
... -
1 hif !H'lWti 1 111111y feedback systems by constant-M and constant-N 1.0 1 -
I!! !ill U( j,.i) ,pl,1111, the Nichols chart, and direct formu las.

t
!,l,iti1NC'I

i 0.5

y, t:[1111:,1111 , l11d / HEE Standard Dictionary o/ Electrical and Electron-

/ I '/r1111 r J11d cd Ncw York: The Instilute of Electrical and Electron-
iU i f11r111n 1,, 1977
~lrl Mt 111111 I> G. Schultz. Linear Control Systems. New York:
"' ' l Gl'[I\\ 11111, 1969.
1 D'AU 111111 < 11 lloupis. Linear Control System Analysis and Design:
2nd ed. New York: McGraw-Hill, 1981.
'01,11,11111,1111/ 11111/ Modem,
o L-._
0
__.L_ _
2
..J..._--'----'--...J.:---'.':---
4 6

(b)
8 10

Figure 6.40 The c losed-loop frequency response mag-

12
... -
UiliE,11 ,mal I B. Tuteur. Control System Components. New York:
MPJn1w 11111, 11>S8 nitude curves for Problem 6.1.
 Problems 38 t
4

( 6.1) 6.2 Determine the gain margin ( GM) and phase margin ( P M) for c, ll h
of the two systems having the open-loop frequency response curvl''
for magnitude and phase shown in Figure 6.41. Assume negativr
unity feedback.
1. 2
- (6.2) 6.3 Determine the closed-loop bandwidth, closed-loop peak magnitudl'
gain margin, and phase margin for each of the following systcm,
The given open-loop transfer functions are fed back negativcly 111
each case to form the closed-loop systcms.
2 4 6 8 10
.. -
12

a. G(s) = s(s
4
+ 1)

t -90'
O'K 1
2 4
1 1
6
1
8
1
10
. -
1
12
b. G(s) = s(s + 2)

c. G(s) = (
100
)
10

~
s .r + 1
( 6.2) 6.4 Suppose we want to ensure that the closcd-loop systcm peak magni
- 180"
tude ( Mp) is less than or eguar to .!.4 Cor each of the systcms m
Figure 6.42. Determine thc range o( a that satisfies this condition
- 210 r--------~;-----------
5
-
nc.,1 +
:& ) ' 1 "
h + ll
Y(.)

t
1:
3
(al

G
R!.I + 100 Y(.)
2t-
< + ,.,

o .____,L__ ~_ _ _...__ _. __ _,1__ _

(b)
o 3 5
.. ___. Figure 6.42 Closed-loop systems for Problem
6.4.

-
0 1----...__ _...__ _...__ _...__ _...__ _
O 1 2 3 4 5

.. (6.2) 6.5 Determine the three ranges of K that satisfy the following critcria

t -~ ,
for the system given in Figure 6.43.

R(~l + K Y(.I
~ - 180' ---1:1----+-1
sis +9l

-270'
(6)

Figure 6.41 The open-loop frequency response curves for

ProblC'm 6.? Figure 6.43 The closecl loop systcm for Prob
 Problems 383
IH !! !!J l 1(H 1111111 y l>omuln

R(s) + Y(s)
il j i ~Yh l1;111 handwidth ( BW) is less than 10
,iI .
I
i nup ~y1 11 111 111ak magnitude (Mp) is less than 1.2.
!ti!F t ii, </'Al) ,~ ~, culer than 50.
(~)
1 - a
2

i1, ..11,d ,,1vo111c<:hanism of Figure 6.44 that has

(i)
!Ylii ll} 1 J('7, volt!>, n = 10 turns, R = 10 il, and
11. ,1 ~ iJ 1 /, , () 001, and B eq = 0.008 in SI units.
Y(s)
!l!!l i [lil:~ ni po, rtivc K for which the following three K< - 1)2
IHliillfli l!'.i! -h rt 1111\hcd simultaneously: :1!..1 (a + 1)2 (s2 + 1)

t; lt uii ;v1 1 of O,( t) due to a step input, that is, due

'" ' " '
1~) , 1/ I, 11111 w cater than 30%.
1
(ii)
1l11mJ :11) 111 "hclwccn 40 and 50.
1!.,1 dv-~1,,1, 111111 duc to a unit ramp input, that is, dueto

f -
~"" '~
1 ! l.! 1, 11 11ut l&ll'atcr than 0.1 radian.
11

~l --/ ~ : I J<Q -
R(s) +

' // Beq

/ (iii)

t.. t
/
/ o.
/
Figure 6.46 Closed-loop systems fo, Problem 6.8.

]' the above expresson by comparing the result with that obtained
from (6.12). In evaluating the above expresson, first compute MP
1 1111

from (6.9).
,no no 111, ,11,11 ,orvomechanism described in Problem 6.6. (6.3) 6.8 Consider the closed-loop systems of Figure 6.46.
a. Provde rough sketches only of Bode magnitude and phase plots
, ,11'\\1011:, are often useful in the design of a feed-
11 11111 , ,
for the open-loop transfer functions. Show asymptotes in ali
y~ if 1i1, e 1t11 \ ll\ h approximation between phase margin (PM)
cases.
11tl liw prn t 11111p111tudc ( M1, ) of the closed-loop frequency response b. Sketch polar plots of GH(jw) in ali cases.

PM = sin - ( ;J (6.3) 6.9 Sketch linear plots, Bode diagrams, log-log plots, and polar plots for
each of the following open-loop transfer functions for s = jw.
lO(s + 1)
1 11 I 1l1 ~V fH'111 .,how11 in Figure 6.45, demonstrale the accuracy of a. GH(s) = -s-+- -
2
lOO(s - 1)
~1 ,,(, +
10
1)
Y(s)
b. GH(s) = s 2 + 29s + 100
lOO(s + 10)
c. GH(s) = 3
s(s + 1)
(6.4) 6.10 Use s-plane vectors to construct contour maps in the F(s)-plane for
the con tours shown in Figure 6.47 for each of the given F( s)
iIWO \i tl.'11> 1ho :iystem for Problem 6.7.
384 Stability Analysis in the Frequency Domain Problems 3ft',

R<s> + Y(s)
:t ) ' 1 (s + 5) (s - 1>2
s-plane +J,-----1
n i 1
-2 - 1 10 +1 +2 +3 +4

-jr----_J
1
(a)

(i)
R(s) + Ks Y(s)
l:) )o 1
,2 +2s+2

.~-plane +1 -
'~1
----, s - 1

I 1 1 (s + 3)(s + 4)

-2 -1 1 2 3 4 (b)
' 1 1
__ _ _J
- ; l.- L- Figure 6.49 Systems for' Problem 6.12.
...
(6.5) 6.12 Construct the polar plots for the open-loop frequency responses ol
(ii)
the systems given in Figure 6.49, and apply the Nyquist criterion to
Figure 6.47 The s-plane contours for Prob- determine the number of closed-loop poles in the right-half planc
lem 6.10. Draw a plot for positive K and anothcr for negative K . Specify your
ranges of K in a table for different stability conditions.
functions. Determine the number of encirclements of the origin in (6.5) 6.13 Consider the feedback control system of Figure 6.50. Sketch thc
the F(s )-plane in each case. polar plot in the GH(jw )-plane, and use the Nyquist criterion to
JO(s - 1) determine the number of closed-loop poles in the right-half s-plane
a. F(s) = 2
as K vares from O to + oo.
(s - 3)
2 + t)
JO(s - 1) R(s) ,, Kb+l)
Y(s)

b. F(s) = s(s - 1) (s - 5)
s- 3
(6.5) 6.11 For the system of Figure 6.48, determine the range of positive K Cor
stability. Sketch the Nyquist plot (roughly) for K = 2 and K = 10,
and use the Nyquist criterion to determine whether the closed-loop
system is stable for each gain. A Bode plot (particularly the phase Figure 6.50 The feedback control system for
part) may be helpful as a preliminary step, but this is not necessary. Problem 6.13.

( 6.5) 6.14 Sketch Bode diagrams for the open-loop plan! of the systcm in
R(s) +A
---,,~\ l: J
I K(l - s)
(s + 2) (s + 3)
Y(s)
Figure 6.51, and then apply the Nyquist criterion for positive K

R(s) + K Y(s)
l:) ' 1
s2(, - 1)
I

Figure 6.48 The closed-loop system for Problem

6.11.

Figure 6.51 The system for Problem 6.14.

 u, 111,, F,oquuncy Oornain 1'1111,1()111:l :JU'/
11
i ,,1111,; 1lic 1111111111.:1 o clo)ed-loop poles in the right half Jf) + K(, + 10)2 Y(,l
i f!h\ii( .21. + 1) < + 100)
I
,1r~ li ~J11111111 1 oo for the system of Figure 6.52,

(al

Y(s)
R(s)
+ (i) 1 K 1
Ylsl (2 - -;,-, + .21

(~!..!)
1 -
2

(b)
Jil(jo1 t .5:1 lho system for Problem 6.15.
Figure 6.54 The systems for Problem 6.17.
!O .v,h "" o Figure 6.53, apply the Nyquist criterion to
1

L!r.1r rn1i111J 1111111hcr of closed-loop poles in the right-half s-plane

1h1
,;.. vd, oo to + oo. You are not to use the Routh table
11 0 111
(6.5) 6.19 For the systems of Figure 6.55, use the Nyquisl criterion to de-
termine the number of closed-loop poles in the right-half of thc
"' i.li N n:111 uf thc problem. Sketch the root locus plots to verify
s-plane as K varies from - oo to + oo. Tabula te your results in a
V!lll" ~111111111,
table showing the different stable and unstable ranges of K.
+ 11
K(s Y(s)
~ i)3""

1
,, Y(>l
1

t :_~
R(sl KO - s/ 100)
,,
.r ,(1 - s)2

.i
- - '
'

L
(a) (a)

1/(1) 1 Y(sl
( 1:) ,, K(2 - s) 1
, 3($ + 1)

K<s + u2 Y(sl
~ 7(.~
(b)

i :111ure 6.53 The systems for Problem 6.16.

(b)
td 1 l(ew:11 thc 1mtructions of Problem 6.16 for the systems of F igure
'~ ~I Figure 6.55 The systems for Problem 6.19.

w ,l (!;I I \l'plv 1l1l Nyqui)t criterion to the closed-loop systems of Problem (6.5, 6.20 Severa! methods may be used to investigate the stability o f feed -
back systems. Using either the Nyquist criterion or the root locu~
;,';;., Sp11:1fy a table the number of closed-loop poles in the
in 5.3)
11yli1 li,1 lf I planc as K varies from - oo to + oo. Show the polar technique (not both), determine the stability of the system in Figure
1,1111 111 1h1 (ill( Jw )-plane, and indica te clearly values Cor ali real-axis 6.56 as K varies from O to + oo. Specify the number of close<l-loop
, , ., ,1111., ln<l1ca1e in your table those ranges of K Cor stability. poles in the right-half s-plane as K vares.
 Problems ;11.11_1
388 Stability Analysis in the Frequency Domain
ji ji
R(sl + (i) 1 K(. + 2) (s + 3) 1 1
Y(s) G(jw)plane Gtj,,,J .plane
'l: ... (s + l)(s-1)2 '
JO JO
-2 -2 o

-ji

Figure 6.56 The feedback system for Problem

6.20. -j2

( 6.6) 6.21 For the G(jw )-plane polar plot given in Figure 6.57, construct
severa! constant-M circles that intersect the polar plot and sketch a
linear closed-loop magnitude frequency plot of M versus w. Use (al (/,)
(6.9) to compute MP for this second-order system, and compare the
result with the max.imum value of your curve obtained graphically. ji JI
G( jw)plane GC jw)plane
-2 -1
JO
-2
Y(sl
R(~l + 10
~~--~l:) ... ,
s(s + ll

-j2
(al

(e) (d)

Figure 6.58 Polar plots of G(jw) for Problem 6.22.

(;(Jcolplane
(6.6) 6.22 For cach of the polar plots of G(jw) given in Figure 6.58, determine
O, jO
thc gain margin, phase margin, and thc maximum value of the
-4 - 2 <.,, = 4
-8 -6 , = 31 closed-loop frequcncy magnitudc curve.
(6.6) 6.23 The closed-loop system bandwidth may be found graphically for
those systems for which M( w = O) = 1 by constructing the con-
stant-M circle for M = 0.707 and noting the bandwidth as thc
frequency at the intersection of the polar plot. Determine and
-j4 closed-loop bandwidth for thc two cases given in Figure 6.59.
(6.6) 6.24 Use Newton's method to compute the phase crossover (wr,) and
gain crossover ( wcd frequencies for the following plants having
-j6 ncgative unity feedback. Determine thc gain margin ancl phasc
margin for each case.
1
-jB
a. G(s) = s(s + 4)
I 5
b. G(s) =
(b) s + 10s 2 + S + 1
3
1
c. G(s) ~ - ,
Figure 6.57 The feedback system and associated ( ,f 1 1)
 in iliu r 11tq11unc y Domain Problems 391

l 1- 10 1 Y(sl
0( Jw)-plane Rlsl + (E)
-1 s($ + 2)
,
~
1
l. 1 1 .
';10
O

1 y (a)

I{I
I - /
-;-i:
/
Y(sl
/
R(s)
+ (i)
I s(s + 1)
- )2 /
111)
--- _ _.,,,
//

(u) (b )

Figure 6.60 Closed-loop systems tor Problem

ji 6.25.
O( jw)-plane

j 1,1 12 ,
l J
Ylsl
:r:i:~

. ~/l
5
o 2 J R (s)
s(s + 1) (s + 2)
I
I
-JI
I
I
I
J2 I Figure 6.61 The system for Probtem 6.26.

,.\ - j3

/
/
/
/
/
/
(6.8) 6.28 Using constant-M and constant-N contours, plot curves of the
closed-Ioop frequency response magnitude and phase for the nega-
tive unity-feedback systems with p lant frequency response functio ns
... 1 , )~ -- -
given by
(b)
10
lelj lllf) tl,!>9 Polar plots ot G(jw) tor Prob- a. G(jw) = )W
. ( )W
. + 2)
I\Jl!I li ;1,\
3
1 inr 11Hi111 t ht dosed -loop system peak magnitude (Mp) for the
b. G(jw) = jw(Jw + l)(Jw + 2)
\'1i l i,' 111 M 111 l 1g111 6.60 by
( 6.8) 6.29 Repeat the instructions of Problem 6.28 using the NichoIs charl.
U~111, (/, M) and (6.60).
(6.8) 6.30 Determine and plot M(w) iri decibels versus w for the systcm of
l!, ( ( ,111p11i111~ {and then using (6.9).
Problem 6.21 by using the Nichols chart. Detennine and plot </>( w)
(1 ( 'u!l~i,J,~ llw , y,1cm of F igure 6.61. Find MP by using (6.63) to versus w for this system by using
tl!!II 1111 11 11,111011 fo r wP' using Newton's method (if necessary) to
jn, f11 ,.,, ,, 1111d ~ubstituting that value into (6.60) to determine Mr a. Constant-N circles.
b. The Nichols chart.
!1 f 1'10 y,h-111, of Figure 6.62, sketch both root locus and Nyquist
( 6.8) 6.31 Determine and plot curves of M( w) and </>( w) versus w for the
pi, il N 1~ 11 v1111~-, from - oo to + oo. Determine the phase crossover
~n,p1 ( 111 y 111 t11 d1 case by selecting an appropriate method from systems of Problem 6.28 by using
l ,iljl,1,~ / '\111n111:11 izc in a table the results obtained by applying the a. Direct formulas in (6.81).
lIY(11ii n1 1 t1irnon to thc given systems. b. Drect formulas resulting from (6.45).
 392 Stability Analysis in the Frequency Domain

+ Ks2 Y(s)
1
;-+ """jj6

(a)
PART 111
K(< + 1)2
7-
}'(s)

CONTROL SYSTEM
+ 2
,. ;.ocijz
DESIGN
(b)

+ K
(.+2)3
Y(sl
..
JO
<(s+J-;- +41
(d

Figure 6.62 Closed-loop systems for Problem

6.27.

(6.8) 6.32 Verify that the results in (6.84) are obtained by substituting (6.83)
into (6.81) for ali w.
(6.8) 6.33 Determine the partial derivatives in (6.85) for w = O, l, 2, and 3 for
the closcd-loop system having a plant frequency response function
givcn by
10
G(jw) = -.-( -.-+-)
JW JW a

and evaluate these derivatives for aN = 1. As in Figure 6.39, plot

nominal, incremental, and exact magnitude and phase curves for
.a = 0.4.
 :5
,...
u
ex;
") t;
'-4 L-d
..,;,6,f:il!/ o _.J
4','y4~};~ ~ w

7
-
"' /,4[,1(/l,,,w::~'

J'
:;t:N
1 #:f}
vr.zs1~
.,\~-:.,. :..~1t
::it {ii ii,,,:.~
8 'E!.~~r;{: ~V--:J I '-~ :.;:
'l,_.~
?.l'~~ti~ ~;~(
,......., ~

"-
L :
~
t'
e
. '-::'
P

1- .
e

Classi~al Design ...

"'

by Series Compensation
The control engineer's possibilities for design are far more extensive than the
comparatively routine methods of analysis. As described in Part JI, analysis
techniques are often preset step-by-step procedures used to determine system time
responses (Chapter 4), stability (Chapter 5), and frequency responses (Chapler 6).
In contrast to the analysis goal of obtaining the output response when given the
system and its input, the goal of design is to determine one of the severa! possible
systems yielding a desired output response for a given input Analysis enables us
to explain what can happen with an existing system under known circumstances;
design requires that we create a new system that meets certain slandar<ls of
perfrmance. The control design task requires creativity and ingenuity in selccting
both the structure and parameters of a feedback controller to be placed around
the plant such that given time response or frequency response performance
specifications are satisfied.
A mastery of analysis procedures prior to attempting control systems design is
important. because (1) it provides sorne experience on the output responses to be
expected from certain classes of plants and controllers with given inputs, (2) it
provides the background concepts, for example, root locus and frequency re-
sponse techniques, necessary for solving design problems, and (3) it permits the
control engineer to verify subsequently that the designed system model does meet
the required performance specifications.
Controller design can be both challenging and rewarding. The challenge is
selecting a controller structure and parameters that satisfy the design require-
ments and are relatively insensitive to planc parameter and input variations. A
reward is observing the successful performance of the designed controller when
applied to the actual plant in its operating environment. Furlher field tcsting may
reveal a need for refining the plant model descrption. If so, a redesign of the
controller may become necessary.
The material for design in this book is described in two chapters. Chapter 7
presents root locus and frequency response techniques for the classical design of
series compensators to be inserted directly into the control loop. The resulting
closed-loop system is of higher order than the original plant model. Chapter 8
desc~ibes state variable design using optimal control theory, which requires that
more variables than only the plant output are accessible. Often, as in ple
395
 Design Speclflcatlons Rovl'lltml 3QT
396 Classical Design by Series Compensation

placcment or state variable eedback (SVFB), all state variables are required or
control. The closed-loop system resulting from state variable design is the same 8 pllll('

ordcr as the original plant model.

7.1 DESIGN SPECIFICATIONS REVISITED

T hc time respo nse and frequency response design specifications defined in

C hapters 4 and 6 are cxamined in this section with respect to whether low or high
values o f the forward loop gain K are required . Recall that time response
spccificatio ns are percent overshoot, rise time, delay time, settling time, and
steady-sta te error. Frcquency response specifications are closed-loop system band-
width, systcm p eak magnitude, phasc margin, and gain margin.

7 .1.1 Percent Overshoot and Steady-State Error

.
Dcpending o n thc order and structure of the closcd-loop system, the adjustment Figure 7.1 The root locus sketch fo r (7.1) with
o f the forward loop gain K for an acceptable percent overshoot and steady-statc K > O.
error can bccome a design tradeoff situation. For example, considera second-order
Therefore,
T ypc I systcm that has negative unity feedba ck with open-loop poles at the origin
and - p 1, that is,
e,s = 1/ K,, = p 1/ K ('7.4)
K
G(s) = s ( s + P 1) (7.1)
which requircs a sufficiently Jarge value of K to ensure that e,. is not grcatcr tha11
sorne specified value e., . In brief, the percent overshoot design spccilic:1t10 11
Figure 7.1 shows thc root locus plot of thc closed-loop potes in the s-plane as K requires a low value of K and the steady-state error specification a h1gh valm
vares fro m O to + oo. The two loci start o n the open-loop poles, move along the of K . .
real axis toward cach othcr, and forma breakaway point at s8 = - p 1/ 2 with a
value of K equal to pU 4. For larger values of K, the underdamped time response
fo r a step input has a percent overshoot given by (4.25) in terms of the damping EXAMPLE 7.1
cocffi cient t. r
Lines o f constan! in the s-plane are defined by
F or a system defined by the G(s ) in (7.1) with p 1 = 2 and negativc untl'i
feedback, determine the ranges of K satisfying _each of the following dcsign
(J = cos - 1 (r)
(7.2)
specifications.
w he rc (J is the anglc shown in Figure 7.1. lf the pcrcent overshoot (%OS) is
%OS s 15%
requircd to be less than or equal to sorne specified value %0Sm""' then K must be
sufficiently .small ( ~ Kmax) to cause the closed-loop poles to lie withn or on the
bound aries o f the wcdge-shaped rcgion in thc s-planc. e"I u ni1 ~ 0.05 radians (7 '1)
ramp
J7or thc T yp c 1 system having the G( s ) given in (7.1). wc determine the
<.tcadv- tatc error ( e., ) due to a unit ramp input as 1/ K ,,, where K ,, is the velocity
error cons t3,nt given by In .satisfying the first requirement in (7.5), we use (4.32) lo determine L....
corrcsponding to a 15% overshoot as 0.517 and thcn relate the range o f K to f hy
K,, = lim sG(s) = K/ p 1 (7.3) identifying K as the square o f the natural frequency w~ and p 1 ac; 2t,," .
s- 0
 Oesign Specifications Revisited 399
398 Classical Design by Series Compensation

Therefore, We may satisfy ali three specifications simultaneously by selecting the most
restrictive (lowest) value of Kmu For example, we would select K =,K3 for the
Pi 1 case shown in Figure 7.2, since any value higher than K 3 violates the third
! = 2/K = /K (7.6) requirement in (7.7). Incidentally, we note that any value of K higher thhn K 2
violates the second and third requirements in (7.7) and any value higher than K 1
from which K ~ Kmax = 1/!.;.;., = 3.74. violates all three requirements.
Since ess = 2/K from (7.4), we may satisfy the second requirement in (7.5) by
setting K 2::. 2/0.05 = 40. Obviously, the conflicting design requirements in (7.5)
cannot be met simultaneously by adjusting only the gain K. EXAMPLE 7.2 .

Determine the allowable range of K that simullaneously satisfies the three design
specifications
7 .1.2 System Peak Magnitude, Phase Margin, and Gain Margin MP ~ 1.5
Frequency response specifications for Mp, PM, and GM can often be satisfied by PM 2::. 40
selecting a suitably low value of K. Figure 7.2 shows a typical G(jw )-plane plot (7 .8)
GM2::.4
for three values of K corresponding to the three equalities for the design
specifications given by for the negative unity-feedback system having a plant transfer function G(s)
Mp ~ MP .... given by
K
(7.9)
PM PMmin G(s) = s(s + 2)(s + 40)
GM 2::. GM,nin (7.7) Figure 7.3 shows the G(jw)-plane plot for (7.9) with K set equal to 250, an
approximate value selected by examining the phase margin specification. We first
MPmax

G(jw),plane
G( jw)plane
OM,,,,,,

-2

w = 8 .9
-11
1
\
\
\
'
K1 K2 K3

Figure 7.2 A typical polar plot of G(jw) for three values

ot K corresponding to Mp, PM, and GM design specifica-
Figure 7.3 The polar plot of G(jw) in (7.9) tor K = 250.
tions.
400 Classical Design by Series Compensation Design Speciflcoll<>111111:,vwit,i, 1

determine the gain crossover frequency w 0 c that yields a phase margin of 40 for The gain margin specifieation GM ~ 4 in (7.8) is eas,ly appl1c-d hv 11 111~ 11t., ,
(7.9) from (6.2) as Routh table to determine the maximum K for stabi lity ( K*) a, I IW I h, 11 1, ft
for K = 251.1, we have a gain margin of K* / K = 3360/25 1.1 1 111, \\ 1111 lt 11,
PM = 40 = 180 -1/G(Jw c) 1 0 well within the spcc!fication. More?ver, jt is easy to determine O , A IMO ,, , t ti,!
range of K that sattsfies GM ~ 4 m (7.8). In summary, thc ranp,c 111111 \,111 .111 , d1
= 180 - l[-90 - tan - 1 ("'t)- tan - ("';t)]I (7.10) three design specifications simultaneously is O $ K $ 251 .1.

which reduces to

tan-l ( W;c) + tan - l ( W;n = 50 (7.11) 7.1.3 Rlse, Delay, and Settling Times and System Bandwldth

Taking the tangent of both sides of (7. 11) and using the two-ang]e formula yields Consider the time response y( t) Cor a second-order underdampcd sy.~l \'1 11 li .11111 ,
"'oc+Woc
- -
unit step input. We recall that y(t) was given in (4.16) as
e- r.... ,
2 40 50 (7 .12) y(t) =1- ,.---;-;; sin(w,t + O)
l _ wOC )( wOC ) = tan
( 2 40
1- 2 r
Although exact normalized curves of rise time. dela y time, and sel 11,ni 111111 11, 11
which can be solved to give w0 c = 2.14 radians/second. We now solve presented in Figure 4.6, we are now interested in examining rdaird ti, 11 11
K specifications by forming conservative estimates based on o nly lhc 111111, ,111d
- --== =-= ==2 = 1 (7.13)
woc/w~c + 2 /w~c + 40
2 lower exponental envelope curves denoted by y/(t) and y,- (t), rc~pt'( 11v1 l_v .
gven by
to obtain 251.J as the value of K for which PM = 40. Therefore, the range of K y/(t) = 1 + e - fw.1
satisfying the second requirement in (7.8) is O $ K s 251.1.
lt appears from Figure 7.3 that the phase margin design spccifieation is the Ye- (t) = 1 - e -fw., ('/ I '/)
most restrictive of the three given requirements. To examine this eonjecture For example, we may approximate the rise time t, by the time rcquim l ,11 lli.
further, we calculate the values of MP and GM corresponding to K = 251.1. lower envelope y,- (t) to move between values of 0.1 and 0.9, that is.
U sing (6.64), we solve for wP Crom
d d
tr = tiy,-(t) - 0.9 - tiy,- (t)-0.1
( K - 42w;) -d ( K - 42w;} + wP(80 - w;) -d [ wP(80 - w;)] = O (7.14) 1 1
wP wP
= [-fw,,
-- In (1 - 0.9)]- [ -- tn (1 - 0.1)]
-fw,,
which simplifies to yield
1
3w; + 3208w; + (6400 - 84K) = O (7.15) =-ln9 1
fw,,
(7 I K)

For K = 251.1, we obtain wP = 2.136 radians/second and compute MP from

Similarly, approximate expressions for delay time f and settling time t , are
G(jw) 1 1 l
MP = 1 + G(jw) K - 251.1 I = tly (,) - os = - ,.- ln(l - 0.5):: -- ln2
1 w-w,- 2.136 ~ - ~w" ~"'n
K 1., =- 11y,-<1>-1 - 8,
= -- -,.
1 In 8, =- 1 In (1)
- ( / 11>)
(7 .16) y:u> - 1+8, - ~w,, fw,, 6,
2 2
/(x - 42w;) + w;(so - w;) K-251.1
.,,-2.136
where 26, is the width of the settling time band, that is, 1 - 6, $ y (t) ~ 1 1 h,
which gives MP = 1.46. Thereore, the design requirement Cor MP in (7.8) is
barcly satisfied Cor K = 251.1, indicating that only a slightly higher value of K
1
would be allowed if PM ~ 40 were not also a design requirement. As an altemate exprcssion for rise time, Fortmann and Hil2 [l J$uggest the approximation 1, ;;, 2.,\ j ,,.
402 Classical Design by Series Compensation Oesign Specifications Revisited 403

These approximate expressions for I ,, Id and t, are independent of the gain K G(;w)-plane
for the second-order system described in (7.1) and Figure 7.1, since fwn = p 1/2.
However, for the third-order system described in (7.9) the real parts of the
}2,-
dorninant complex closed-loop pole pair are functions of K and, hence, these
values of I,, Id, and 1, depend on K. M = 0.701
Suppose design specifications are given as
t, .$ 1,....

Id.$ Id...,.
- 2 - JI fHv}Pw,i 1 2 1 3
I, .$ 1,.... (7.20) 1
1
1
Substituting (7.18) and (7.19) into (7.20) and rearranging yields 1

,rl
1
- ln9
-rw n
<--
- t r.nu.
1

1 - j2
1
1
-fw< - ln2 1
n - --
Id ..... 1

Figure 7.5 Using the constant-M circle for M = 0.707 to

- fwn .$ - In (1/8,) (7.21) determine system bandwidth.
Is.,...
the most negative of the right-hand sides of the inequalities in (7.21). Using this
Figure 7.4 shows the acceptable (shaded) region in the s-plane corresponding to most restrictive condition guarantees that ali three design specifications are
satisfied simultaneously.
A design requirement for the closed-loop system bandwidth B W may be
s-plane expressed as

BWmin .$ BW .$ BW.,,ax {7.22)

where the lower bound BWmin ensures that the system bandwidth is large enough
to enable the system to respond to ali command inputs of interest and the upper
bound BWmax is selected to prohibit excessive amounts of high-frequcncy noise
from interfering with proper system operations. Figure 7.5 shows the detcrmina-
1 tion of system bandwidth for closed-loop systems having unity gain at w = O by
1
1 using the constant-M circle for M = O. 707 in the G (jw )-plane.
1
1
1
1 EXAMPLE 7.3
j - In 2
1~-
1 ldmax For the second-order system of Example 7.1 having the G(s) given in (7.1) with
p 1 = 2, determine the ranges of K satisfying each of the design specifications
-In (116,) listed in Table 7.1.
1rmax 18,nax Of the nine requirements to be satisfied, ranges of K for the percent ovcrshoot
and steady-state error criteria were deterrnined in Example 7.1. The three remain-
Figure 7.4 The s--plane region satisfying t,. td, and t5 '
design specifications simultaneously. ing time response specifications, that is, rise time, delay time, and settling time,
 404 Classical Design by Series Compensation 1 Design Speclficntlow, lll)vl(llhJ(.I "u

are independent of K but are satisfied Cor the given sysccru, :1111111!1118 1,i !I
design formulas in (7.21). Design formulas Cor closed-loop i,y~ll-111 p 11 111 '>\"
tude, phase margin, and system bandwidth can be obtaincd hy 1111111111 ti,,
o algebraic inverses of the corresponding analysis equations in (6 9), (<, 1 1) ,,11,J
~ ~ ~ ; ..r (6.7), respectively. As described in Example 7.1, the design formula 111 (,,,,., 111
~ ~ ~ lO ,-.. terms of %0Smu was derived from (4.25) as (4.32). The gain mnrAin is 111111111,: fi,,
:: VI VI VI ~ ;: this example, since there are no jw-uis crossings for K > O h ,., 11111 11"'' ,t,I~ li
~ ;!. .-1 .-1 ....1 ;: ~ ~ ! : satisfy ali design requirements of Table 7.1 simultaneously hy acl111\lr11i, 11111\ lh!
g>
l1l
<"i
VI 1 1 1 ':;
,v
,-.VI ~1 VI VI gain K.d We emphasize again that the design formulas
. in Tahle 7 1 ,111
a: ~ N N N ~ ~ ~ ~ ~ for un erdamped second-order systems descnbed by (7.1).
1
'c:i: '
c:i: '
c:i:
~ I
1 1 1
I 1 1 : EXAMPLE 7.4

; ( ~ Results from Example 7:2 pr~vid~ a rang~ O .s K .s 251.1 for sac_i..,fy1ng 1111: ''."'''
,r \\...J ~, ei frequency _re~ponse specificallons m ~7.8) s1multaneously. Deter~m~ clw (p11~~1hl\
~ V\ _.. s: more restnct1ve) range of K that sat1sfies not only these three cnceria h111 lils,-, th,
-
1
M ! ~ first four time response specifications listed in Table 7.1. What " 1111
,...: ,-.. : ~ ~ n "ti bandwidth for the minimum and maximum vals of K in this rangc'l
\) -a ~ ,. ._ : _J ~ .,.~ ~ ..r + Figure 7.6 shows the approximate s-plane region for which che lime 1r p1111~,
'"'- ~ ~:, cE_ : g : ): ~, [ '-~
'-8 e " specifications are satisfied. This approxirnation is based on using thc 1111111111 ,111
V'\
'-' J1 E~ ~
1:
* -' E
-t>
e
~
., Q.
'- -
'
.... 1
..-..
E +
-
.,. Et complex poles to yield an approximate second-order system. The intcrs!'c 111111 111
] &;: '#. ~ :::,. ::S, 1 .... ~ ~E 1 ~ the wedge-shaped region from Figure 7.1 with the most restriccive of clw 111111
~ ~ e3 "g ~ 51 51 5 :::,- .... , ~ -~ N semi-infinite regions for t,, td, and t., provides the allowable region ro, 111ect11 11'
o -~ II .... 'b VI
I q 1 ..-+ ~ '----;::."" ali four requirements simultaneously. In this case we see from TRhle 7. 1 1h.11 1111.
VI VI .- O J' . . h . 1 (1 ~ 0 . .
411
GI
e,,
<I>
o -...E -e
" .._, -
+ 3e 3e 3e .:, N
es e '(!) e :
EE
::,,,- sett mg tune reqmrement, avmg - n /o,)/t, ..... = - .75, IS the mosl ll'.'-1111'.
e
111
~ "'
t,
.,_,, .,_,, .,_,,
1 1 1
, 1N
.-">
:::::
N
e *
g ~ :::..
IX)
tive

..l a: 1 1 1 1 ; 1 1 1 1 1 Toe root locus for positive K is sketched in Figure 7.6. We next determ11H' th.11

~
-g <:: N N N : <:: <:: ~ es : e " range of K Cor which ali three closed-loop potes are inside the shaded (acceptahlc)

~
: J ':: ;: ':: <1> J .j l ~e e! s-plane region. If we llssume (and later verify) thal the root locus emerges fro111
~ ~ ~ ~ ,: ~ ', Che shaded region through the %OS boundary, chen we can easily show from thl
5 1 anglc requirement that 0 1 + 02 + 03 ce 180 or 02 + 03 = 180 - 01 = O. Tak1111~r
1 1
1
, the tangent of 02 + 83 , we have

e 1 tan 2 + Can o3
+ 03 ) = - -o

..
/l; o o o u111 VI o
~-- 1 ,i
.Q> 6 5l 5l ~ ... ::,,, 5l i tan{02 ----
;
o
e .::
Q>~
~
....
q
(")
o
..-
q -j ~
'<tlO o
U"!
..-
o
'<t '<t
ij'i '-
u
. 1 - tan 02 Can 03

~ ~ ~~

1
..-. o~'Ea.
,._
G>
v1
en
0
v1 .J>
-' vi
VI -":>"'
.....,
VI-:: o.
e E
VI
~~
< <
Q..
<
(!)

..- ,,
:
:, --
/J + -. p--
W en ":! '-. , 2- a 40 - a
..J
al
~
" * '
-;:;
(ll.. , . ,-:
:
;
":
1
=
1-
(
--
/J ) ( fJ
---
) = tan O
1- ' ' 2-a 40 - a
1
i:11 , where fl = a tan O = a tan [cos 1 <fm,n)J, and f nun is determined as 0.517 from thJ
%OS design formula in Table 7.1. Solving (7.23) yields a = 0.96 and fl 1 ~X
Since - a = - 0.96 is less than - 0.75 from the settling time design specifka1m11 1 ~
 406 Classical Design by Series Compensation Design Specifications Revisited 407

s-plane

G(jw)-plane

o
.,o 2
-1
"' .. 2
O
w = 1
1
w = 08
1
w = 0.6
\ 1
w = 0.5
'k -jl
w = r-4
M = 0.707

w = 0.3

K = 136.9 K = 36.8
(BW :;: 2.31) (BW ::; 0.59)
~
-12

Figure 7.6 The 9-plane regon satisfying %OS, t,, td, and t.
design specificatiOf!S simultaneously. Figure 7.7 Using the M - 0.707 circle to determine system bandwidth fo r
Example 7.4.
wc havc shown that the assumed emergence of the root locus tluough thc %OS
boundary is indecd corrcct. Using s-plane vectors, we compute K,nu as method to solve for accurate bandwidth values for each K from

1
K max = /a 2 + /J 2 V(2 - a}i + p i {(40 - a} 2 + p i (7.24) G(Jw) 11 K
2
= fi, (1 .26)
11 + G(jw) ... -w = V(K - 42wi)2 + w2 (80 - w2 ) w-BW

which yields K m:u = 136.9. Moreover, we determine K on by s-plane vectors

drawn to the point aJong the real axis where the shaded region is entered by the The corresponding values of BW for Kaun = 36.8 and Km.u= 136.9 are ap-
locus starting on the open-loop pole al the origin ( - 0.75 + JO). Therefore, proximately 0.59 radians/second and 2.31 radians/second, respectively.

Kon = (0.75)(2 - 0 .75)(40 - 0.75) = 36.8 (7.25)

7 .1.4 Summary and Deslgn Prevlew
and the range of K that satisfies ali seven criteria under consideration is
l6 8 .:; K ~ 136.9. The design formulas listed in Table 7.1 are valid Cor underdampcd systcm:,
(O < f < 1) or for higher-order systems having dominant closed-loop polc pairs
hgure 7.7 shows the constant-M circle for M = 0.707 in the G(jw )-plane with
polar plots given for Km,n = 36.8 and Kmax = 136.9. We may use Newton's that can be approximated by underdamped second-order systems, as demon-
strated in Example 7.4. Moreover, the acceptable s-plane region o Figure 7.6 for
 408 Classical Design by Series Compensation The Series Compensation Approach 41111

Y(sl where Z 1(s) is the complex impedance of the resistor-capacitor parallel comlm
tion, that is, ~
Series Plan!
compensator 1
z,(s) = -y-
-+ se
R1
Figure 7.8 The configuration for series com-
pensation .
By substituting (7.29) into (7.28), using Vo(s) = R 2 I(s), and rearranging,
closed-loop pole locations is also based on a second-order system approximation. obtain the voltage transfer function Gc(s) as
In contrast, we should recognize that design in, the G(jw )-plane is not restricted
to second-order systems but is valid for general higher~order systems. l
s+-
The purpose of this section has been to demonstrate conflicting requirements V0(s) R 1C
that can emerge when a single parameter is to be adjusted to satisfy severa! design Gc(s) = V(s) = R + R2
specifications simultaneously. For example, low values of the forward loop gain K , s+---
are often required to satisfy design spccifications on percent overshoot, rise time, R1R2C
delay time, settling time, system (frequency response) peak magnitude, phase
s+z
margin, and gain margin. The system bandwidth often requires a band of
relatively Jow values of K. In contrast, a design specification on the steady-state s +p .~
error due to a unit ramp input requires large values of K. We conclude'that the
adjustment of K alone is not ordinarily satisfactory to meet ali of these conflict- Jn (7.30), we have defined z and p as
ing design requirements simultaneously. (
1
z=--
R1C

.... ,.. 7.2 THE SERIES COMPENSATION APPROACH R1 + R2

p= (I
Thc traditional method of using series compensation in closed-loop systems to R 1R 2C
satisfy several design rcquirements simultaneously is presented in this section.
Figure 7.8 shows the configuration for series compensation. The form of series Observe that the ratio a= p/z can be detcrmined as
compensator to be considered is
1+ 'TS
a= p/z = (R 1 +R 2 )/R 2 (7 12)

Gc(s) = 1 + 'T2S (7 .27)

Figure 7,9b shows the zero and pole locations of the phase-lead network al , {:
and - p, respectively. Since a > 1, the relative positions are as indicatc<l.
which has unity gain for low frequencies. The design problem is first to select -t,,,,.We may also express (7.30) in the standard form for use in construct111~ Hnd
proper values of 7'1 and 7'2 and then to specify the elements of the physical device diagrams with s = jw as
itsclf. We describe configurations of electrical networks having these desired
propertics in this section and present specific design procedures in the following
sections.
w _ V0 (jw) _ (2-) 1 + jw/(l/aT)
Gc(J ) - V,(jw) - a 1 + jw/(1/T)
where
7.2.1 Phase-Lead Networks
1 R 1 R 2C
CoJsider the phase-lead network of Figure 7.9ll. The current /(s) resulting from ' T=-= - - -
thc applied voltage V;(s) can be expressed as p R1 + R2

V,(s) . In othcr words, the zcro is locatcd al z 1/ aT and thc polc; al p

l(s) = ( ) (7.28) f.i~urc 7.9, shows Dodc magni111dc and pha!-c diagrams for (;.( f:J)
R 2 + 7: 1 s
 410 Classical Design by Series Compensation The Series Compensation Approach 411

R The maximum value of 1/) occurs at the geomctric mean of the two comer
frequencies 1/T and l / aT. This maximum phase 4>( w,,,) = 4>m is

+ e
~
+ s-plane 4>,..= tan- (aT T~ )-
1
tan -
1
(r T~} '
/ (s)
V,(s) R2 V0(s) = tan - 1 (/a)- tan- 1 (1{;) (7.38)
-p -,
Taking the tangent of 4>,,. in (7.38) and using the two-angle tangent formula, we
have
(u) (b)
ra - i; ra
1
tanipm = 1 + {;(1 {a)
T
t
;~....
o
o
w
- = ~(ra
2
- 11/a ) = ira
a- 1

In Figure 7.9c, as the ratio a between tbe two corner frequcncies approadH:s
(7 .39)

'l 20 log10Wu) unity, the two corners coincide, the compensator pole and zero cancel, and the
0
N - amount of phase lead is reduced to zero. As a becomes very large, the cornees
separate by a large amount and 4>m approaches 90. As a rule of thumb, we
1' normally try to limit a to the range 1 < a < 15, since values greater than
approximately 15 tend to pemt unwanted noise .disturbances to entt:r and
influence the system dynamics adversely. The precise lioting value of a aJlwed
t 90"
depends on the particular application.
~
11
4'm Using the identity sin 'Pm = tan 4>ml + .;.-02'Pm) , we may form
/Ci
a- l

w-
: sin 4> = - - (7 .40)
m a+1
() 1 w,..
a1' 7' Solving (7.40) for a gives the design equation
(e)
1 + sin4>,,,
a= (7.41 )
Figure 7.9 Phase-lead series compensation (a) network, (b) zero and pole 1 - sin 4>,,,
s-plane locations, and ( e) Bode magnitude and phase diagrams.
Equation (7.41), rather than (7.39) or (7.40), will be useful in the design proce-
We necd to solve for the maximum value of the phase curve as a function of a dure, since we will want to determine the value of a Cor which a given additional
and T. Denoting the phase of G, (jw) as 1/), we write phase lead ( 4>m) is obtained. The remaining problem of dcciding whcre to place
the two corners of the compensator will be treated in subsequent sections.
tp = / G,(Jw) = tan - 1 (aTw) - tan - 1 (Tw) (7.35)
Thus, setting the derivative of 4> with respect to w equal to zero, we have 7.2.2 PhasJ L~g Networks
d4> 1 aT T The phase-lag network of Figure 7.10a has the voltage transfer function givc:n by
(7.36)
dw ... - ... = 1 + (aTwm}2 - 1 + (Twm}2 = O
m
1
Vo ()
s
R + -sC
where w,.. is the frequency at which 4> achieves its maximum. Solving (7.36) yields i
Gc(s) = V,(s) = 1 (7.42)
1 R1 + R2 + -
wm = rra (7 .37) sC

~
412 Classical Design by Series Compensation
Phase-Lead Design by Root l ocu i 111 ;1
R1
(with s = jw) as
+ l (s)- 1 +
s-plane
s + z)
G(s)=a--
(
R2 c s +p
V(s)

el V0 (sl
-z - p
.
Gc(Jw) =
1 + jw/ (1 / aT)
. .
o
(n)
T (b) The s-plane zero and pole locations are shown in Figure 7. IOI>
diagrams in Figure 7.10c.
1

t
1
<.,"
o
T

1
aT
w- 7.3 PHASE-LEAD DESIGN BY ROOT LOCUS
o
ii
E
o
N
----- - -
- 20 1og,oa ---- - ------- --.............-- - We describe in this section how to design a phase-lead series C'omprn,:11111 11\i
using the root locus technique (1-6). Initially, we convert thc design rtq111111111111
into an allowable s-plane region and then verify that the simple adju1-111u111 111 ih,
gain K will not permit us to satisfy the design requirements. Wc ntxc ., p111 fv 1111
desired dominant closed-loop pole locations and attempt to pfau 1lw h 111
network zero and pole such that the root locus passes through lhr,1 d1,1a, .i
points. S<rlecting the location of the network zero automatically fi xcs 1h1 el,''""

t
o .,_ result, since the lead network pote must subsequently be placed to y1d d 11 rn 1'
angle of 180 for the open-loop compensated plant Gc(s)Gp(s) at lhe cl1,111d
location of the closed-loop pote. The gain K of the compcnsaccd pl:1111 "

~"
.... calculated from s-plane vectors, and the velocity error constant K 0 , for exn111plt .
- 45
is then determined. Suppose we are attempting to guarantee that thc stcady ~' ""
3 error for a unity-feedback Type 1 system resulting from a unit ramp inp111 d <1t,
~ -90 not exceed sorne specified value. We can compute the steady-state error as tlw
(e) reciproca) of K v If the resulting ess from the compensator poie ancl ,n 11
placemenl is too large, we establish new desired dominanl closed-loop polcs and
Figure 7.10 Phase-lag series compensation (a) network, (b) zero and pole repeat the above design procedure.
s-plane locations, and ( e) Bode magnitude and phase diagrams. Figure 7.11 shows the steps of the recommcnded design algorithm. In Step 4,
the compensator zero is placed somewhere to the left of lhc second open-loop
lf we define z, p, a, and Tas pole. If this zero were placed precisely on this pole, we would have cancclla1 11u1
compensation. Dorf (2) recommends placng this zero immediately hclow tl11 /
z = l/R 2C desired closed-loop pole location. While this choice makes the resulting co,1111111.1 1
tions slightly easier, we suggest a zero location somewhere to the Ieft of thi~ 11111111
to improve the assumption that the desred quadratic poles tend to do mina1t 11 ..
p = l/ (R 1 + R 2 }C
closed-loop response. We show the limits of this approximation in thc foll11,101g{
examples.
a= Ri/(R 1 + R 2 )
I
EXAMPLE 7.5
T = (R 1 + R 2 )C (7.43)

Consider the postional servomechanism with the field-controlled de mol<11 , h, 111 11 .

then (7.42) can be rcarranged in standard forms for root locus and Bode diagrams
n Figure 7.12. We showed in Section 2.1 that the plant transfe~ functio11 rnn l,11 t
 Phase-Lead Design by Root Locus 415
414 Classical Design by Series Compensation
Rr
Start
+ +
) , \llmnN 1

Convert des,gn recuirements

into allowable s-plane region
Step 1
vp K1,
v1 r) Lr
I

81.--
JL
/
./
/
/
Yes
Step 2 (Je
11,

n - turn 11-lurn

Modify allowable
s-plane reg,on V
Select des1red domrnant
closed-loop pole locations
Step 3
t; li,11
Select compensator Figure 7.12 Positional servomechanism with field-controlled de motor.
zero placement Slep4

Mod1fy desired
expressed in the form
dorn,nant closed-loop
pole locations Calculate compensator
Slep 5
pole loca\ion Kv (7.45)
Gp(s) = s(l + 'TmS )(1 + -r,s)
Compute ga1n K by
, Step 6
lill
s-plane veclors No phase-lead
cornpensat ion where K" is the Bode gain corresponding to the velocity error constan! for this
is needed Type 1 system. The mechanical time constant -rm is usually severa! times larga
than the electrical time constan! -r., which places the comer frequency al 1/r. at a
No higher frequency than the one al 1/-rm. The parameters in (7.45) are given by
Step 7
-r = J /B and -r, = LIR1. The gain Kv is equivalent to KpKAKr(N1/N2 )BR 1,
111
where these parameters can be identified from the transfer function of Figure
2.12d.
Suppose we consider the case where -rm = 0.5 and -r. = O. We have assumc:d
Step 8 that the inductance of the motor field circuit ( L1 ) is negligible, resulting in a
second-order plant described by

Ku (7.46)
Select network
parameters Step 9 Gp(s) = s(l + 0.5s)

End
which is equivalent to

K
(7.47)
Figure 7.11 Flowchart of an algorithm for the root locus design of phase-lead
compensation networks.
.,. Gp(s) = s(s + 2)

where K = Kj-r,., = Kj0.5 = 2Ku.

lt is required to design a phase-lead network for a negative unity-feedback
system having the plant transfer function in (7.46) or (7.47) subjecl lo the design
416 Classlcal Design by Series Compensatlon Phase-Lead Design by Root Locus

K = 40 $-planc

., .plane

- 1
/
/ -p

" t = (-p - PI + z)/ 2

K = 40

Figure 7.13 Allowable s-plane region to satisfy

%OS .s 15 and t, .s 1.0 second.

Figure 7.14 Root locus plot for a typical phase-lead network

spccificatio ns given by pole / zero ptacement.
/%OS~ 15%
In Step 3, we arbitrarily select the closed-loop polc Iocations for the phase-lcad
I, ~ 1.0 second compensated system at - a J/3. We choose a = 3 arbitrarily and calculate /1 a<, (
a tan O, where O = cos - 1 (!min) from (7.2), and !min is determined from the %OS l1
/ es,I unit ~ O.05 radians {7.48) desi n formula in Table 7. l. Altematively, we can show that /l cqual,
ramp
a l - !!in/ fmin Both methods yield /3 = 4.97. Thc points - 3 j 4.97 lie on lht t
Figure 7.13 shows the allowable rcgion in the s-planc (Step 1) to satisfy the boundary of the allowable s-plane region determined in Step 1. I
%OS and , , design requirements in (7.48). We have superimposed the wcdge- In Step 4, we place the zero somewhere to the left of the second pole. A !\pu .ti
shaped region shown in Figure 7.1 for a %OS ~ 15% with .the s-plane region lying phase-lead network zero placement is shown in Figure 7.14. Once thc 1r111
o n or to the left of the vertical Jine o = - In 9/ t, = -(In 9)/ 1 = - 2.2, which location has been arbitrarily specified, we can compute thc angle Or from '
was identified for the genera] case in Figure 7.4.mSince K" must be equal to or
grcater than 20 to satisfy the steady-state error requirement (Ku ~ 1/e,.,,,,.. = r../ZerosoCGp,-4..s) - E/PolesoCG,GP(s ) = (2k + 1)1 80
1/0.05 = 20), we require K ~ 40. Figure 7 .13 shows that the closed-loop pole
locati&ns fo r the uncompensated system with K = 40 le outside the s-plane
O, - [OP + 01 + 02 ] = (2k + 1)1 80 (/ 4'))
region for which %OS .s; 15% and t, < 1.0 second (Step 2). Therefore, it is
evident that compensation is required to meet ali thrce design specifications
simultaneously. where k is z.ero or any positive or negative integer. Values of O,, 8 1 , and O, t,11 1 h,: f
418 Classical Oesign by Series Compensation Phase-Lead Design by Root Locus 419

dctermined as open-loop pole, the second places the zero at - a = - 3 as suggested by Dorf in
(2), and the third puts the zero arbitrarily at -6. We may use (7.51) and (7.52) 10
ol = tan - 1 (-/3 ) show that the zero can be placed no farther to the left tha.n - 7.977 lwithout
z-a
violating the rule-of-thumb restriction that the ratio a = p/z between pole and
zero Jocations must not exceed 15. At z = 7.977, (7.51) and (7.52) yield p = 119.6,
01 = 180 - tan - 1 (~) ~~gi~ .
K(s + :,.977)
GG (s) = - - - -- - (7.53)
82 = 180 - tan - 1 (-/ -3 -)
P1
(7 .50) e P s(s + 2)(s + 119.6)
Evaluating K at s = - 3 j4.97 by s-plane vectors (Step 6) gives K = 488.0. The
where the desired dominant closed-loop poles are at - a j/3 with a = 3 and value of K., is Kz/p 1 p = K/2a = 488.0/30.0 = 16.3, S!nce GPp(s ) may be
/J = 4.97, and the second open-loop pole is at - p 1 with p 1 = 2. By using (7.50) expressed in standard Bode form as
in (7.49) with k = -1, we find that 488.0(7 .977) (l + s/7 _977)
2(119.6)
op = ol - 01 - 02 - (2k + 1)180
G,GP{s) = s(l + s/2)(1 + s/119.6)
13 16.3(1 + s/7 .977)
= tan - 1 ( - -)- 121.1 - 101.4 + 180 (7.54)
z-a
s(1 + s/ 2)(1 + s/ 119.6)
4.97 )
= tan - 1 ( - - - 42 .5
" (7 .51) where a = p/ z = 119.6/7.977 = 15.0. Therefore, thc steady-state error for a unit
z- 3 ramp input is 1/K., = 1/16.3 = 0.061 radians, which is too large (Step 7) to meet
the third design specification in (7.48).
The compensator pole location at s = - p may then be determined in Step 5 by Since e,, is not small enough in Step 7 of the algorithm (Figure 7.11), we enter
w,ing the feedback loop in the flowchart, which modifies the desired dominan! closed-
/3 loop pole locations. We next place the desired dominant closed-loop poles at
p =--+ a - a j/3 = - 4 j6.62, which is again on the boundary of the acceptable %OS
tanOP
and t, region in the s-plane. The resulting equations corresponding to (7.51) and
(7.52) for OP and p are
_ 4.97
- tan O + 3 (7 .52)
6.62 )
p O
P
= tan - 1 ( - -
z- 4
- 47.9

Table 7.2 shows the results obtained by using the phase-lead algorithm of 6.62
Figure 7.11 . The first try places the zero at -p 1 = - 2 to cancel the second p=--+4 (7 .55)
tanOP
Selecting z arbitrarily as 6.5 yields p = 20.9, a = 3.2, and K = 137.3. For this
T ABLE 7.2 Pole / Zero Placement Attempts for Example 7.5
choice, G,GpCs) becomes
Do1111nant Poles z 8P p a K K, Bss 137.3(s + 6.5)
3 1 / 4.9/ 2.0 58.9 6.0 3.0 33.7 5.6 0.178 GpP( s ) = s(s + 2}(s + 20.9)
3.0 47.5 7.6 2.5 39.9 7.9 0.126
6.0 16.4 19.9 3.3 89.3 13.5 0.074 21.3(1 + s/6.5)
(7 .56)
7.977 2.4 119.6 15.0 488.0 16.3 0.061 s(l + s/ 2)(1 + s/20.9)
4 1 /6.62 6.5 21.4 20.9 3.2 137.3 21.3 0.047
9.42 2.8 141.2 15.0 858.9 28.6 0.035 which yields e.. = 1/K., = 1/21.3 = 0.047 radians, and the steady-state error
design specification in (7.48) is satisfied. An additional calculation in Table 7.2 for
 420 Classical Design by Series Compensation Phase-Lead Oesign by Root Loc;w, 11:, 1
140.----.....---,....-~....-~....-~....-~....-~....-~....-~~
The remaining step is to determine parameters for the phasc-lead comprn, 11 i,,11
l .~ network (Step 9) in terms of z and p. Since these are three network p:11a111dnr,
1.00
(R 1, R 2 , and C) and only two design pararneters (z and p), the resultini ~0111111111
is not unique. Often the capacitance C is selected arbitrarily Cor convcn1l'1111 111
t 0.80
physical realization. lf we select C as 1 F for this e,cample, then (7.31) and ( / ll)
Redesign usong 4 % overshoot in Step 1
may be used Cor the choice in (7.57) to yield R 1 = 172.4 Kl and R 2 12. l K U
We must also adjust the value of K to be 600, which includes the garn fat 1111
a -= 15.0. Figure 7.15b shows the resulting phase-lead network ancl thc add,11011111
O 20 gain (a,. 15.0) needed to ensure Kv~ 20. In a positional servomecha111sm, M11 h
o.oof 1
O 40 O 60 0 .80 1 00 1.20 1.40 1.60 1 80 2.00
as in Figure 7.12, this network is usually inserted at the amplifier input and 1111
amplifier gain K,. is increased by a factor a.
Time Csccl--+-
(a)

EXAMPLE 7.6
R1 = 172.4 KH

Consider again the system of Figure 7.12 but with 'Tm = 0.5 and T, = 0.02S. \~ hirh
+ yields a third-order plant described by
e.e~,
~
40 1
V,(s)
C= l F / ~ 1( Gaino
= 15.0)
1 aV0 (s) wfs + 2) K ,
G ( ) IJ
Rz = 12.3 KO ' s = s(l + 0.5s )(1 + 0.025.r)
or equivalently

K
(I>) G,(s) = s(s + 2)(s + 40)
Figure 7.15 Phase-lead design results for Example 7.5.
where K = K.,/ 'T,,,'T, = 80Kv. The problem is to design a phase-lead network to
thc modificc.l dcsircd closed-loop pote locations shows that the zero can be placed satisfy the design specifications in (7.48).
no farther to the left than - 9.42 for a ~ 15. Figure 7.16a shows a root locus sketch for the uncompensated system with thl'
In Stcp 8, we examine the time response of the pbase-lead compensated system allowable s-plane region for %OS and t, requirements (Step 1). Wc rcqu,rc
to determine whether it does indeed meet the required performance specifications. K ~ 1600 to meet the steady-state error requirement, since K,, must be greatcr
We observe that the actual %OS may be somewhat greater than 15% because of than or equal to 20, as in Example 7.5. Obviously, the %OS and t, requiremcnts
the dominant closed-loop pole approximation in Step 3 and the presence of a zero cannot be satisfied by adjusting only the gain K (Step 2), and therefore a
in the compcnsator-plant transfer function. Computer simulation results in Figure compensation network is needed. Observing the similarity with Example 7 .S. wc
7. 15a show that the design in (7.56) yields a 28% overshoot. Redesigns using 10% select the desired dominant closed-loop pole locations for the compcnsate<I
and 4% in Step 1 provide overshoots of 19% and 15%, respectively. Thus, a system as - 4 j6.62 (Step 3) and select the compensator zero at - z = - 6 ~I
suhstantial safety margin, that is, the use of 4% (with a = 5.0) rather than 15% in (Step 4). Figure 7.16b shows a sketch o the resulting root locus plot. As in (7 51)
Step l , was needed in this example to obtain a design that meets the given we determine ()P from
,;pccifications. The resulting G,Gp(s) becomes
()p = ()z - () l - fJ2 - fJ3 - (2k + 1)180
600(s + 5.8) ~

G,GP(s) = s(s + 2)(s + 86.8)

= tan - ( z6.62
_ )1 - 121.1 - 106.8 - 10.4 + 180
20.0(1 + s/5.8) 4 z-6.S
{7.57)
s(l + s/2)(1 + s/86.8)
6.62 )
for whic:h " , /z 1S.O. = tan - i
( 6 _5 _ 4 - 58.4 = 1 J .O
422 Classical Design by Series Compensation Phase-Lead Compensation Using Bode Diagrams 423

Therefore, in Step 5 the compensator pote at - p is determined from

s-plane

6~ .
p= ~ - +4=~2 I (7.61)
~~

Using s-plane vectors in Step 6, we determine~ for the phase-lead compensatc:d

system as K = 9636.1. Moreover, we compute Kv as K/(80a) =
9636.l/[80(38.3)/6.5) = 20.5 and e.. in Step 7 as 1/Kv = 0.049 radians, which
meets the design specification in (7.48). In Step 8, we find from compuler
-0.5 - j6.2 simulations that the actual %OS is 30% for this design attempt. Therefore, we
must impose more stringent design requiremenls and return to Step 1 to begin the
redesign procedure. The succeeding steps are identical to those examined here and
in Example 7.5 and will not be repeated. Once an acceptable design is obtained,
- jl6 R 1 and R 2 can be calculated in Step 9 and the phase-lead network inserted for
series compensation.
(a)

7.4 PHASE-LEAD COMPENSATION USING BODE DIAGRAMS

s plane
Design by using Bode diagrams allows us to satisfy frequency response specif:a-
tions directly without converting them to desired closed-1oop pole locations in the
s-plane (1-6). In fact , the designer's choice between root locus and Bode design
procedures can usually be based on which design specifications are given. Pri-
marily, we examine phase margin or peak magnitude requirements in this section
and show sorne correspondence between the resulting polcs and zeros of phase-lc:ad
networks designed here and those obtained in Section 7.3.
Consider the polar plot in the G(jw)-plane of Figure 7.17. The plot for the
uncompensated system lies too near the -1 + jO point as measured by either

G(jwJ-plane

- 1 + jO

I
I
Uncompensaled Cornpensa ted
/
/
/
/

- ---- --
/
(b) /
/
/
Figure 7.16 Aoot locus plots for the (a) uncompensated and (b)
phase-lead compensated systems of Example 7.6.
Figure 7.17 A typical polar plot for uncornpensated
and phase-Jead compensated systems.

li
 424 Classical Design by Series Compensation Phase-Lead Compensation Using Bode Diagram'l 111.:,

t ____ -----------= - - - - Start

1 20fog()ll

io Ster, l
ff

'~ aT
w,,, 1
T
... - Determine PM l or the
oncompensated system
and setect <I>,,.
Strp l

Yes Sir , .l

90 S1rp ~
Determme a
~

1 Determine "'m and T Slrp ',

1
No phase-lead
O

"'-
com i:i.nsation
1 w,,, 1 ,s needed
- No
aT r ),ltp , ,

Figure 7.18 Bode diagrams for the phase-lead series compensation

network with gain a included.
Selecl nP.twork 511'p I
pararneters

phasc margin (PM) or peak magnitude (Mp) specifications. Placing a phase-lead

series compcnsation network in the forward path can often give acceptable values End
for thesc design specifications. We need to sclect the pote and zero o the
pha~e-lead network such that the maximum phase lead occurs at the new gain Figure 7.19 Flowchart of a Bode phase-lead design procedure.
crossover frequency, that is, at the frequency where the phase margin is measured
for the compensated system.
margin. Thjs permits us to decide just how much additional phasc lcad wc nccd 1t,
We set the gain K Cor the uncompensated system to giv.e the desired steady-state
meet the design specification. If the design specification is met with no additic111al
error, for example, for a unit ramp input. To maintain the low-frequency gain
phase being needed, then Step 3 tcrminates the design algorithm. If (j>nr > O. wt
when the series compensation network is added, we increase the arnplifier gain by
a factor of a, which is the ratio between the network comer frequencies. The proceed to design the phase-lead network. In practice we add approximatcly 5 t< i
10 (or even more in sorne cases) to this preliminary value of </>,.,. This addi11011:1I
rcsulting Bode diagrams are shown in Figure 7.18.
amount is necessary because the phase curve usually is decreasing somcwhat 1,t
Figure 7.19 shows a flowchart of the phase-lead design procedure using Bode
the gain crossover frequency for the uncompensated system. We will find thnt 1111 ,
diagrams. lf we are designing a phase-lead network for a Type 1 system subject to
new gain crossover frequency les to the right of the gain crossover frc4m11 v f,,,
stcady-state error and phase margin specifications, then in Step 1 we compute
thc uncompensated system, as shown in Figure 7.20. Thus, the phasc 111:111.111 ,11
the new gain crossover frequency would be smaller than previously i111lw1,t , d
1
unlcss we include additional phase to offset this decrease in phasc. A ,, 1111,
Kv... = l/e,,.... I unil (7 .<i2)
negative slope usually rtquires only 5 to 10 additional phasc. On lhl' 11tl1,,
ramp
hand, a sharply decreasing phase curve in the vicinity of the gain n11,,11, ,
I frequency for the uncompensated system requires a much greater ad<l,1 11111 nll
We adjust th,c forward path gain K to yield this vaJue of Kv and then attempt to phase for </>,.,. In fact, it may not be possible to add enough phasc for sy.~t1111i- ,, itli'
satisfy the given phase margin design requirement. In Step 2, we determine the extrcmely large phase slopes in this vicinity. In such cases, the phasc-laz d,,111 111
gain crossover frcquency for t-he uncompensated system and thc associated phase the following two sections is more appropriatc.
426 Classical Design by Series Compensation Phase-Lead Compensation Using Bode Diagrams 427

40 ~~~~~~~~~~~~~~~~.......-~~~ ..... We calculate the new phase margin for the compensated system and compare it
with the design requirement in Step 6. If a larger PMc is needed, w increase thc
4>m appropriately and repeat Steps 4 and 5 until the specification is mc:uFinally,
we determine the network parameters R 1, R 2 , and C in terms of the design
t 20 parameters a and T (Step 7) by using (7.32) and (7.34).
~
.:::,
e:>
o
j EXAMPLE 7.7
o
oll
0 .5
1
1
1
2
1
46
~ 1
20
-
1
N 1
- 10 1oa,oa
1
... ----
Cornpcnsated
lt is required to design a phase-lead compensation network for the system in
(7.46) of Example 7.5 such that the phase margin is greater than or equal 10 40
1
Uncompensated
!G(;wl l'',VII and the steady-state error due to a unit ramp input is less than or equal to 0.05
-20 IG(Jw) = GP(Jw)J radians.
1 1 We construct the Bode magnitude and phase diagrams, as shown in figure
7.20, for Ku = 20, since Ku = 1/ess .... = 1/0.05 = 20., The value of K in (7.47) is
2Ku = 40. Since this system is a second-order case, we may identify K as w!,
oI n"1 1 1 1 1 1 1 1 compute t as - p 1/2 wn = 2/(2/40) = 0.158, and then use the expression in
1 2 4 6 10 , 20
... ---- (6.12) to obtain PM Cor the uncompensated system as PM., = 18.0. Equation
t (6.10) gives the gain crossover frequency as 6.2 radians/second. Because of the
~ fentle negative slope al the gain crossover frequency, we add 5 as a safety
~ -90 "~ margin to the difference in the required 40 and the 18.0 jusi obtaincd. Resulcing
Compe11~ted
!G(Jwl = G,G(Jwll values of 4>m and a, obtained by using (7.41), are given by

4>,,. = 40 - 18.0 + 5 = 27
- 180'
1 + sin 27
a= = 2 7 (7 .64)
l - sin 27
Figure 7.20 Application of the design procedure in Figure 7.19 for
phase-lead design. We now locate the new gain crossover frequency al that point where the Bode
magnitude curve has a value of - 10 log10 a = - 10 log 10 (2.7) = - 4.3 decibds.
Stcp 4 solves for the value of a from (7.41) as We set 20log 10 IG/Jw) I = - 4.3 and obtain

1 + sin 4>m 40
(7 .65)
{7.41) c,Uw>I = = 10 - 43120 = 0 .613
a= l - sinq,m 2
w/w + 4
Ncxt, since the maximum phase being added occurs at wm, we wish lo place wm at Squaring (7 .65) and rearranging, we have
thc ncw gain crossover frequency. Figure 7.18 shows that the Bode magnitude
c urve has a value of 10 log10 a at "'m Thus, in Step 5 we locate that frequency on 2
w4 + 4w 2 - ( ~) =O (7.66)
thc Bode magnitude curve for the uncompensated system in Figure 7.20 where the 0.611
magnitude is - 10 log 10 a. By placing wm at this frequency, we will have used the
phase-lead network most effectively, that is, added the maximum phase at that which has the solution
frequency where it contributes most in satisfying the phase margin design ~
rcquirement. 2
40 )
Once w,,, is determined, we solve for T from (7.37) as w = \/ - 2 + ,, <2>2 + ( 0.613
1
T =-- (7.63) (7 .67)
wm/a = 8.0 radians/second
 Phase-Lead Compensation Using Bode Diagrame-, I\ZII
428 Classical Oesign by Series Compensation
4 0 ~ - - ~ - - - - . - - - ~ - - - - - . . . . - - - - . . . - - - - . ----T'~
Thus, we choose wm = 8.0 radians/second and compute T from (7.63) as
1 1
T = --,=: = r-.-:, = 0.076 second (7.68)
wmva 8.0v2.7
2r 1 1 ~ 1~
fro m which
1 1 '
t
- = - - = 13.1 radians/second ::,
T 0.076 <.:>
o 1 2 ,., _RO, ....
1 1.31 o

E.
~,
- = - - = 4.9 radians/second (7 .69) N
Comprn'-11"1
rc<., ...1 = ~;,r: ,1,,..,1
aT 2.7 1
-20
Thcrcforc. thc compcnsatcd system has the combined compensator-plant transfer
function given by
20( 1 + s/4.9) - 40
(7.70)
G,G/s) = s(l + s/2)(1 + s/13.1)

Finally, we compute the phase o Gpp(jw) at w = wm = 8.0 radians/second to 1 1 1 1 1 1

l_g 5 1 2 4 6 10 ,v v uv 1
ohtain -90" 1 1 1 1 1 I I 1
/ Gpe(j8.0) = -138.9 (7 .71)

from which the phase margin for the compensated system is

t
(7 .72)
~ - 180"
PM,. = 180 - 138.9 = 41.1
1 hus, we have shown that the dc.~ign specification o PM ~ 40 is satisfied by the
~ompensated system. -270
lf we arbitrarily selcct the capacitance C as 1 F, then from (7.32) and (7.34),
wc have
Figure 7.21 Bode magnitude and phase curves for Example 7.8.
1
aT (4.9)
R1 = - = = 204.1 Kl
e 10 6
A phase-lead compensation network is to be designed to meet the same critcria
R1 specified in Example 7.7, that is, PM ~ 40 and essl unit S 0.05 radians.
R2 = - - = 120.1 Kl (7.73) ramp
a- 1 We determine K,, = 20 as before to meet the steady-state error requirement anti
note that K = 80Ku = 1600. Using this value of gain, we construct the B0<k
Rccall once again that the arnplifier gain o 40 must be increased by a factor of
magnitude and phase diagrams in Figure 7.21. The gain crossovcr frcqu,n, v
a = 2.7 to yicld the rcquired transfer function in (7.70).
cannot be ound by using simple closed-form expressions in this third-ordc, r a"
lnstead, we must solve
EXAMPLE 7.8 1600

w/w 2 + (2)2 /w 2
+ (40)
2
=1 (l M)
Consider again the negative unity-feedback system of Example 7.6 that has a c.,- c..,G('

pla nt tfansfcr function given by Simplifying (7.74) yields

K 2
(7.59) !( wc) = w~c + 1604w~c + 6400wc - ( 1600) =O (7 /'i)
.Gp(s) = s(s + 2){s + 40)
430 Classical Oesign by Series Compensation Phase-Lag Compensation by Root Locus 431

which may be solved numerically in only a few iterations by using Newton's This exarnple on phase-lead compensation for a third-order system has demon-
method. The resulting value of wGc is 6.1 radians/second, which gives PM., = strated the difficulty in solving for the gain crossover frequencies Cor both
9.4. Therefore, since PM., < 40, a compensation network is needed to meet this uncompensated and compensated cases. Although formulas in ternlS of , and wn
design specification. were available for the second-order system of Example 7.7, no such simple
As a first attempt, we select a 5 safety margin to add to the difference in 40 closed-form expressions could be used for higher-order systems. Newton's method
and 9.4 as enabled us to solve iteratively for the unknown frequencies.
'Pm = 40 - 9.4 + 5 = 35.6 (7.76) This example has also shown that a safety roargin of only 5 may nol be large
enough and one or more additional trials may be needed. While a 10 safety
In Step 4 of Figure 7.19, the value of a from (7.41) is 3.8, and - 10 log 10 a = -5.8. margin was barely acceptable in the present example, much larger safety margins
We next solve for wm in Step 5 from might be required in sorne applications. A limiting value of 4>,,, is reached when
1600 the ratio a = p/z becomes 15.
IG(jw ) 1 = = 10-S.S/20 = 0.513 (7.77) In brief, two limitations of series compensation using phase-lead networks are
m w,,,,,/w! + (2) 2 Vw! + (40)' (1) the requirement of increasing the forward loop gain by a factor of a and (2)
the restriction that a = p/z must not be greater than 15. In the first cai,e,
which can be simplified and solved by Newton's method to yield w,,, = 8.6 boosting the arnplifier gain as required can sometimes result in a saturation
rndians/second. From (7.63), we determine 1/T = 16.8 radians/second, from condition for which linear analysis and d~sign procedures are invalid. In the
which l/aT = 4.4 radians/second. Thus, this first design attempt yields the second case, plants having sharply decreasing phase curves in the vicinity of the
combined compensator-plant transfer function gain crossover frequency tend to require larger safety margins. However, it is
20(1 + s/4.4) the total 4>m, rather than only that part due to a safety margin, which can result in
G/J,(s) = -s(_
l _+-s/-2-)(_l _+_s_/40-)(-1 + s/16.8) (7 .78) an excessive value of a. Therefore, for sorne systems we sce that it may not be
possible to satisfy the design specifications by using phase-lead networks.
whcre the arnplifier gain in (7.78) has been increased by a factor of a = 3.8.
Sclling s = jw in (7.78) and calculating the phase of G/J,(jw) Cor w = w,,, = 8.6
radians/second enables us to determine PM, = 36.8, which is too small to meet 7.5 PHASE-LAG COMPENSATION BY ROOT LOCUS
thc design specification as determined in Step 6 of the algorithm.
We next retum to Step 4 and arbitrarily add a safety margin of 10, instead o The phase-lag network of Figure 7.10a can be used for series compensation to
5 as in (7.76), to give 4>m = 40.6. The new value o( a is 4.7, - 10 log 10 a = -6.1, satisfy the sarne design specifications of many cases for which the phase-lead
and w"' = 9.1 radians/secdnd. We proceed to determine 1/T = 19.8 radians/sec- network was used in Sections 7.3 and 7.4 (1 - 6). There are selected cases for which
ond and 1/aT = 4.2 radians/second. lf the Corward loop gain K is increased by one network is better suited Cor series compensation than the other. At the end of
a factor of a= 4.1, we may express the resulting G/J,(s) as Scction 7.4, we indicated systems where thc phase-lead network is not applicable.
Moreover, we can find systems where the phase-lag network is not applicable, as
20(1 + s/4.2} wc will see in the following section.
G,.G ,(s) = s(l + s/2)(1 + s/40)(1 + s/19.8) (?.?9) The motivation for phase-lag compensation by a root locus procedure can be
described by first determining the forward loop gain K II for the uncompensated
which yields PM, = 40.2 as desired. Curves Cor the design in (7.79) are shown as system. This gain is calculated by using s-plane vectors drawn to the dominant
the compensated case in Figure 7.21. closed-loop potes that are selected to satisfy a %OS design specification. A typical
Compensation network pararneters Cor (7.79) are determine from (7.32) and
plant transfer function is given by
(7.34) as
aT (4.2)
1 G,(s) = K.,(s + z} (7.81)
Ri = -e = = 238 Kl s(s + p 1 )(s +Pi)
10 -6
Phase-lag places a pole-z.ero pair near the origin on the negative real axis 3t
R1 s = -z and at s = -p = .!az, where O< a< l. Using (7.44), we express the
R2 = -- = 64Kl (7.80)
a-1 combined plant-compensator transfer function as
where C has been selected arbitrarily as 1 .F. Ka(s + z}(s + z 1 )
Gp,(s) = ( )( )( ) (7 .82)
s s + p s + p I s + Pi ...
 432 Classical Oesign by Series Compensation Phase-Lag Compensation by Root Locur.

Wc than calculate thc value of Ka as approx.imately K,, from s-plane vectors Start
drawn to the dominant closed-loop potes. Therefore, we may select a as ap-
proximately K II divided by the desired K- that is, by that value of K nceded to
s:1tisfy the stcady-state error design requirement. However, we might expect the Convert design requ irements
into an allowable a-plane region
calculated Ka to be even larger than K ,,, since the s-plane vector from the
compensator pole is slightly longer th#an the one from the compensator zero. An
Determine K,, at the dominant 1oscd
offsetting efTect results from the root locus for the compensated system passing loop peles on the region boundary
slightly to the right of the intended dominant pole locations because of the
acld ition o f the compensator pole-zero pair. For this reason, we often determine a
hv
Yes
K,, ( 1 ) (7 .83)
a= Kdcsired 1 + SM
whcre SM is a safety margin of 10% or so, which decreases a to account for the
new locations of the dominant closed-loop poles of the compensated system. Calculate o as K0 d1v1ded by thc
Figure 7.22 shows a ft owchart of the step-by-step procedure for phasc-lag desired value of K and mod1fy to
include a safety marg_i~
ccm1pcnsation by thc root locus method. In Step 1, we convert the %OS dcsign
~pecification into an allowable s-plane region. We procced in Step 2 to construct ,.
thc root locus for the uncompensated systern and then to calculate K,, at Determine compensator le<O
dominan! closed-loop poles selected on the boundary of the allowable s-plane and pole locahons

rcgion . We use K,, in Step 3 to determine whether the steady-state error design
spccificalion is salisfied without requiring series compensation. lf it is not Determine nE"w dom,nant closed-loop pnle
,atisfied. we calculate a from (7.83) with SM = 10% = 0.10. In Stcp 5, we locahons al the reg,on boundary ami
calculate the resuttmg 1(
:1rhi1rarily locate the compensator zero on the real axis at a point that is 5% to No ph,1'" I;,~
l O'\~ of thc dislance from thc origin lo the second open-loop pole and place the comrcne.~11,on
1s ncrded
compcnsator pole at s = - az, where O < a < 1. We next determine new domi- Movc zero
to r,ght and
nan! closed-loop pote Jocations at thc boundary o the allowable s-plane region decrease o
hy c:onstructing the root locus Cor the compensated system. The value of Ka (and
hencc K) is thcn obtained from s-planc vectors drawn to thcse new closed-loop
roles. In Step 7, we calculate e,, and test to see whether it meets the design
!-pccification. If it does not, we decrease a [corresponding to a further increase in No
the safety margin SM in (7.83)} and retum to Step 5. lf e11 is small enough in
Stcp 7. wc next obtain the lime response curve by computer simulation and
determine whcthcr ali design rcquirements have been met. H the %OS is excessive,
w e <lccrease z, that is, move thc zero to the right, decrease a, and retum to Step 5
to calculatc thc new pole location. Once the design requirements are satisfied in
s,~""
Stl'p 8, wc complete the phase-Jag design with the calculation of network
parameters in Stcp 9. Rearranging (7.43) gives End

1
R2 = - Figure 7.22 Flowchart of an algorithm for the root locus design of phasc l,111
zC compensation networks.
1'
1
R =-- R (7 .84)
1 pC 2

whcrc thc valuc of C is selected arbitrarily as bcfore.

 Phase-Lag Compensation by Root Locus 435
434 Classical Oesign by Series Compensation
j2

EXAMPLE 7.9
splane

Design a phase-lag series compensator for the negative unity-feedback system

with plant transfer function given by

K - I
'
jl

G (s) - (7.47)
P - s(s + 2)

to satisfy the design specifications

%OS s; 15%

e,,1uni, s; O.OS radia ns

ramp
(7 .85)
.L~ :1-1 -l~
- 2

T his system was considered earlier in Example 7.5 with an additional design
constra int on rise time. Since time responses for phase-lag compensated systems
- jl
are generally much slower than those for phase-lead compensated systems, design
rcq uirements on rise time are not included.
F igure 7.23a shows the root locus plot for the uncompensated system described
by (7.47) and the allowable s-plane region for closed-loop p oles. This region
differs from the one in Figure 7.13 only because the rise time d esign requirement -1 - j l.656
has been omitted here. We refer to Step 2 (Figure 7.22} and select the dominan!
j
closed -loop poles at - a jfl We choose a = l arbitrarily and compute /3 as -j2

tY tan [cos- (imin)l oras a 1 - i~n/!rnin to obtain fJ = 1.656. The points - 1

1 (a)

./ 1.656 are located on the boundary of the allowable region. We next use s-plane
Figure 7.23 (a) S-plane vectors for the uncompensated
vcctors to compute K u for the uncompensated system as and (b) phase-lag compensated systems of Example 7.9.

Ku = [ Vl 2
+ (l.656)
2
J2 = 3.74 (7 .86) pensator-plant transfer function becomes

Ka(s + 0.2)
wh ich is the value shown in Table 7.1 for a 15% overshoot requirement and (7 .88)
p 1 = 2. Therefore, since K 0 = K/p 1 = K / 2 = 1.87, we determine e,, for the
Gpp(s) = s(s + 2)(s + 0 .017)
uncompensated system as 1/K. = 1/1.87 = 0.53 radians and note that this value
of K., is much too small (Step 3). Figure 7.23b shows the s-plane vectors. for computing the value of Ka for the
Thc calcula tion of a in Step 4 is achieved from (7 .83) by tirst dividing K u in compensated system. This calculation utilizes the new dominant closc:d-loop polc
(7 86) by tha t value of K which yields e.. = O.OS radians, that is, by K = 2K 0 locations at -0.90 jl.49, which are determined from (7.88) as the intersecton
2( 1/O.OS) = 40. Arbitrarily choosing 10% for SM in (7 .83) yields ru of the %OS allowable s-plane region boundary with the root Jocus of tht:
compensated system. In other words, s = 0.90 jl.49 satisfies the two equa tions
3 .74 ( 1 given by
= 40 1 + O.l) = 0 .085 (7.87)
~
fij a= tan [cos - 1 (!min)] = 1.656
In Step 5, we arbitrarily place the compensator zero at -0.2, which is 10% of
thc distance from the origin to the second open-loop pole (al s = - p 1 = - 2). /GPP(s) L--,HJP= (2k + 1)180 (7.89)
T hcrefore, the compensator pole is at - az = - 0.017, and the combined com-
436 C11w:;lcal Dcsi911 hy Sodos Cornponsntlon Phase-Lag Compensation by Root Locus 431

1.40 ~--~--.....----...----,---.------,
j2
~ P h a s elag des1gn in (7.90)

~plane

t Phase. fag design in (7 .9 1)

Comparable phaselad des1gn

lrom Figure 7,15

1 00 2.00 300 '1 00 5 00 6 00

Time (secl----
(al

-2
'
..._..... R1 = 1 9 Mil
-Wv o
1
1
+ +

~~
R2 = IOOKO
v,c.~> . ,~voc.~) + ~}

-ji e= IOOF T -o
o

-0.90 - jl.49

(t,)

-j2: Figure 7.24 Phase-lag design results for Example 7.9.

(b)
Computcr simulation results are shown in. Figure 7.24a for the dcsign indicated
Figure 7.23 (Continued) in (7.90) and for GPp(s) given by

he valuc of 0.90 was obtained numerically by moving along lhe line having a 2.00(s + 0.1)
negativc slope al an anglc 8 = cos - 1 (fmin) with the negative real axis in the Gpp(s) = s(s + 2)(s + 0 .005)
J-planc. Using a = 0.90, we easily find fJ = 1.656(0.90) -= 1.49. Using the s-plane
vcctors in Figure 7.23b drawn to s = - 0.90 + Jl.49 yields Ka= 3.393, from
which wc may express (7.88) as ,
20.0(1 + s/0.1)
3.393(s + 0.2) = s(I + s/2}(1 + s/0.005)
GpP(s) = s(s + 2)(s + 0.017)
Figure 7.24n indicates lhat thc %OS is 30% for the dcsign in (7.90) and 14% fo,
20.0(1 + s/0.2) the design in (7.91 ). The time curve for phasc-Jead design from Figure 7. 15 is abo
{7.90) shown lo emphasize th,at phase-lag designs yield significantly slowcr time rt
- s(1 + ;2)(1 + s/0.017) '
, sponses than phase-lead designs.
Selecting C equal to 100 F, wc use (7.84) to oblain R 1 = l.9 MU and
which givc~ K,. = 20.0 and e,, = O.OSO rad ians. Thus, tl1e steady-state error R 2 = 100 K!l in Step 9 of the phase-lag design algorithm. Thesc clcmcnt vahw,
d c~ign spccification has bccn satisficd. for (7.91) are shown for' the nctwork in Figure 7.24b.
 438 Classcal Desgn by Series Compensaton
Phase-Lag Compensaton by Root Locus 439

a-plane
'I $'-planc I
10

j8
j2
6

j4

j2 ji

~
-40
it:: 1 1
-6 -4 -~!ti~
-j2 O
o
-4 1 11"
-40 -p

-ji

-12

(a)

Figure 7.25 Root locus plots for (a) the uncompensated

and (b) phase-lag compensated systems of Example 7.10.
(b)

Figure 7.25 ( Continued)

EXAMPLE 7.1 0
according to (7.83), may be selected as
For 1he 1hird-order system of Examples 7.6 anJ 7.8, designa phase-lag compensa-
tion network to satisfy the specifications in (7.85) of Example 7.9, thal is, 137.8
a= - -( 1 )
= 0.078 (7.92)
%OS ~ 15% and e,. (dueto a unit ramp) ~ 0.05 radians. For convenience, we 1600 1 + 0.10
recall here that the plant transfer function is
where SM has bcen chosen arbitrarily as 10% = 0.10. As in Exampk 7.9, we
arbitrarily sclcc t z = 0.2, and p may be computed to be p = az = 0.078(0.2) =
K
(7.59) 0.0156. As before, the new dominant closed-loop poles (for 1he c.:ompens.11ed
GP(s} = s(s + 2}(s + 40) system) must satisfy (7.89) for G,G/s) given by

Figure 1.25a shows the root locus plot for (7.59). We determine the dominant Ka( s + 0.2)
7 93
closcJ-loop poles at the intcrsection of the root locus with the boundary of the Gp,(s) = s(s + 2)(s + 40}(s + 0.0156) ( - )
a llowable s-plane region for %OS ~ 15% as - 0.96 jl .59. At 1hese closed-loop
polt:s, we next use s-plane vectors to calculatc Ku = 137.8, from which Kv = 1.723 which yields s = - 0.85 j1.4l Figure 7.25b shows the s-plane vcctors used to
and e,,= 0.58 radians. Therefore, e,, is much too large in Step 3 of Figure 7.27. determine Ka as 120.3, from which K = 120.3/a = 120.3/0.078 = 1542.4, K,, ~
We de1crmine the desired value of K as 2(40)/0.05 = 1600, from which a, K/80 = 1542.4/80 = 19.3, and eu = 0.052 radians. T he test in Step 7 of Figure
7.22 fails, and we must decrease a fur1her and return to Step 5.
 440 Classical Design by Series Compensation Phase-Lag Compensation by Bode Procedures

Wc next select 15% for the SM in (7.83) to yield

Start

137.8( 1 )
= 1600 1 + 0.15 = 0.075 {7.94) Set K to satisfy
e,. requirement Step 1

which gives p = az = 0.015 for z selected arbitrarily as 0.2. We again determine

Determine PM" for the
the dominant closed-loop potes for the compensated system as - 0.85 jl.41. uncompensated system Stcp 2

We proceed to use s-plane vectors to determine Ka = 123.6, which gives

I23 .6(s + 0.2) Yes

Step 3
Gpp(s) = s ( s+ 2)( s+ 4-" 1
.

20.6(1 + s/0.2) Determine the new "'Ge

for the compensated syst em Step 4
(7.95)
= s(l + s/2)(1 + s/40)(1 + s/0.015) (include a safety margin)

Therefore, e,s = 0.049 radians, which satisfies the design requirement and passes Place the zero al a value
of one-t enth wcc Step 5
the test in Step 7.
In Step 8, computer simulation results indicated that the %OS for the design in
.
(7.95) is 27%. A redesign yielding an overshoot of 15% has the transfer function Decrease Comput e a and determine No phase-lag
the pole placement compensation Step 6
givcn by '"GC is needed

80.0(s + 0.1)
No
Gpp(s) = s( .t + 2)(s + 4,-" Step /

20.0(1 + s/0.1)
{7.96) Select network
s(l + s/2){1 + s/40)(1 + s/0 .005) parameters Step 8

Again selecting C = 100 F, we determine in Step 9 values for R 1 and R 2 as

End
1.9 M~2 and 100 KU, respectively, which completes the required design of the
phasc-lag network.
Figure 7.26 Flowchart of the Bode phase-lag design procedure.

definiton, too low at that frequcncy ( wGc) where the Bode magnitude curve
crosses the O db leve!. There are two basic series compensation approaches fo
correcting this situation: phase-lead r phase-lag compcnsation. We may us,
7.6 PHASE-LAG COMPENSATION BY BODE PROCEDURES phase-lead compensation to add sufficient phase without altering the magnitudc.:
curve greatly. We described the Bode design. of phase-lead networks in Section
T hc Bode dcsign of phase-lag series compensation networks is perhaps the
sirnplest of the four procedures described in Sections 7.3 through 7.6 [1-6). As in
7.4. On the other hand, we may use phase-lag compensation to attcnuate thJ.,...
magnitude curve, resulting in a somewhat lower gain crossover frequency at whid
Section 7.4, typical design spccifications for which Bode procedures are most the phase curve yields the proper amount of phase to mect the phase margin
uscful include rcstrictions on the phase margin and on the steady-state error requirement. We describe1the details of this second approach in this section.
m..ulti11g from a unit ramp input. We begin, as before, by constructing Bode Figure 7.26 provides a flowchart of the Bode phase-lag design procedure, anl
magnitude and phase diagrams for the uncompensated system whcn the forward Figure 7.27 shows a typical case. The first three design steps are identical to thos,
loop gain K is selected to barely satisfy the steady-state error requirement. If the for the phase-lead design algorithm in Figure 7.19. lf the phase margin requirc-
phase margin for the unco!11pensatcd system is too small, the phase curve is, by ment is not satisfied in Step 3, we proceed in Step 4 to determine that frcqucnc)I
 Phase-Lag Compensation by Bode Procedures 443
442 Classical Design by Series Compensation
arbitrarily and then calculating R 1 and R 2 from (7.43) as

aT
R2 = - I
C

3
t R1
l - a)
= ( -a- R,2 (7.97)
:::,
~ 20
ff
o
N EXAMPLE 7 .11

Design a phase-lag compensation network to yield PM : 40 and e ss due to a

unit ramp input :,; 0.05 radians for the negative unity-feedback system wilh plant
-20
transfer function given by
K
{7.47)
G/s) = s(s + 2)
O n ('\')

't~'
0.05 0.1 0.2 0 .5 2 This system and design requirements were considered earlier in Exampk 7.7 for
t 1
Uncompensated
phase-lead compensation, where Steps 1, 2, and 3 yieldcd K = 40 and P M,, = 18,
which gives ess = 0.05 radians but is not large enough to satisfy the design
11-90
~
IG(iw) =
---....__
..G"(iwll
specification on the phase margin. Figure 7.27 shows Bode diagrams for this <.:ase.
PMu If we include a safety margin of 5 in Step 4, we seek to determine the w for
which
- 180.- - - _ _ ,_ - - ...- - - -
/ G/jw) = - 180 + (40 + 5) = - 135 (7 .98)

Figure 7.27 Application of the design procedure in Figure 7.26 for

phase-lag network design. Using (7.47) in (7.98) allows us to solve easily for w as 2 radians/second, whit:h
we select as the new gain crossover frequency for the compensated system ( wc;c ) .
Therefore, in Step 5, we arbitrarily place the zero (I/aT) at one-tenth "'ce, that
at which the phase is large enough to meet the design specification, provided the is, 1/aT = 0.2. In Step 6, we determine a as
new gain crossover frequency ( wcc) occurs at that point. An additional amount
of phase, perhaps 5 or so, should be included as a safety margin~ since the
phase-lag network that achieves the desired attenuation also Iowers the phase 1
fi = 0.141 (7.99)
curve slightly at "'ce In Step 5, we arbitrarily place the compensator zero at a = 40 = 10
frequency of one-tenth the desired new gain crossover frequency "'ce l.9 Step 6,
we determine a to provide the necessary attenuation as the reciproca) of IG/jw)!, "'ce/w~c +. 2 2 '-'cc- 2
cvaluated at the new gain crossover frequency for the compensated system
2
( wcd The compensator pole (at 1/T) is Iocated by multiplying the zero We next calculate 1/T as (1/aT)a = 0.0283, from which
location (1/aT) by a. lf PMc is not large enough in Step 7, we decrease "'ce
(incr.:ase the safety margin) and return to Step 5. However, if PM,. is adequate, 20(1 + s/0.2)
(7.100)
we proceed to determine the phase-lag network parameters by selecting C Gpp(s) = -s(_l_+_s-/2)(1 + s/0.0283)
...
In Step 7, we calculate /cpP(j2) = -139.9, yielding a PM,. of 40.1, wlch
2
Alterna tively, i( A is the necessary attcnuation in decibels, we set - 20 log10 a = A and determine a satisfies the design specification. In Step 8, we arbitrarily select C = 100 F and
as 10 - ,,;20
444 Classical Design by Series Compensation Design Comparisons and Extens,on,

use (7.97) to give by setting C = 100 F and computing values o R 1 and R 2 as

aT 1/0.2 aT 1/0.183 -
R2 = - = = 55 KO
R 2= C 100 . JO 6 = 50 KO e 100. 10- 6

R1 = ( -1 -a) R1 =
(1-0.124)
_ : 55 KO = 388 KO
= 1- a) = {1 - 0.141) 0
-
0 124
R1 ( -a- R 2 0.141
50 KO = 304 KO (7.101)

which completes thc phasc-lag network design.

EXAMPLE 7.12 7.7 DESIGN COMPARISONS ANO EXTENSIONS

Dcsign a phase-lag compensation nctwork to satisy PM ~ 40 and e.,I .Unll

s; 0.05 In this scction we make phase-Iead and phase-lag compensation design compari-
radians for the negative unity-feedback system having ramp sons by first dcscribing the usefulness and limitations of each approach and thcn
identifying selccted cases for which only one o the design procedures is applica
K ble. We introduce the lag-lead compensatfon network as an extension of thc
Gr(s) = s(s + 2)(s + 40) (7.59) concepts in Sections 7.3 through 7.6 and suggest the use of automated digital
computer design routines, perhaps involving computer graphics, for this com
, his problcm was considered for phase-lcad compensation in Examplc 7.8, pensation design.
\\ hcn it was found the ess = 0.05 radians requires K = 1600, and the value o
ph:i~c margin for the uncompensated system is 9.4, which is too small (Step 3).
i\p.ain induding a safety margin of 5 in Step 4, we use (7.59) to caJculatc wcc 7.7 .1 Phase-Lead and Phase-Lag Deslgn Comparlsons
from
For both phase-lead and phase-lag design, we sclcct bctween root locus and Bode
/ Gr (Jwr;c ) = - 90 - tan 1
( wcc/2) - tan 1
( wGc/40) procedures according to thc design specifications to be satisfied. As indicated in
the opcning remarks of Section 7.4, root locus procedurcs are more directly
applicable when given time response performance specifications such as %OS, risc
= - 180 + (40 + 5) = -135 (7.102) time, dclay time, or settng time. We can use the root locus approach to place
dominant closed-loop pales within s-plane regions in which these design specifica- .
,\.hich yields wGc = 1.83 radians/second. We select 1/aT = 0.183 and calculate tion5 are only approximately satisfied. We then select compensator zcros and 1 ,
potes to yield a sufficiently large error constant, for example, K,. for Type l
systems, to satisy steady-state erro, requirements. Computer simulations of time
l
= 0.124 responses are needed to enable us to properly cvaluale thesc designs and t _.
= 1600
(7.103)
determine whether a redcsign is required. On the other hand, we find that Bode .,_...._.,.....
procedures are more dircctly applicable when frequency response performance
1
Wc;c/w~c + 2 2 /w~c + 40 2 .,Gc I 83 specifications such as phase margin, gain margin, system peak roognitude, N
system bandwidth are given. Jnitially, we set the forward-loop gain as the valuc 1 . 4
frorn which 1/T = (1/aT)a = 0.0226. Therefore, needed to mcet the steady-state error requirement and proceed to eithcr incrca~c
the phase (phase-lead) or ifecrease the magnitude (phase-lag) in the vicinity of tht
20(1 + s/0.183) gain crossover frequency (for PM specification) for the compensatcd system. 1
I {7.104)
GPr(s) = s{l + s/2)(1 + s/40)(1 + s/0.0226)
' We follow thc~ samc proccdurcs whcn a gain margin rcquircmcnt is spccificd. cxccpl t"ond,111111
frorn which /GPr{J.83) = - 140.0 and PMr = 40.0. We complete the design musl be satisficd at thc phase crossovcr rcqucncy for thc compensatcd systcm. Scc Probkm 7 1R
 446 Classical Design by Series Compensation Oesign Comparisons and Extensions 447

R
TABLE 7.3 Phase-Lead and Phase-Lag Dealgn CharactertsUcs

Phase-/ead design:
Provides series compensation by the placement of dominant s-plane closed-loop
+ C
'+
'
poles to satisfy time response specifications and by additional phase to satisfy
frequency response specifications. R2

Yields faster rise, delay, and Sl!ttling times. V,(sl V 0 (s)

Results in an increased system bandwidth.

TC2


Requires additional forward-loop gain.
Not applicable to those systems requiring more than approximately 61 o
additional phase at the new gain crossover frequency. o~---= Figure 7.28 A lag-lead network.

Phase-/ug design: A phase-lag compensated system attenuates the frequency response magnitude
Provides series compensation by increasing the error constant for given dominant in the vicinity of the gain crossover frequency, resulting in a reduce<l system
s-plane closed-loop poles to satisfy time response specifications and by magnitude bandwidth and, consequently, a slower time response- that is, highcr values o
attenuation to satisfy frequency response specifications. rise, delay, and settling times. Phase-lag design is applicable only for those cases
Yields slower transicnt responses. in which there exists sorne frequency having the desired phase. Problem 7.20
Results in a decreased system bandwidth. provides a situation where phase-lag compensation cannot be used.
Not applicable to those systems having no frequency at which the phase exists to
yield the desired phase margin.

7. 7 .2 Lag-Lead Compensation
As a useful extension of the procedures developed thus far in the chaptcr, we may
When both time response and frequency response specifications are given for a combine phase-lead and phase-lag design concepts to yield a lag-lead network
pa rticular design, such as %OS, PM, and steady-state error (e., ) requirements, we that retains many of the desirable features of the separate networks. In lag-lead
select what we consider to be the most stringent subset of these specifications for design, we both attenuate the magnitude curve (from lag design) and provide
which either root locus or Bode procedures apply, perform the required design, additional phase (from lead design) at the gain crossover frequency . The ad-
and then verify that the remaining specifications are also satisfied . Iterations may vantages of lag-lead compensation are (1) the system bandwidth is incrcascd
be necessary to meet ali specifications simuhaneously. (yielding a faster transient response) beyond that which would be achicved with
T able 7.3 summarizes the characteristics of phase-lead and phase-Jag com- only phase-lag compensation and (2) the requirement of additional forward-loop
pensation. As indicated earlier, a phase-lead compensated system has faster rise, gain inherent in phase-lead compensation can be avoided.
delay, and settling times anda corresponding larger system bandwidth. How large Figure 7.28 shows a lag-lead network for which the voltage transfer function
the system bandwidth should be depends on the particular application. If it is too can be expressed as 4
large, high-frequ ency noise can become a problem; if too smalJ, the speed of the
time response can become too slow. The rule of thumb that limits a in phase-lead
design to a value of approximately 15 is intended to prohibit excessive system V0 (s) (1 + a 1T1,s )(l + a2T2 s )
G((s) = -V-(s-') = - (l_ +_ T-s-)(_l _+_T_ s_)_ (7.106)
bandwidths. This value of a = 15 limits the allowable amount of phase that can 1 2
he added ( <f,,,,) to approximately 61? Problem 7.19 shows a case where phase-lead
design is not possible because the phase is decreasing rapidly at Wcc for the
~
uncompensated system, indicating the need for such a large safety margin that
<P,,, > 61 would be necessary. In brief, the limitations on phase-lead design are 4 T hc lag-lead network is sorne times designated as a lcad-lag network becausc the lead p:irt ( R 1 :ind
the requirement for additional forward-loop gain (by a factor of a) and the C 1 ) is encountered at the input terminals. We prefer the tcrminology lug-lead bccau,o: 1h~ lag clfoct
restric tion that <Pm .$ 61 to prohibit an excessive system bandwidth. occurs al lower frcq uencies.
448 Classical Design by Series Compensation Oesign Comparisons and Extensions 449

where
OT = R2C2
40
a 2 T2 = R 1C1
12 = 1
20'
T1 + T2 = R 1C1 + R 2C2 + R 1C2 (7.107)
3
:.:,
The z.ero and pole resulting from the lag nctwork (R 2 and C2 ) occur at l/a 1T1 o
and 1/T1, respectively, where O < a 1 < 1. Furthermore, the z.ero and pote result- .,.
o
.!2
0
1004 0. 1 0.2 0 .4 20 40
ing from the lead network (R 1 and C1) occur at l/a 2 T2 and 1/T2 , respectively, o
N
where a 2 > l. In brief, we require 1/T1 < 1/a 1T1 < 1/a 2 T2 < 1/T2 - 20
Suppose the design specifications include a phase margin requirement, a limit
on the steady-state error due to a unit ramp input (Type l system), and a
closed-loop system bandwidth that is the same for compensated and uncom- -40
pcnsatcd systcms. We may use the ftowchart described in Figure 7.29 for lag-lead

Start -60'
I 1 1 1 1 . -~

Construct Bode diagrams with K set 0.1 0 .2 0.4 20 -10

to satisly the e,,, requi rement
Step 1
--1--1---
Compensated
IG(i,a) = Gi;Gp(i,o)I

Yes
]
~-180
Step 2

-270
Determine <l>m . Step 3
and calculate n 7

Figure 7.30 Bode magnitude and phase d iagrams illustraling lag-lead network
design.
Calculate <te
and T2 Stcp 4

lncrease
<>m
design. In Step 1, we set K to satisfy the steady-state error requirement and plot
Determine a 1 No lag-lead thc Bode magnitude and phase curves, as in Figure 7.30. We determine PM,, in
and T 1 compensation Step 5
is needed Stcp 2 and compare it to the phasc margin specification. If PM11 is )arge enough,
we do not necd to add a compcnsatiot) network, and thus we termnate the
procedure at this pont. If PM11 is too small, we determine the amounl of
No
Step 6 additional phase needed from

4'm = PMrnin - PM.. (7.108)

Setect network
parameters Step 7
where PMmin is the rninin1um aUowable phase margin according to the design {

End
specification. Since thc new gain crossover frequency for the compensatcd systcm 1c
--1 S:
...
occurs at a slightly lower value than the corresponding "'ce for the uncom-
pensated system (presumably with greater phase at that new frequency), wc
Figure 7.29 Flowcha_rt of an algorithm for lag-lead network design. usually can apply (7.108) wthout an additional safety margin. As in phase-lcad
 450 Ctassical Design by Series Compensation Design Comparisons and Extensions 451
ilcMgn, we next calculate a 2 in Step 3 from
and the requiremenl thal the allenuaton for large w should be approximately the
1 + sinq,m same for both compensated and uncompensated syslems. The planl transfer
a2 = (7.109) function for this negative unity-feedback syst~.m is given by ,
1 - sin<Pm
K
In Step 4, we locate the new Wcc by using GP(s) = s ( s+2)( s .+ 40) (7 .59)

IGP (Jwcc} 1 = /;;; (7.110) Phase-lead and phase-lag designs were consid~ed in Examples 7.8 and 7.12 for
this system and the first two design requrements.
As a n approximation, we may express The typical Bode magnitude and phase curves used for illustration in Figure
7.30 apply to the current design problem. In Steps 1 and 2, we calculate K,, = 20
l/T2 = wcc{i; (or K = 1600), construct the Bode diagrams just described, and determine PMu
= 9.4 , as in Examples 7 .8 and 7.12. In Step 3 we calculate a 2 = 3.07 from
l/aT2 = (l/T2 }a 2 (7.lll) (7.109) and in Step 4 determine "'ce= 4.6 radians/sccond from (7.110), that is,
1600
Mo reover, we arbitrarily select the zero corresponding to the lag part of the (7 .116)
uctwork in Step 5 at Wcc/w~c + 22 /w~c + 402 = ,/3.07
l/a 1T1 = 0.1(1/a 2 T2 ) (7.112) Therefore, from (7.111), we have
l, o m which
1/T.2 = '-'ce;;; = 4.6,/3.07 = 8.0
1/T1 = (1/a 1 T1 )a 1 = (1/a 1T1 }/a 2 (7.113) l/a 2T2 = (1/T2 )/a 2 = 8.0/3.07 = 2.6 (7 .117)

., i nce a 1 = I/a 2 In Step 6, we determine whether PMc is large enough. If not, we

.
In Step 5, we use (7.112) and (7.113) to calculate
1ncrease </>,.. and recalculate a 2 from (7.109) in Step 3. If so, we proceed to select
lag-lead network parameters in Step 7 by choosing C 1 arbtrarily and then using 1/a 1T1 = O.l(l/a 2 T2 ) = 0.1(2.6) = 0.26
1/T1 = (I/a 1T1)/a 2 = 0.26/3.07 = 0.085 (7 .11 8)
a2T2
R1=-
We may now express Gpp(s) as
C1
20(1 + s/0.26)(1 + s/2 .6)
T1(l - a 1) + T2 (1 - a2) 7 9
C2 = - -- -- -- - - G,Gp(s) = s(l + s/2)(1 + s/40)(1 + s/0.085)(1 + s/8.0) ( .ll )
R
From (7.119), we determine PMc = 44.5 (at "'ce= 4.6 radians/second), which
a 1T1 satisfies the phase margin design requirement.
R =- (7.114) Selecting C 1 = 1 F arbitrarily, we use (7.114) in Step 7 to yield '
2 C2
1
which are determined from (7.108).
( 2 6) -
R1= . = 385 Kl
10- 6
1
EXAMPLE 7.13
(0 .085) - 1 -
1
(0.26) - + (s.or: 1 - (2.6) -
C2 = = 42.1 F
385 K
1>csign a lag-lead compensator network to satsfy
(o26)- 1

PM 40
R2 = 42.1. . 10- "' =91.5 Kl
6
(7 .120)

e.. l unii .:s; 0.05 radians (7.115) which completes the lag-lead network design.
ramp
452 Classical Oesign by Series Compensation Problem'i '15'. t

7 .8 DISCUSSION ANO SUMMARY REFERENCES

An important feature in the practica! design of phase-lead, phase-lag, and Jag-Jead l. T. E. Fortrnann and K. L. Hitz. An Jntroduction to Linear Control Syst11111
compcnsatio n networks is the sensitivity of the system performance to each New York: Marce) Dekker, 1977.
nctwork parametcr. For cxample, if design specifications on %OS or PM are 2. R. C. Dorf. Modem Control Systems, 3rd ed. Chapter 10. Reading, Ma<;c;,;
given, it is uscful to calculate their sensitivities with respcct to R 1, R 2 , and C for Addison-Wesly, 1980.
phase-lead and phase-lag designs and with respect to R 1, R 2 , C1, and C2 for 3. B. C. Kuo. Automatic Control Systems, 4th ed. Chapters 8 and 10. Englewood
lag-lead design. Having determined these sensitivities about nominal network Cliffs, N.J.: Prentice-Hall, 1982.
parameters, the control system designer can then determine the deleterious effects 4. J. J. D' Azzo and C. H. Houpis. Linear Control System Analysis and Dcs,gn
on system performance resulting from the (hopefully slight) mismatch when using 2nd ed. Chapters 10 and 11. New York: McGraw-Hill, 1981.
olT-the-shelf components. A careful sensitivity analysis during the design stage S. A. P. Sage. Linear Systems Control. Chapters 5 and 6. Charnpaign. 111.
can often prcvent the necd for a system redesign after extcnsive field testing has Matrix Publishers, 1978.
hcgun. 6. J. L. Melsa and D. G. Schultz. Linear Control Sy.5tems. Chapter 10. Ne,,
The series compensation procedures described in this chapter for single-input, York: McGraw-Hill, 1969.
singlc-oulput systems are directly applicable to decoupled multivariable control 7. D. H. Owens. Multivariable and Optima/ Systems. London, Engl:md
systems [7). Such systems have the same number of inputs and outputs with Academic Press, 1981.
dccoupling accomplishcd by multivariable feedback to allow the first output to be 8. H . H. Rosenbrock. State-Space and Multivariable Theory. New York: Wilcy,
inlucnced by o nly the first input, the second output by the sccond input, and so 1970. '
forth. Thcrefore, series compensation can be introduced independently in each 9. A. G. J. MacFarlane, editor. Frequency-Rdponse Methods in Control Sy.ttcm1 I
scparate channel of the decoupled multivariable control system. Additional New York: IEEE Press, 1979.
analysis and design procedures for multivariable systems are described by
Rosenhrock (8) and MacFarlane [9).
Wc can sclect from a wide range of computational aids to automate the steps of
1hc dcsign algori thms prescntcd in this chapter. We can casily place the algorithm PROBLEMS
stcps on programmable calculators, which have an advantage o portability for
classroom testing purposes. We might also choose to forro control system design (7.1) 7.1 For the system in Figure 7.31, determine the rangcs of K
software packages on digital computcrs ranging rom personal computers to satisfy each of thc design specifications given by
largc-scale centralized computcrs. Finally, we could employ interactive computer
graphics facilities not only to perform the system design itself, but also to
simulatc thc system's time and frequen cy responses and verify that ali sensitivity %OS~ 20%
rnnsiderations have bcen included.
In summary, we have introduced the series compcnsation approach in this e.,I unit ~ 0.02 radians
ramp
chaptcr for dcsigning feedback control systems that satisfy givcn time and
frequency response performance specifications. We presentcd algorithms for
phasc-lead and phasc-lag design based on both root locus and Bode procedures. RC.tl + 1( ) '(<)
l:) F,(~)
Wc illustrated each design procedurc with two examples of a positional ~ 1 <(< + 4)

servomcchanism (second-order and third-order systems). We described a lag-lead

dcsign procedurc that combincd the phase-lead and phase-lag networks to yicld a
passive network having the advantages of each. Finally, we discussed the impor-
tancc of sensitivity analysis for the designed control system, the extension of
dcsign rcsults to uncoupled multivariable control systems, and the variety of Figure 7.31 The system for Problem 7 .1.
computational aids that can be employcd for design purposes. In the following ...
chapttr, we compare this transfer-function feedback design based on dynamic (7.1) 7.2 What are conservative estima tes of t,, t ti and t, for thc second onlt
series compcnsators with state variable feedback design based on optima) control system in Figure 7.32 when K = 1.0? Determine the ranges nf ,
thcMy. satisfying each of the other design specifications listed in Tahlc 7
 Problems 455 t
454 Classical Design by Series Compensation
( 7.2) 7.6 Repeat Problem 7.5 for a phase-lag network with
H(s) + K <, + 1) Y (s)
< - 1) (s ,.. V0 (s) = 1 + s/20
G, ) V;(s) 1 + s/10
'
( 7.3) 7.7 Design a phase-lead network by the root locus procedure for a
Figure 7.32 The system for Problem 7.2. negative unity-feedback system having a plant transfer funclio n
given by
(~ 7 1) 7.3 For the negative unity-feedback system having
K
G (s)- -
K ' - s(s + 4)
G ( s) = -s (_s_+_l_)(-s -+ -40-)
The design specifications are

determine the range of K that simultaneously satisfies the three

%OS .s 20%
design specifications given by t, ~ 3.0 seconds

MP .s 1.6 e,.I unit ~ 0.04 radians

ramp

PM ~ 4.5 Specify values of R 1 and R 2 , if C = 1 F. Use computer simu-

lations to verify that the design specifications are satisfied.
GM~ 5 ( 7.3) 7.8 Determine the smallest value possible for e,. in Problem 7.7 if the
dorninant closed-loop potes are to have a negative real part of - 6.
( 7 1) 7.4 Determine the region in the s-plane for which the following specifi- The ratio a = p/z must not be greater than 15. ....
cations are satisfied simultaneously. ( 7.3) 7.9 Repeat Problem 7.7 if the plant transfer function is given by

%OS .s 20% K
G,(s) = -s(_s_+_ l_)-(s- +
- 40- )
t, .s 4.0 scconds
( 7 .4) 7.10 Design a phase-lead compensation network if
tJ .S 2.0 seconds
K
t, .s 6 .O seconds G,(s) = s(s + 4)

and it is required that PM ~ 45 and e,, dueto a unit ramp is 0.04

For thc system described in Problem 7.3, what range of K (if any) is
radians or less. Determine network parameters if C = 1 F.
needed to satisfy each of the specifications?
(7.4) 7.11 Designa phase-lead compensatin network if
7.5 Suppose we have designed a phase-lead network with
K
V.(s) s+10 ~
Gp(s) = s(s + 4)(s + 80)
G,(s) = :.(s) = s + 20
and the design specifications are PM ~ 45 and e,, due to a unit
ramp input is 0.04 radians or less. Determine nctwork paramctcn, 1f
What values o R 1 and R 2 should be selected if C = 1 .F? R epeat C = l F.
for C = 10 F.
l~
456 Classical Design by Series Compensation Problems 4,,1

( 7.5) 7.12 Design a phase-lag network for the system of Problem 7.10 if the ( 7.7) 7.18 Design a series compensation network for the negative unil y fI
plant and steady-state error requirement are unchanged but PM ~ back system having 1

45 is replaced by %OS~ 12%. Determine network parameters if

K
C = 100 .F. Use computer simulations to verify your design.
Gp(s) = s.(s + 2)(s + 40)
( 7.5) 7.13 Design a phase-lag network Cor the system of Problem 7.11 if the
plant and e,, requirement are unchanged but PM ~ 45 is replaced subject to
by %OS~ 12%. Determine network parameters if C = 100 F.
Verify your design by using computer simulations. GM~ 5

( 7.6) 7.14 Design a phase-lag network for the system of Problem 7.10. De- e,,I ,mit ~ 0 .05 radians
termine network parameters if C = 100 F. r amp

( 7.6) 7.15 Design a phase-lag network for the system of Problem 7.11. De- ( 7.7) 7.19 Design a series compensation network for the negative unity-fcrd
termine network parameters if C = 100 F.
back system having
( 7.3, 7.16 Let the design specifications for a negative unity-feedback system K
7.5) having G/s) = s(s2 + 20s + 106)

GP(s) = - . -
K
subject to ..
PM ~ 50
be given by e.,I unit ~ 0 .05 radians
ramp

%OS~ 20%
( 7.7) 7.20 Design a series compensation network for thc ncgativc unity fr,d
] e,.,l 11nit ~ 0.06 radians
back system having ... J : 1 1
t rarnp
K
GP(s) = s2(s + t)
Design both a phase-lcad network ( C = 1 F) and a phase-lag
nctwork ( C = 100 F) to meet these dcsign specifications. Show subject to

( 7.7) 7.17
time response curves by using computer simulations.
.
Dcsign a series compensation network for the negative unity-feed-
%OS~ 20%

back system having e.,I 11ni1 ~ O.OS radians/sccond

paral>ola

GP(s) = ~ ( 7.7) 7.21 Designa lag-lead network if

K
subject to
Gp( s) = s( s + 4)
%OS~ 30%
and it is lequired that PM ~ 45, e,, dueto a unit rnmp inp11t 1
PM ~ 45 0.04 radians or less, and the lag-lead compensated systcm is to ~
t
approximately the same attcnuation for large w as thc 1111tt-n"""'"'
e,.I unit ~ 0.05 radians pensated system. Spccify network parametcrs.
ramp
 Reformulation of the Design Problem 459

8
- TABLE 8.1 Typlcal Performance Measures

Criterion Mathematial Expressioh

t,
Integral squared error (ISE) J(u) =
f e dt 2

State Variable Design Using Integral absoluta error (IAE) J(u) =

lo
I,

J, edt
lo

Optimal Control Theory Integral time squared error (ITSE) J(u) =

f (t - t )e dt
lo
I,
0
2

Integral time absolute error (ITAE) J

J(u) = (t - t )Jeldt
lo
I,
0

l:xperience with alternative design concepts provides the control engineer llexibil-
Mnimum energy J(u) =
J u2 dt
lo
I
i ty in approaching the design task. This knowledge can be useful, for example,
when deciding between a percent overshoot constraint and the integral of squared
Mnimum fuel J(u) =
Jltudt
e1ror as a criterion for aircraft autopilot design. Moreover, decisions can be made Mnimum time J(u) =
f dt
lo
more easily on whether a dynamic or static controller is appropriate for a given I,

application. Finally, the design approach may be chosen to simplify al~orithmic Quadratic in state and control J(u) =
J (xrox + yu )dt
lo
2

:.teps for large-scale system design or to utilize automated design routines. In

hrid, familiarity with alternative designs permits choices regarding (I) the aspects
ol system behavior to be emphasized, (2) controller structure and characteristics, referred to as a performance index, cost functional, or penalty function, thc
., and (3) operational steps leading toan acceptable controller. performance measure indicates how well the system performs between an initial
The series compensation techniques of Chapter 7 are based on classical s-plane time t O and a final time t1. Our design goal is to minimize the performance
(root locus) or frequency response (Bode) methods that utilize only the plant measure (1 - 4). ...
o utput for feedback with a dynamic controller. In this final chapter on design, we The selection of a performance measure is based on which aspects of systt:m
employ modern state-space procedures that require the availability of all state behavior in a particular application are deemed most important to the control
variables to form linear static cntrollers. Rather than using percent overshoot or system designer. If the output time response is required to approach its final value
seuling time as performance specifications, we select performance measures rapidly, then errors in the state variables should be weighted heavily in the
formed as the integrals of weighted functions of the state variables and plant performance measure. On the other hand, if a major design concern is to conserve
input. After minimizing these performance rneasures, we can compare the result- energy or fue! in a spacecraft application, the plant input should be a promincnt
ing closed-loop systems with those obtained earlier frorn classical design proce- part of the performance measure.
durcs. We identify several typical performance measures in Table 8.1. The ftrst two
criteria are based on minimizing integrals of the squared error or absolutc value
of the error. Time weights are incorporated into the ncxl two criteria to cmphasizc
a reduction of errors occurring later in the time response. The next thn:e
8.1 REFORMULATION OF THE DESIGN PROBLEM performance measures focus on minimizing energy, fue!, or time. The last entry in
Table 8.1 is a quadratic performance measure composed of the integral of a
The relative quality of a system design can be determined by a performance weighted combination of the squares of state variables and the plant input. Both
rneasure that provides a figure o/ merit as a single numbcr. Consider the general the selection of the type of performance measure and the choice of wcights are
performance measure given by often decisions to be made by the control system designer.
~
J(u) = j' g(x,u,t)dt
1
(1)
' 8.1.1 Parameter Optimization
where g(x, u, t) is a functional of the state x, the plant control u, and time t. Also When the controller structure or configuration is specified, we rnay adjust
458 controller parameters to minimize the given performance mcasure. We rdcr to

11 j
l'
1
1
460 State Variable Design Using Optima! Control Theory Reformulation of the Design Problom

this design procedure as parameter optimization. For exarnple, we prove later in

1
this chapter that the optima) feedback control law for a linear plant with a s + 1
quadratic performance measure is itself linear. Al this point, we arbitrarily
assume a linear controller configuration and apply parameter optimization.
The mathematical operation involved in solving this linear quadratic control
problcm by parameter optimization evaluates the quadratic performance measure
given by
Figure 8.1 The linear system for parameter optimization
J(u) = j\xrQx + yu 2 ) dt (8.2) in Example 8.1.
'
and x 2 (0) = O, we apply a unit step input [r(t) = 1], which drives x 1(t) towa;
under thc assumption of a linear controller
+ 1 as t -+ oo . As a performance measure, we select
u= krx (8.3)
J( u) = f( e
o
2
+ yu 2 ) dt
Suhstituting (8.3) into (8.2) yields
0
={ ((1 - x 1)2 + y(l - x1 - K,x 2 ) 2] dt
J(u) = J\r[Q + ykkr]x dt (8.4) o '
' From either (8.5) or Figure 8.1, we observt?that wn = 1 and f = 1/ 2 for t
We then select the controller parameters in the vector k to minimize J(u) in .(8.4). system without tachometer feedback. When tacbometer feedback is inserted, "''
In general, we refer to the resulting feedback control law as a specific optima! might obtain either an underdamped, critically damped, or ovcrdamped systr 111
control law because the controller configuration has been specified. Moreover, we time response. An underdarnped response has the form
say that the design is suboptimal if one or more of the controller parameters in
(8.3) have been fixed and the optimization is performed by adjusting the remain-
e- r,,
ing free controller parameters. While the specific optimal control law in (8.3) may y(t) = x 1(t) = 1 - ,.----;; sin(wdt + O)
be cithcr optima! or suboptimal for the linear quadratic control problem, depend-
V1 - f,2
ing on whether ali parameters in k are free for adjustment, this linear controller
can only be suboptimal for those problems in which the optima! controller
where r, is the systcm damping coefficient and wd and o satisfy
configuration is nonlinear.
wd = /t - r,2

O= cos - 1{r,)
EXAMPLE 8.1
We form the derivative of y(t) in (8.8) as
Consider thc linear system of Figure 8.1 having a plant transfer function given by
e- r,,
Y(s) 1 y( t) = xi( t) =; ,.----;: sin wdt
U(s) = s(s + 1)
(8.5) V1 - t,2
Using (8.8) and (8.10) in (8.7) gives
Thc error e is the diffcrence in the input r(t) and the output y, which is fed back
negatively as shown. We now add tachometer feedback to this system to form a 2
linear controller u(x) as
r,
J(K,) = [ ~ ( [ ~ sin(wdt +
1-
0)]
, u = e - K, y

= r(t} - x 1 - K,x 2 (8 .6) e- r,, e- r,, ] 2)

+y -;---;: sin(w,l + O) - K, r;- sinwdt dt
where we havc selected phasc yariables x 1 = y and x 2 = y. Starting with x 1(0) = O
[ V1 - t,2 V1 - t,2
462 State Variable Design Using Optima! Control Theory lntroduction to Optimal Control 463

The performance measure J in (8.11) is to be minirnized by adjusting the Although these gains may also be found by parameter optimization, we prefer the
parameter K, in Figure 8.1. However, according to (4.46), the relationship direct procedure of the following sections for their calculaliop.
between K, and f, is 2f,wn = 2fwn + K,w~. For wn = 1 and f = 1/2, we have
'
f,=t(K,+1) (8.12)
8.2 INTRODUCTION TO OPTIMAL CONTROL
Consequently, we have an underdamped response (O < f, < 1) Cor -1 < K, < l.
We note that the system response is unstable Cor K, < - 1 and overdamped for We presented the concept of a performance measure in the last section as a means
K, > l. Substituting (8.9) and (8.12) into (8.11) yields an integral for J as a of assessing the relative quality of a system design. Once a particular controller
function of K, and y. Expanding the integrand by using trigonometric identities, configuration is assumed, we may use parameter optimization to determine thc
we can evaluate the resulting expressions with the aid of integral tables to obtain best feedback solution (lowest value of J) uoder that assumption. A basic
problem is to determine which of ali possible configurations yields the lowc.:!>t
J(K,) = 1 + 2y + (K, + 1)2 value of the performance measure J. Even if we do not choose to build that
(8.13) optima! controller structure for our particular application, we would neverthekss
2(K, + 1)
lilce to know how the controller we are using compares with this optima! solution. 1
C urves of J versus K, are shown in Figure 8.2 Cor y = O, 0.5, 1, and 1.5 with
minimum points occurring for K, = O, 0.41, 0.73, and 1, respectively. For
y > 1.5, mnima of J occur for values of K, greater than unity and, hencc, 8.2.1 Problem Formulatlon
corresponding optima! system time responses Cor those performance measures are
In deriving necessary conditions for the general optimization problem, we con-
overdamped.
sider the performance measure (1 - 4) definoo by

The results in Figure 8.2 are generally suboptimal since only one state variable
J(u) = { 'g(x,u,t)dt (8.14)
{x 2 ) is fed back through an adjustable gain determined by parameter optimiza-
tion. We observe that it may be possible lo obtain even lower values of J(u) in Let the plant be dcscribed by
.
{8.7) by feeding back both x 1 and x 2 through adjustable gains k 1 and k 2 .
x = /(x, u, t) (8 .15)
5
with known boundary conditions on the state x(t) at , = O, that is, x(O). Wc
assume that the class of plant inputs is unconstrained and, consc4uently, anv
value of u is admissible.
The optimiz.ation problem is to minimize the performance mca,urc J( 11) i11
(8.14) subject to the dynamic plant equations in (8.15) ami thl.' "11u.i11d
boundary conditions.
;. Locos ol minima
as yva, ies
::,

8.2.2 Euler-Lagrange Equatlons

As a preliminary step in the solu1io11 11l U11 ~ 1111f1111\ lf111nn l 11i!ilit! t 111,11111
problem, consider the pcrfor~1.111,c 1111 is1111 : t1ivcn liy
o.__________,.__________,..__________,._
-0.5 O 0.5 1.0 .1( 1 ) lii [ 't:( 1
1
1 i1, I i t/i
K,-
Figure 8.2 Curves of J versus K, tor Example 8.1. where y{t) is a rn11 1111111 111 s ;;,~f1L11 ' f1111UH)11 I 1 1m i!W 111t, \'!i l !111 / 11 li
I 464 State Variable Design Using Optima! Control Theory lntroduction to Optimal Control llf1:,

r term. As a necessary condition to minimize J(y) in (8.16), we set the dcriva 11vl
(8.20) equal to zero and simultaneously set e equal to zero, thcreby forcing ,.,,
to be the optima! solution y(t). Since this condition must hold for any arhitr.11y
y(tol
variation 'IJ(l), we obtain

t
y(/) g_!!_(g)JI
[ay =0
y{I~
dt ay y-v

which is rcfcrred to as the Euler-Lagrange equation for

variable y in (8.16). For the vector case, (8.21) becomcs

,_ [ag !!_ ( ag )]! =0

ay - dt ay _., .
1
Figure 8.3 lllustrating the time variation 11(t) about an
optimal solution y(t). We are now ready to solve the optima! control problem formulated in (8 14)
pieccwise continuous derivatives. Let y (t) be an optima( solution and TJ(t) an
arbitrary variation about this optimal solution, that is,
and (8.15). We begin by defining a function L as
.
L{x, >.. 11 , t) = g(x. u, t) + >.T[f{x. 11, t) - x]
y( t) = y(t) + n1(t) (8.17)
whcre >. repres~nts an 11-vcctor o( Lagrangc multipliers which serve as difTcrent
whcrc f is a small, positive scalar constant. We rcquire that y(t) must satisfy thc
constraints associatcd with the n plant equatio ns in (8.15). The n-vector >,.,
givcn houndary conditions on y(t), that is, y (t0 ) = y (t 0 ) and y (t1 ) = y(t1 ).
fun ction o time t, since the state x is itsel a unction o f t.
Thcrcfore, 11(1) must satisfy 11(10 ) = 11(11 ) = O. We indicate a typical time
Considcr the performance measure given hy
response o y( t), y(t), and 11(t) in Figure 8.3.
To obtain a necessary condition for the optimality of y (t), we first substitute
(8.17) i nto (8.16) to give J(u) = j''L(x , >. , u, t) dt
'
J(y) = f''g(y + ETJ, y+ E1J, t) dt {8.18)
' Jt is well known that the minimization of (8.14) with thc add itional constraints 111
(8.15) is cquivalent to the minimization of (8.24) with no additional constrain.
Wc thcn determine thc derivative of J( y) with rcspect to e as (l - 4]. Using (8.23), we write the Eulcr-Lagrangc cquations or (8.22) as

-di =
dt
f''[-aya
to
8
11
8
+ -a. iJ ] dt
ay
(8.19)
d(8L)
-
....:_ - 8L
-
dt ax ax
Intcgrating (8.19) by using thc parts formula of elementary integral calculus, we
ohtain d( ol) 8L
dr a~ = a>.
-di =
de
f'''[ -aya 1J -
8
a 8
-d ( - . ) 11 ] dt
dt ay
-d(iJL) -ol
i-
-
a
-
iJu
I''' f''[-aay8 - -dtd ( -.aay8 )]
dt
ag 'IJ +
= -. 'IJ( f) dt {8 .20)
ay lo
whcre van at,o ns about optima( values have been includcd for thc stalt ,
Lagrangc multipliers >., and plant input u. Evaluating the cquations in (8 2'iJ
whcrc the first tcrm is z.ero because 11(1 0 ) = 11(t ) = O, leaving o nly the integral
 lntroduction to Optimal Control 467
466 State Variable Design Using Optimal Control Theory
equations as
these optima! values (denoted by the superscript ) yields
x = -3x* - sx
. ag
- },,. = - 1 +{ -ar) TI >,,. X= -2x + 3X* '(8.31)
ax ax
Using any of the state transition matrix methods of Section 4.4, we determine thc
x = r(x , u, t) unforced solution of the equatioos in (8.31) as

= ( !e- s, + }es, {e- s, - ies') ( x*(O))

ag 1
O=-
au
ar) TI >,,.
+{ -
au (8.26)
x(t))
( X*(t) }e- s, - !es' }e- s, + 1es, A*(O)
(8.32)

The third equation in (8.26) is an algebraic equation lhat relates u to x * and >.. We are given x *(O) = x 0 and x (oo) = O as boundary conditions. The fin,t o f
The fin,l two equations are each of order II and, consequently, require a total of these is satisfied automatically in (8.32). Moreover, we may use x *(oo) = O to
2n boundary condilions. Usually, sorne of the known boundary condilions are al solve for X*(O) by first writing the component equation for x*(t), that is
t O and sorne al t1. For this reason, we refer to the probkm of determining time
solutions for the Euler-Lagrange equations as a two-point boundary value prob- x*(t) = (1e- 5 ' + fes')x(o) + {1e s, + 1e 5 ')X(O) (8 .J3)
lcrn.
As 1 -+ oo, we neglect terms iovolving e- s, in (8.33) and set ~ *(t) = O to ob1.1in

EXAMPLE 8.2 o= }e 5'x(o) - !e 5 'X(o)

Considcr the first-order linear planl described by = Bx(O) - 1X*(O))e 5' (8.34)

.x = -3x + 4u (8.27) We set the bracketed term in (8.34) equal to zero and solve for X*(O) in 1cnns of
x *(O) to yield
with boundary conditions given as x(O) = x 0 and x(oo) = O. Let the performance X(O) = }x*{O) = {x0 (8.35)
measure to be minimized be ,
Therefore, the time solutions for x*(t), X*(/), and 11*(1) are
J(u) = {(x 2 + u 2 ) dt (8.28)
o x(t) = x 0e- 5'
The problem is to determine the optima! control u that will transfer the state x x(,) = x(o)e - 5' = !xoe s,
from x 0 (al t = O) to the origio (al t = oo) while minimizing J(u).
Wc first form the functional L in (8.23) as
u(1) = -2A(t ) = -!x0 e - S 1 (8.3)

L(x, A, u, t) = x 2 + u 2 + X( - 3x + 411 - i) (8.29) The solution to the Euler-Lagrange equations in (8.30) provides the optimal
open-loop control functioo u*(t) in (8.36). We show this open-loop rcsult in
Using (8.26), we obtain
Figure 8.4a.
Seldom are we able to determine the optima! closed-loop control configu ra1ion
-X = 2x - 3X* from the calculus of variations approach. Yet for this first-order systcm, we see
.x = - 3x* + 411* from (8.36) that ~

u(x) = kx* = }r * (8 17)

O= 211* + 4X* (8 .30)
Figure 8.4b shows this linear feedback structurc.
We solve the third equation in (8.30) for 11* and then rewrile the first two
I
lntroduction to Optima! Control 411'1
468 State Variable Design Using Optima! Control Theory
"""'I
11 (1) = -! X('/! -SI X
1 and set this derivative equal to zero to obtain k equal to - } or + 2. Wc
choose k = - t to yield a stable system. The corresponding value of J in (8.38'1
is xJ/8. In summary, we have used the Euler-Lagrange etuations to determine
the optima! open-loop control function u(t) for a first-order linear system with a
quadratic performance measure. Moreover, we have determined that the optima
feedback control structure is a constant gain (k = - !) and have verified this'
(a)
optimal value of k by parameter optimization for an assumcd linear feedback
X
configuration.

8.2.3 The Linear Quadratic Control Problem

1
-2 Consider the linear, time-varying plant described by
k
( b) x = A ( t )x + b( t) u
1
04~ with initial conditions x(t 0 ) = x 0 . We use the quadratic performance mcasurc,
already given in (8.2), that is, :o 1

0.3~ J(u) = f (xrQx +

1
to
yu
2
) dt

t
.J(k) 0 .2,\
where Q is a positive semidefinite matrix, y is a positive constant. and both 10
and t are specified. The problem is to determine thc optima! open-loop control
1
function u*(t) and the optima! closed-loop control law u(x) Cor thc lineai :
quadratic control problem.
o \~ The optima! open-loop control function u(t) can b e obtained by solving thc
Euler-Lagrange equations. ldentifying g and fin (8.2) and (8.39). we may use thc
third equation in (8.26) to yield 2yu + },.Tb(t) = O, from which
o
- 3 -2 -1 o +l
1
k- 11* = - - )tb(t)
(e) 2y

Figure 8.4 Optima! (a} open-loop and (b) closed-loop Performing the indicated operations i.n the second and first equations in (8.26)
solutions with (e) a curve of J(k) versus k for Example
and substituting (8.40) for u* yields
8.2.
1
Suppose we had initially selected a linear feedback configuration u = k x with x = A{t)x - - b(t)br(t)>-.
thc intcnt of using the parameter optimization procedure of the last section to 2y
obta in the o ptima! value of k. Evaluating J(u) in (8.28) for u= kx gives
J(u) = {' [xle2(-H 4k)t + k 2x5e2(- 3+4kJ1] dt
J.. = -2Qx* - AT( t) >-.

which are the required Euler-Lagrange equatio ns.

2 This formula tion would be complete if 2 11 houndary conditions were known .
x~(l + k )
(8.38) W e havc provided x(t 0 ) = x 0 as 11 of those bou ndary conditions. lf 11 is spccik d
2(3 - 4k)
and x(t ) is free. we state wi thout proof that transversality cond itions rcq11111
1
F igu re 8.4c shows a curve o f J( k ) versus k . Wc next calcula te di/dk from (8.38)
 470 State Variable Design Using Op.timal Control Theory lntroduction to Optimal Control 471

>..(11 ) = O for optimality. If 11 = oo, this condition corresponds to x(oo) = O, as in

the scalar case considered in Example 8.2. In solving (8.41) using these boundary
conditions, we obtain x*(t) and >..(t), from which u*(t) can be obtained from
(8.40).
To determine the optima! closed-loop control law u(x) for the linear quadratic
control problem, let us assume that there exists a symmetric, time-varying,
posilive definite matrix P(t) satisfying
-------l 2 _ _ _ _...,
>.. = 2P(t)x (8.42) ----- 2------'
for ali x. Using (8.42) in (8.41) yields (a)

} T
P + Q + A 7 (t)P + PA(t) - -Pb(t ) b (t)P = O (8.43) , (1) = - e - 31 + x
'I

which is referred to as the matrix Riccati equation (1 - 4]. There are n(n + 1)/2
coupled component differential equations in (8.43), with an identical number of
unknown variables in the matrix P. The boundary condition for (8.43) is
P(r1 ) = O. Substituting (8.42) into (8.40) gives the optima! closed-loop control
law u*(x) as
----~ ~-
(b)

l T 1 .. Figure 8.5 Computer simulations of (a) the Euler-Lagrange equations and (b) the
u(x) = - - (2Px) b(t) = -- - b1 ( t)Px (8.44) optimal open-loop solution for Example 8.3.
2y y
From (8.41 ), we have
For the special case where the linear plant is time-invariant and the process is of
infinite duration, that is, 11 = oo , the II by II matrix P is constant. Since P is zero
for this case, (8.43) becomes the nonlinear algebraic matrix equation given by
x: = xr
.. 1 xr = - 2x: - 3xf - ,\~
Q + A7P + PA - - Pbb7P = 0 (8.45)
'I X1 = - 5x*1 + 2.\2
X=
2 -sx2 - .\*1 + 3.\*2 (8.48)

EXAMPLE 8.3 wilh boundary conditions x:(O) = l , xf(O) = O, .\!(11) = O, and .\~(1 1) = O.
Figure 8.5 shows a computer simulation diagram for the fourth-orda system of
Let the plant be described by equations in (8.48). Using any of the methods of Section 4.4, preferably 1h<'.
impulsed-integrator procedure for determning the resolvent matrix, wt: may
X1 = X2 obtain the state transition matrix <l>(I) to yield tht: form o f the time solution for
x2 = -2x1 - 3x 2 +u (8.48) as

r.
(8.46)

with x 1(0) = 1 and x 2 (0) = O. Suppose we select a quadratic performance x:(1) </>12 </> 13 <f>14
measure (with 11 = oo) as xf(t) = </>22 </>n </>24
rx:io))
</>21
(8.49)
.\1(1) 'Pn
J(u) = tfo 00

(5x + 5xi + u 2
) dt (8.47)
,\i (I)
</>31

</>41 </>42
</>33

cp43
'P34

</>44
.\;(o)

We wish to determine the optima! open-loop control function u*(I). where we have inserted x:(O) = l and xr(O) = O in (8.49). We determine the
r 472 State Variable Design Using Optima! Control 1:"heory lntroduction to Optimal Control 4/ t

characteristic equation for the closed-loop system in Figure 8.5a as , - - - ----~1

1 1
r----- - ---------- - , 1
1 1 ( ) 1 + J:2 X( f
1 u I l
s
4
- 10s 2 + 9 = O (8,50) 1 1 -
1 1
1 1
which yields roots of - 1, + 1, - 3, and + 3. Therefore, each element cf,;1 of the 1 1
statc transition matrix in (8.49) has the form : Control law :
'- - - - - - 1 ~ 1
4';1(1) = K 1e- , + K 2 e+1 + K 3e- 3' + K 4 e 3'
~ - - - - - - - - - -'- - ~~' - - - J
(8.51)
'-~ 1 1
Wc cxprcss the component cquations for At(t) and A;(I) in (8.49) and then use
thc remaining boundary conditions At(oo) = A;(oo) =Oto obtain Af(O) = 6 and
A;(O) = l. The resulting time expressions for xt, x, A!, and A; are Figure 8.6 The linear plant and optimal feedback control law for
Example 8.3.
xt( I) = 1.5e- ' - 0.5e- 3'
Simplifying (8.55) gives
x!(t) = -1.5e- , + 1.5e- 3'

3'
2.5 - 4p,2 - 2pf2 P11 - 3p ,2o,Lo 2p22 - 2p12;22) = ( ~)
A(t) = 7.5e - ' - 1.5e - ( P11 - 3P12 - 2P22 - 2P12P22 2.5 + 2P12 - 6p22 - 2P22 O

>.;(,) = e-J, (8.56)

(8.52)
1
from which
U, ing (8.40). wc obtain the optima! open-loop control u(t) as

u(t) = -A;(t) = -e- 3' (8.53)

P = (o\ 0.5)
0.5
(8.57)

which is shown in Figure 8.5b. We observe that P is positive definite. From (8.44) and (8.57). the optima) f
In solving for the optimal closed-loop control law using the matrix Riccati closed-loop control Iaw is given by
cquation, we first identify
1
u(x) = - -(O 1) ( 3 0.5)(X1)
0.5 X2
A= (
-2
O _;) b = (~)
0.5 0.5

= -xi - X2 (8 .58)
Q = (25 'Y = 0.5 (8.54)
/5) Figure 8.6 shows a computcr simulation diagram for the linear system in (8.46) \
with its optimal feedback control law in (8.58). A comparison may be made with
Using (8.54) in (8.45) gives the simulation diagram in Figure 8.5b, which provides an open-loop control
function u(t) that is valid only for the given initial conditions x 1(0) = l and
x 2 (0) = O. On the other jland, the closed-loop control law u*(x) in Figure 8.6 i~
( 205 /5) + (~ -2)(P11
-3 P12
P12)
P22
+ ( P11
P12
P12 )( O
P22 -2 -~) optima! for ali initial conaitions for the state vector x. Moreover, as demonstratcd
i~ Section 3.4, noise disturbance effects are minimized with a feedback configura /!
oon. .
1 ( P 11
0 .5 P 12
P12)( O
P 22 1
){o 1)( P11
P12
P12) =
P22
(O ~) (8.55)

1
 ,ff4 State Variable Design Using Optima! Control Theory
lntroduction to Optimal Control 475

EXAMPLE 8.4
1 Simplifying (8.62) yields

l
We retum to the design problem of Example 8.1, where we used parameter
uptimization to adjust a tachometer gain K, for suboptimal control. We now
1 2
1 - YP12 P11 - P12 - YP12P22 '
want to determine the optima) feedback control law u*(x) obtained by feeding = (g g) (8 .63)
1 ' 1 2
Pu - P12 - 2p,i - 2JJi2 - - P22
hack, through appropriate gains, both state variables for this second-order plant.
We can apply (8.45) to solve for the matrix P and then obtain u*(x) from (8.44). 1P12P22 y

Using the direct phase variable form shown in Figure 8.1, we express the state
variable equations as from which we obtain

X = Xi
p= (/y+ 2/y {y ) (8.64)
i = -xi+ u (8.59)
{y - y+ VY2 + 2/y
2

We again use the performance measure defined earlier in Example 8.1 as

It is again easy to show that the matrix P is positive ddinite as r..:quirc.:d . From
(8.64) and (8.44), we determine u*(e) as

J(u) = f(e 2
+ yu 2 ) dt (8.7)
u(e) = - ~br/>e
y
where the error e is the difference between the unit step input [ r( t) = 1J and x 1 , 1
that i:,, e= 1 - x 1 Since J(u) is in terms of e and not x, we must have = vY e + ( - 1+ /1 + 2/{i )ez {8 65)
differential equations describing e. Letting e 1 = e and e 2 = , we now define a
new set of state variables as
Figure 8.7 shows the optima) closed-loop controller configuration as a fum:tion of
the performance measure parameter y.
e1 = e2 We may compare the results from Example 8.1 for the suboptimal controller
configuration of Figure 8.1 with corresponding results for the optima! feedback
2 = -e2 - u (8.60) control law u*(e in (8.65). We determine the mnimum value of J(u), denoted as
J , as y + 2{i . Curves of J versus y are shown in F igure 8.8 for optima! ami
We obtain the second equation in (8.60) by calculating e1 = - xI from e 1 = 1 - x 1
suboptimal feedback structures. The optima! feedback control law has both e and
and then identifying ei as -xi and 2 as - x2 , which from (8.59) is equal to
- (-x 2 + u) or -ei - u. fed back through gains identified in (8.65), while the suboptimal controller
utilized Cor feedback only through an adjustable tachometcr gain K,. We
Using (8.60) and (8.7) with e = e 1 , we identify
observe that J for the optima} controller is lower for ali valucs of y except
y = 1, for which the two controllers are identical.
A= (g -n b=(_~) Q = (~ g) (8.61)
r(I); l +
X1

Using (8.61) in (8.45) gives

(~ g) + (~ O)( P12
-1
P11 P12)
P22
+ (P11
P1i
P12 )(
P22
O
O -n ~
-1 + ~ I \y

_ ~( P11
y P12
P12 )(
P22 - 1
O}(o -l)(P11
P12
P12) = (O
O g) (8.62)
P22
Figure 8.7 The optima! feedback control law for Example 8.4 .
 476 State Variable Oesign Using Optima! Control Theory Oesign Comparisons 477

y
r(/) +
1
1
1
1
1
___ _ ___ J1

t
J'

(o)

V
r(I)
1
1
1
1
~- ----.O.:>. . . .___ _1.0_,______1...__
o ~

5 2 1~ 1 1:
_J
y- ------
Figure 8.8 Comparisons between the optimal
solution of Example 8.4 and the suboptimal solu- (h)

tion of Example 8.1.

Figure 8.9 Development of linear SVFB control for Example 8.5.

examples: the first compares optima! control and SVFB control (pole placement).
and the sccond compares optima! control with series compensation.

8.3 DESIGN COMPARISONS

EXAMPLE 8.5
Kalman (5) considered the inverse optimal control problem o determining when
a linear closed-loop system is optima!. The matrix A and vector b completely Let the transfer unction of a linear plant be given by
describe the linear plant, and the matrix P and scalar y define the optima!
feedback control law in (8.44). If these descriptors of the linear plant and 1 l
feed back control law are specfied, we may solve (8.45) for the weighting matrix Q
G(s} = (s + J)(s + 2) = s + 3s + 2
2
to determine the performance measure for which this closed-loop system is
optima!. These remarks also apply to thc time-varying case, exccpt that the matrix
Q. wh1ch may also be time varying, may be determined from (8.43). Therefore, we
We are to design a linear SVFB control law as shown in Figure 8.9a using thc
conclude that ali linear systcms are optima! according to sorne quadratic perfor-
direct phase variable f(jrm of state variables such that the resulting closed-loop
mance measure to be determined from either (8.43) or (8.45).
poles are placed al - 1 and - 3. Moreover, we are to determine a quadrat,c
Wc use the foregoing concept of an inverse optima! control problem in this
scction to aid in comparing results obtaincd by the methods of optima! control, performance measure for which optimization thcory yields this same feedback
statc variable feedback (SVFB) control, and series compensation. We present two control law.
 Design Comparisons 479
478 State Variable Design Using Optimal Control Theory
Solving (8.73) for Q gives
Using the procedures developed in Section 3.6, we obtain a closed-loop transfer
function for Figure 8.9a as
Q= (
5y 6y -:
5y
Pu) (8 .74)
Y(s) K
6y - Pu
'
(8.67) We may select p 11 = 6y in (8.74) without loss of generality. We then obtain
R(s) = s + (3 + Kk 2 )s + (2 + Kk 1 )
2

5y o) ~ (8 .75)
Requiring pole placements at -1 and - 3 yields a desired closed-loop transfer Q =(o 5y
function as
Therefore, we have an optima! feedback control law that is identical to the linear
Y(s) 1 I SVFB control obtained in Figure 8.9b when we select a quadratic performance
R(s) = (s + l)(s + 3) = s 2 + 4s + 3 (8.68) measure in (8.2) given by

Equating (8.67) and (8.68) gives K = 1 and

J(u) = y f 0

(5x + 5xi + u 2 ) dt (8 .76)

3 + Kk 2 = 4 We see that optimal control with a selected quadralic performance measure yidds
the same resulls as pole placement for linear time-invariant plants. Moreover,
optima! control provides a generalized formal for the control of time-varying o r
2 + Kk 1 =3 (8 .69)
nonlinear plants.
from which k 1 = k 2 = l. This linear SVFB control law is shown in Figure 8.9b.
For thc corresponding optimal control Jaw, we first idcntify for the plant
EXAMPLE 8.6

A-(
- - O
2 -31)
Consider the linear plant with phase-lead series compensation shown in Figure
8.10a. In Example 7.5 of Section 7.3, we determined values of thc parametcrs a
and T to satisfy design specifications of %OS .$ 15% and t, s 1.0 second for a
b = ( ~) (8.70) step input and ess s 0.05 radians for a unit ramp input. We pinpoint in this
~ example the general diffi.culty that prevents us from using optima! control [in
Since we want to obtain the same control law as for the SVFB control case above, particular (8.45)] to yield the same feedback control Jaw as phase-lead compensa-
we require tion.
We let a and T remain unspecified here and attempt to determnt: J(u ) in

u*(x) = - X - X2 = ( =!r(;J (8 .71)

terms of these parameters. As shown in Figure 8.10b, we define state variables as
X= X2

Equating (8.71) and (8.44), we obtain x2 = -2x 2 + Ku

1
(-1) 1(pll i = - -x 3 + (8 .77)
-1 = - y P12 P12)()
3 e
(8.72) T
1 P 22
Letting e 1 = e= 1 - x 1, e 2 = e, and e3 = e, we may express an alternatt: form
which yields p 12 = p 22 = y. Using (45), we may write of state variables as
e = e2
~

Q + (l -3
-2)(P11Y Y)+
Y
(P u Y)( O I)
Y Y - 2 -3
e2 = -2e2 - Ku
1
e3 = - e - - e3 (8.78)
-~(~)(y y)=O (8 .73) T
 480 State Variable Design Using Optimar Control Theory
Design Comparlsons '18 1

+ lloT _ K_ l_ Let us arbitrarily choose the feedback control law to be

+ 1/T -< + 2)

(a)
u(e) = 1
P ) T( el)
e2
( (1 - 1/ a)/ T e3

r(r) = J + where p can be selected such that the matrix P is positive definitc and th(I
resulting matrix Q from (8.45) is positive semidefinite. Using (8.82) in (8.45) and
solving for Q yields

(b) Q=
(
1 + 2pl)
+ (1 + E)/ K -
p

E+ /133 + Pu/T
/111
p
p

(p
1
+ 1/ K + ( / K - Pu
+ p - 2/K

+ E/ K + E/TK - Pu
E+ /133/lu/T l
(p + E{l + T)/K - PI) :
1 + 2pll/T 1

Figure 8.10 Phase variable selection for the phase-lead series compensated sys-
tem of Example 8.6.
,.
Thcrcfore, we identify A and b as
where we have Jet = (1 - 1/a)/TK. When only the output y (i.e., x 1) is
~ _)) b-(-:K)
1
-2 available for feedback, we can use observer theory (6) to form approximations forl
A= ( (8.79)
- 1 o other state variables. In this example, only x 2 was inacccssible. For higher-orde,
plants, we may need to form approximations. for severa! additionaJ state variable~
before the linear quadratic control problem can be formulatcd and thc rcsultin p,
From Figure 8.10b, we have linear feedback law obtained. Finally, we should recognize that the phasc-lcadi
1
series compensation design yields a specific optima! control law sincc the feed-
u = x3 + (1/aT)x 3 back configuration has been fixed . Toe design is also suboptimal with respect to
the J(u) resulting from (8.83).
= e - [(1 - l/a)/T]x 3

= e1 + ((1 - 1/a)/T)e3 (8.80)

8.3.1 DHlgn Procedures In Retrospect
Using (8.80) and (8.44), we obtain
We summarize in Figure 8.11 the various control system design procedure:
1 (Pu P12
Pul (- OK )
described in this and previous chapters. We identify proportional control, PJD.

r(1 - l~a)/ T) _ -
Y Pn
P12 P22
P 23
P23
p33 0
(8.81) con trol, and SVFB control as design techniques, primarily from Chapter 3, that
can be performed by both transfer functio and state variable methods. Series
compensation in Chapter 7-that is, phase-lead, phase-lag, and lag-lead coml ::I ~
Solvi ng (8.8 1) with y = 1 yields p 12 = 1/K, p 22 = O, and p 23 = (1 - lja)/ TK. pensation - depends on the transfer function procedures of root locus or Bode
Sincc p 2 = O, the matrix P will not be positive definite as required by optimal design. 0n the other Mnd, the optimization methods described in the earlicr
control tteory. Thercfore, we conclude Cor this exarnple that it is not possible to sections of C hapter 8 are based on state variable design. Multivariable contro~ ' '
determine a quadratic performance mcasure J(u) in (8.2) that yields the same system design can be accomplished by series compensation Cor noninteractin
optima! control Iaw as the one specified by phase-lead series compensation. control after decoupling or by optirnal control theory to determine the m-vector
feedback control law u(x) as described in the following section.
 482 ~IJtu Vdrt.lble Da~1911 Using Opt1rna1 Control Theory Multivariable Optima! Control 483

1
r- -- -- -- ----------------------, The Riccati equation for the multivariable case is
1
: Proportional control
Q + A7 P + PA - PBR dB 1P =O , (8.87)

1
1 whcre P is a constant matrix because A and B are not functions of time and wc
PI D conllol SVFB control are considering the infinite-time problem. We observe that (8.87) reduces to (8.45)
1
1 Both transfer funcllon and
for the single-input case (m = 1), where R- 1 =. 1/y. The multivariable optim.il
control law u(x), corresponding to the single-input optima! control law u"'(x) 111
1
i ------- ---- --r------ -------
sldle va11able des,gn
(8.44), is
1 1
1 Senes I Opt,mal
I compensat1on
I
co11trol u(x) = -R - 18 1/'x (8 .88)
1 1
1 1 Therefore, we obtain an optimal feedback control law as the solution to the linc.tr
l Oecouphng theory-
noninteract,ng control
: Optimal control
for mult1vanable
quadratic multivariable control problem.
1 I systems .
1 1 1
1 f 1
L___ Transfer function des1gn __ J_ ___ State variable design ___ J EXAMPLE 8.7
Figure 8.11 Relationships between control system design
procedures. Lel a second-order multivariable plant to be controlled be described by

X = Xi+ U
8.4 MULTIVARIABLE OPTIMAL CONTROL
.x 2 = - 4x 1 - 4x 2 + u 2 (8.89)
Consider the multivariable plant described by
as shown in Figure 8.12a. The performance measure to br.: mjnimized is
i =Ax+ B u (8.84)
0
J{u) = {' {2xf - 2x 1x 2 + 36xi +u+ ui} dt (8 .90)
where u is an m-dimensional control vector and B is an associated n by m
matrix. Lel the performance measure to be minimized be given by
Fro m (8.89), we identify the matrices A and 8 as

J(u) "" f"'{x 7Qx + u7 Ru) dt (8.85)

A=( - O l)
4 - 4
where R is an m by m symmetric, positive definile matrix. Thus, we see that
(8.84) and (8.85) define the linear quadratic multivariable control problem [J-4].
Eukr-Lagrange equations for the multivariable case in (8.84) and (8.85) yield
B=(t ~)=l (8.91)

(8.26) with the scalar u replaced by a vector u to give

Moreover, we identify Q and R from (8.90) as
x = Ax - (t)BR- B >.. 1 7

>.. = -2Qx - Ar>.. Q =(- 2

1
-1)
36
~
u= -(i) R- 1BT>,. (8 .86)
R=(1O )
l = 1 (8 .92)
with x(O) = x 0 and >..( oo) = O. The solution of (8.86) yields the optima) open-loop
control function u*(t) and associated x*(I) and >..(1). We solve for the 2 by 2 matrix P in (8.87) by substituting (8.91) and (8.92) into
484 State Variable Design Using Optima! Control Theory Aeduced-Order Observers 411'-

8.5 AEDUCED-ORDER OBSERVEAS

-uz + % The distinguishing feature o this chaptef has been the assumption that all statC'
variables are available for use in a feedback controller. In this section we show
how to reconstruct any inaccessible state variables by using reduced-order
observers (6). Toe reduced-order observer accepts as its input both the plant

'----------14.-------- . , output y and the plant input u. A further-problem is the need to guarantee that
misalignments in initial conditions between the observer variables and the corre-
sponding plant variables are quickly corrected. We show that eigenvalues of thc
(o)
observer can be selected arbitrarily to resolve this problem.
r------- -------------------- --, Let a multivariable plant be described by
1

, / _____
1 +
1
1
~~--J~_u 1
x = Ax+ Bu
y= Cx

where the r-vector y has components available for feedback. We can sclect a set of
state variables (or perform a linear transformation) that results in the elements of
1 y as individual state variables. Let the remainmg n - r statc variables form thf' .
1._ Control
_ _ _law_ .J1
elements of a vector z, where

z = Ex
1 ~
(b )
Using (8.96) and (8.97), we can forro
Figure 8.12 Multivariable optima! control for Example 8.7.
y = F1y + F2 z + G 1u
(8.87) to o btain
i = F3 y + F4 z + GiU
{
l --
2
1
- 1) ( O
36 + -4
1r (P12
- 4
Pu
P22 -!)
P12) + ( Pu
Pu
P12 )( O
P22 - 4 Let T be an arbitrary n - r by r constant matrix. Defining w as the difTerence

~)( n-1( or(Pu

between z and Ty, we may express w from (8.98) as
) = ( i)
- ( P11
P,2
P12 )( 1
P22 O 1 P12 P12
P22 O
(8 .93)
w = z - Ty
Solving (8.93) gives
= (F4 - TF2 )z + (F3 - TF1 )y + (G 2 - TG 1)u
p = (
- 1
3 -~) (8.94)
.1 = (F4 - TF2 )w + (F3 - TF1 + F4 T- TF2 T)y + (G2 - TG 1 )u
Using (8.91 ), (8.92), and (8.94) in (8.88) yields the multivariable optimal control
as 1 Since z contains state variables that are not accessible, we do not know z(t 0 ) and,
u*(x) = - R - BTPx 1 1 hence, w(t 0 ). However, we can select T suh that the eigenvalues of F4 - TF2 in

-( ~ria )T( 1
J
-1 -1)(;:) 1
(8.134) yield a sufficiently fast settling time for the observer state w. Once (8.99) is
solved for the approximation w(t), obtained by using estimated initial conditions.
we can then forro an approximation z for the inaccessible state variables in z as

=( -JX1+X2)
X - Jx 2
(8.95) i = w+ Ty (8.100),

l
Figure 8.12b shows this optima! feedback controller. Finally, we note that the combined plant-observer is of order n + (n - r )
486 State Variable Design Using Optimal Control Theory Aeduced-Order Observers 487

2n - r, where the eigenvalues of the (n - r)th-order observer are selected arbi- Setting w =z- Ty, we write
trarily.
w=i - Ty

EXAMPLE 8.8 = ( - 2z + Ku) - Tz

Consider the linear plant in Example 8.6 described by = (- 2 - T) z - ( - 2 - T) Ty + ( - 2 - T) Ty + Ku

X= X2 = ( - 2 - T)w - (2 + T)Ty + Ku (8 . l04)

x2 = - 2x 2 +Ku Figure 8.13a shows the combined plant-obscrver with the appro ximation for x 2
formed as in (8.100). The student is asked in Problem 8.18 to verify that 1he
y= X {8.101) transfer function from u to x2 is indeed K/(s + 2). Finally, Figure 8.13b show)
,. the closed-loop feedback systcm resulting when the optimal feedback control law
Recall that in Example 8.6 we compared the series compensation design shown in in (8.82) is applied by using x I from the plant and x2 from thc obscrvcr.
Figure 8.10 with a feedback controUer obtained by optimization theory. A basic
restriction in the optima! controller design was that the state variable x 2 was
inaccessible. Therefore, Jet us define EXAMPLE 8.9

z = X2 (8.102) Suppose only x I is accessible in the multivariable plant of Example 8.7 described
by
an<l thcn determine a first-order observer to yield an approximation .x 1 for x 2 .
X= X2 + u,
Cormponding to (8.98), we use (8.101) and (8.102) to form

y=z
x2 = - 4x 1 - 4x 2 + u 2

y= X (8.105)
i= - 2z+Ku (8 .103)
We set z = x 2 and form a first-order observer to yield the approxima tion x 1
Corresponding to (8.98), we form

y = z + u,

= - 4y - 4z + U2 (8 .106)

Again, with w = z - Ty, we have

w = i - Ty
= (- 4y - 4z + u 2 ) - T(z + u 1 )

= (- 4 - T)z - 4y - Tu 1 + u 2

= ( - 4 - T) z - ( - - T) Ty -t ( - 4 - T) Ty - 4 y - Tu 1 -t u 1

(b>
= (- 4 - T)w - (4 + 4T + T 2 }y - Tu 1 + u 2 (8 .107)

Figure 8.13 Reduced-order observe, for Example 8.8. Figure 8.14a shows the combined multivariable plant and fin,t-or<ler oh~i:r,c.:r
 References 489
488 State Variable Design Uslng Optima! Control Theory
feedback controllers for operations in environments having significant noise
disturbances [10). When only random time curves of system inputs and outputs
are available, it becomes necessary to perform system tdentification by using
techniques based on either continuous-time functions r discrete-time series
~ 1 t t 14 + T +
analysis (11-13). Finally, decentralized control algorithms are being studied or
the hierarchical control of large-scale systems [14,15) and adaptive control
algorithms for systems having models th.at are only partially known (l 6, 17).
.___ _ _ _ 4 1 lnterested students are encouraged to pursiie these advanced topics in the currcnt
'-------+-tT------
(c,)
technical literature [7-17).

SUMMARY
) 1 , 1 T 1-----"'
In this final chapter, we looked at design procedures based on the modero state
variable approach for linear plants with complete state feedback. Initially, we
.r =y.----- reformulated the design problem in terms of the minimization of a performance
ll,ul4+T+
measure and then derived the Euler-Lagrange equations as necessary conditions
ti for optimality. We showed that the solution of these equations yields the optima)
open-loop control function u(t).
We considered the linear quadratic control problem in which the plant is linear
1 1 T 1 and the performance measurc is the integral o a quadratic function of the plant
state x and input u. We derived the matrix Riccati equation and showed that the
'--------------13i-...------------.... optima) closed-loop controller u(x) is a linear function o the plant statc
(b) variables. Morcover, for the infinite-time problem, we showed that the feedback
gains are constant when the plant is a time-invariant system. We presented the
Figure 8.14 Reduced-order obsetver for Example 8.9. extension of these optimal control results to the multivariable case.
In making design comparisons, we showed the equi_valcnce between SVFB
Using this observer output as an approximation for x 2 , we show the optimal control (pole placement) and optimization theory for linear plants by constructing
control law of Examplc 8.7 in Figure 8.14b. the quadratic performance measure that is minimized by a given SVFB control
law. However, since series compensation techniqucs require only the plant output
for feedback, we are unable to determine thc corresponding quadratic perfor-
mance measures for optima! control directly. Using a reduced-order observer to
approximate inaccessible state variables permits us to gcnerate estimates of the
8.6 ADVANCED TOPICS IN CONTROL complete state for feedback and, hence, to complete the desired comparison. We
concluded with a discussion of selectd advanced topics in control thcory.
Thc introductory nature of this book prohibits us from including details of a
number of advanced control systems concepts. However, we can mention these
advanced topics briefty and encourage their study in research programs or in
further courses, perhaps at the graduate level. First, the need for work in ~ '
REFEUNCFS Y. . .1. /1'1 . ti ,., ')
microproccssor control has increased dramatically in recent years because of the
rapid expansion in the development and use of microprocessors and minicom- l. D. E. Kirie. o)timal Control Theory. Englewood Cliffs, N.J.: Prentice-Hall,
puters [7,S,. Robotics has become another extremely popular arca for control
1970.
systems applications that feature design including electromechanical sensors, 2. B. D. O. Anderson and J. B. Moore. Linear Optima/ Control. Englewood
plants, controlJers, and actuators (9). Stochastic optimal control is an arca that
Cliffs, NJ.: Prentice-Hall, 1971.
focuscs on the combined design of state variable estimators and associated
490 State Variable Design Using Optimal Control Theory Problems 491

.l. M . Athans and P. L. Falb. Optima/ Control: A11 lntroductio11 to the Theory
ami l ts Apphcatio11s. New York: McGraw-Hill, 1966.
1 performance measures. Evaluate J( 11*) and explain the practica!
implication of minimizing J( 11) in each case.
4. A. P. Sage and C. C. Whi te, 111. Optimum Systenu Control, 2nd ed. I
a. J(u) = {"'u 2 dt
Englewood Clitrs, N.J.: Prentice-Hall, 1977.
S. R. E. Kalman. "When Is a Linear Control System Optima!?" Transacrions
ASME Ser. D: J . Basic Eng., Vo l. 86 ( March 1964), pp. 1- 10.
b. J(u) = {tu 2
dt
6. D. G. Luenberger. lntrod11ctio11 to Dynamic Systems: Theory, Models , and
Applications. New York: Wiley, 1979.
c. J(u) = l 00
((2 - x) 2 + u 2 ) dt
7.

8.

9.
G. F. Franklin and J. D . Powell. Digital Co11trol o/ Dy11amic Systems.
Rcading, M ass.: Addison-Weslcy, 1980.
U. C. Kuo. Digital Comrol Systems. New York : llolt, R.inehart and Winston,
1980.
Spcial Issue of the Proceedings o/ the IEEE. "Robo tics and Factories of the
Futun:" (July 1983).
'1 (8.2) 8.2 Let a linear plant be described by

with x(O)
x= - x +u
= 1. lf the performance measurc is given by

10. P. S. Maybeck. Stochastic Models, Estimation, and Co111rol, Vol. l. New

York: McGraw-Hill, 1979.
J(u) = l 00
(x 2 + 11
2
) dt

11. D . Graupe. lde111ification o/ Systems, 2nd ed. Huntinglon, N. Y.: Krieger,

express the Euler-Lagrange equations ami idcntify tht: ncccs-,ary
12.
1976.
T . Kailath (guest editor). "Special Issue on System Idcntification and 1
'I
boundary conditions. Solve these equations for the optimal control
Time-Series Analysis." IEEE Transactions 011 A11tommic Control, Vol. 11*( I ).
A<..'- ! 9, No. 6 (Deccmbcr 1974). 1
r, (8.2) 8.3 For the plant and pcrformancc lllasurc in Probkm 8.2, kt II J..,.
13. G. E. P. Dox and G. M. Jenkins. Tune Series Ana/ysis, Forecasting and
Control. San Francisco: Holden-Day, 1970. 1
1
Sketch the curve o J(k) versus k. Compare the value of k ohtj1ned
by setting dJ/dk = O with the one obtaind as the ratio of 1/'.. (t) to
14. M. Athans (guest editor). Special lssue on .. Large-Scale Systems and '
1 x*(t) in Problem 8.2.
Decentralized Control." IEEE Trt111sactio11s 011 Automa1ic Control, Vol.
AC-23, No. 2 (April 1978). (8.2) 8.4 Repeat the instructions of Problcm 8.2 for the plant dt::.cribcd hy
15. A. P. Sage. Methodology o/ Large-Scale Systems. New York: McGraw-llill,
1977. ~ x= 2x + u
16. l. D. Landau. Adaplive Control-The Model Reference Approach . New '
York: Dekker, 1979. with x(O) = 10 and thc performance mcasure given by
17. G. C. Good win and K. S. Sin. Adaptiue Filtering Predictio11 ami Comrol.
00
Englewood Cliffs, N.J.: Prentice-Hall, 1984.
1(11) = fo (2x
2
+ 11 2) dt

'
PROBLEMS
( 8.2) 8.5 For the linear plant in F.xample 8.2 dcsnibcd by

8. 1 Lct the closed-loop system in Figurc 8. 15 have x(O) = O. Detcrminc

the optima! feedback gain k that minimizes each of the following
x = - 3x + 411

-
r(I) ~ 1 + X

~
with x(O) = x 0 , what infinite-time quadratic performam:c mca.,ure
is minimized with negative unity feedback (k = - 1)? In otha
words, fin~ y such thar

!
0

= {' (x 2 + yu 2 } dt
"- --~ J(u)

Figure 8.15 The closed-loop system for

is minimized for k = - 1.
optim1za11on in Problem 8.1 .
d>
 492 State Variable Design Using Optima! Control Theory Problems 493

(8.3) 8.10 Let a linear plant be dcscribed by

..----~ 4 1------,
i 1 X2

u + _L. 1 r
i2 -12x 1 - 7x 2 + u
+ 1

Find the quadratic perfonn~ce measure lhat is'minjmized by the

optimal feedback control given by

Figure 8.16 The linear plant of Probfem 8.6. u(x) = - 2x 1 - x 2

for arbitrary initial conditions on the stale x( t).
( 8 2) R.6 Express (but do not sol ve) the Euler-Lagrange equations for lhe
linear plant shown in Figure 8.16 with a performance measure given (8.3) 8.11 Repeat the instructions of Problem 8.10 i lhe plant is described by
by X= X2
10
J(u) = { (x? + 2x + 4u 2 ) dt X7 = X3

i 3 "" - 6x 1 - llx 2 - 6x 3 + u
( 8.2) 8.7 Let a second-order linear plant be described by . .JI

and the optima! feedback control is

X - X1
u(x}..: -x 1 - x 1 - x 3
i 2 = -3x 1 - 4x 2 + u
( 8.3) 8.12 Find the quadratic performance measure that is minimizcd if thc
with x 1(0) = 1 and x 2 (0) = O. lf the performance measure is given linear plant in Problem 3.29 has an oplimaJ feedback control
:l/; specified by the SVFB control determined there.
( 8.3) 8.13 What first-order linear plant is being considered if the eedhack
J(u) = ((x; + x? + u 2
) di con trol
u(x) = -5x
solve the Euler-lagrange equations to obtain x*( t ), ~ ( t), and u( t).
minimizcs the performance measure givcn by
( R.2) 8.8 Determine optimal feedback gains k 1 and k 2 for the plant and
performance measure given in Problem 8.7.
~ J(u} = fc x 2 + u 2 ) dt?
( 8.2) 8.9 Show that the time-varying feedback control law given by
Assume the initial slate x(O) is arbitrary.
u(x, t) = - [tanh{l - t)Jx
( 8.3) 8.14 What second-order linear planL expressed in direct phase variable
is obtaincd from the solution to (8.43) when the linear plant is form, yields the feedback control
describcd by
u*(x) = -x 1 - 2x 2
x=u 10 mjnimize
' and the performance measure to be mirumized is given by 10
J(u} = { (x; + x? + 2u 2 ) dt?
o
2 2
J(u) = f{x + u ) dt
Assume an arbitrary irutial stale x(O).
o
494 State Variable Design Using Optimal Control Theory Problems 495

( 8.5) 8.20 Consider the linear plant described by

11 +
x1 = - 3.x 1 + u ,
x2 = - 2x 1 - x2 + u
y = X
'--------- 4 !-+------___.
Show that it is impossible to construct an observer to approximate
Figure 8.17 The linear plant of Problem 8.17 for which a the inaccessible state variable x 2 Why?
reduced-order observar is required.

( 8.4) 8. 15 Determine the optimal feedback control that minimizes

J(u) = [~,(x~ + Xi+ u+ ui} dt

o

if the linear multivariable plant is given by

x1 = -2x 1 +x 2 +u 1
x2 = - x 1 - x 2 +u 2

Assume an arbitrary initial state x(O).

( 8.4) 8. 16 What quadratic performance measure is minimized by thc feedback
control

- x1 - 2x 2 )
, u*(x) = ( - 2x - X2

for the linear multivariable plant given in Problem 8.15?

(8.5) 8. 17 Determine a first-order observer to approximate x 2 for the linear
plant in Figure 8.17. Let this observer have an eigenvalue of -4.
~
Show that any misalignment in initial conditions between x 2 (0) and
.x 2 (0) decays as e - 4 '.
( 8.5) 8. 18 Yerify that the transfer function from u to x2 in Figure 8.13a is
given by

X2 (s) K
U(s) = s + 2

Moreover, show that .x 2 is not controllable from the input u.

( 8.5) 8. 19 Let y= x 1 for the linear plant given in Problem 8.11. Determine a
second-order observer to approximate x 2 and x 3. Lec both eigenval-
ues of chis reduced-order observer be located al - 2.
 Answers to Selected Problems 497

1.14 ML 2d 1 8/dt 1 + Bd8/ dt + MgLO = f(t)L

1.15 a. True
b. False- linear system stability does not depcnd on thesystem's input.
c. True (for perfect sensors)

ANSWERSTO Chapter 2
2.1 New data provide an imparta) test for model validation purposes.

SELECTED ,11
2.2

2.3
a. K/ ((1 + R 1C1sXl + R 2C2 s))
b. 1/ (1 + R 1C1s + R ,C2 s + R 1C2 s + R 1R 2C 1C2 s 2 )
KR if(R 1 + R 2 + R 1R 2C(l - K)s]

PROBLEMS 2.4 - [M2 s 2 + (8 1 + B 2 )s + (K 1 + K 2 ))/~(s), (s) .., [M 2 s 2 1- ( 8 1

B 2 )s + (K 1 + K 2 )J[M1s 2 + B 1s + K.J - (B 1s + K 1)
2
+

2.7 a. l /( J s 2 + 8s + K), where J = 1 1 + (N1/N2 ) 2J 2 ,

B = B 1 + (N 1/N2 ) 2B2 , and K = (N1/N2 ) 2K 2
b. I /( Js 2 + 8s), where J = 1 1 + (N1/N2 ) 2J2 + (N1/N3 ) 213 and
Chapter 1 B = (N1/N2 ) 2B 2
1.1 Natural systcm with inhercnt feedback (b, f); natural system with human- 2.10 O. l/ [s(5 + O.lsX0.25s + 0.12))
devised controller (e, g); human-devised system with human-deviscd con- 2.11 0.2/[s(5 + 0.2sX0.2s + 1) + 0.04)
troller (a. c. d. h). 2.13 5.3/(s((5 + o.1sxo.25s 4 0.12) + 1) + 5.3}
1.2 d. Refcrring to Figure 1.2b, the plant block contains the ship dynamics, 2.15 [G 1(1 + G 3 ) + G 3 + G 1G2G3 )/(l + G 1 + G3 + G 1G3 )
sensor(s) are direction sensors, the controller is a combined comparator 2.16 (AG + A)/(1 - GH - AC - AB - A8H + ACGH)
and ruddcr actuator. the system input is the desired course, and typical
disturbances are unexpected currents. 2.17 (G 1G 2G 3G~ + G 1G2G5 + G1G3G4G6 + G 1G5 G6 )/(s), where (s) = 1 +
G 1G2 H 1 + G1G 2G3G4 H 4 + G4 H 3 + G 1G3G4 G6 H 4 + G 1G5G6 H4 + G2G3 H 2
1.3 a. Rcerring to Figure 1.2a, the open-loop controllers can be identified as + G 1G2GsH, + G1GiG)G6H1H2
timers and the plant as the lighting devices with light intensity (bright-
ness) as the plant output. 1
2.19 a. Define six nodcs (l to 6) for thc associated signa! flow graph. Node 1 is
,# the input node and has a unity-gain branch outgoing to Node 2. Node
b. Referring to Figure 1.2b, the system input is a setting for the desired 6, the output node, has a unity-gain branch incoming from Node 5.
light intensity, the controller sensors are optical sensors and compara- Other unidirectional branches have gains: F(2 to 3), P(3 to 4), C(4 to
tive devices, the plant is composed of the mercury vapor lamps with 5), A(3 to 2), G(4 to 3), 8(5 to 4), and D(4 to 2).
light intensity (brightness) as the output. b. Define seven nodes (1 to 7) for the associated signa) flow graph. Node 1
I.S Rcferring to Figure 1.2b, the system input is the desired drilling mud
density, a controller/sensor senses mud density and adds barite, the plant
.! 1
is the input node and has a unity-gain branch outgoing to Node 2.
Node 7, the output node, has a unity-gain branch incomjng from Node
is the mud circulation system, and the output is the actual drilling mud 6. Other unidircctional branchcs have gains: A(3 to 4), C(4 to 5). E(5
density. \ 1
to 6), 8(4 to 3), G(4 to 3), D(5 to 4), F(6 to 5), and two unity-gain
1.13 a. Nonlinear, continuous time, time invariant, detcnninistic branches (2 to 4 and 3 to 5).
b., Linear, continuous time, time invariant, stochastic 2.20 a. (v; - vc,)/R 1 = (ve, - v0 )/R 2 + C2 dvc/dt
c. Linear, discrete time, time invariant, detenninistic (ne, - Vo)/R 2 + e, dvc/dl = 0
d. Nonlincar, continuous time, time varying, deterministic v, - v0 = Ve,

n e
 498 Answers to Selected Problems Answers to Selected Problems 499

- 1/(R 2C1 ) - l/{R2C1 ) ) Chapter 3

b. A = ( - l/(R 2C2) - (l/(R 1C2) + l/(R2C2)] 3.1 a. Steady-state errors
1/( R 2C1) ) ( - 1) b. Steady-state errors, time response shaping
b= ( l/{R 1C2 ) + l/(R 2C2 ) e= O c. Steady-state errors, system scnsitivity
2
d. Steady-state errors, bandwidth considera1ions, relative stabilily
s + [1/(R 1C2 ) + 1/{R 2C2 )]s + l/(R 1R 2C1C2 )
c. 3.8 a. O, 10,0,0
s 2 + [1 /(R 2C1) + 1/(R 1C2 ) + 1/(R 2C2 )]s + l/(R 1 R2C1C2 ) b. 2, oo, 00, 20

2.21 a.
. _(-(R
X -
1 + R2 )/(R 1R2C1 )
( )
- l/(R 2C1 ))x c. l. oo, 80, O
d. 2, 00, 00, 1
-1/ R 2C2 -l/ (R 2C2 )
3.9 a. f., O, O, O
(R 1 + R 2 )/(R 1R2C1)) b. oo,0.1, oo,2
+( u
l/(R 2C'2) c. oo, O, {J/80, O
y= ( =!)\ + (l)v, 3.10 a.
b.
No, poles on jw axis
- 10
b. Define a 0 = l/(R 1 R 2 C1C2 ) and a1 = a 0 (R 1C1 + R 1C2 + R 2C2 ). c. No, polcs in right half of 1hc J-planc
V0 (s)/V,(s) = s 2/(s 2 + a 1s 1- a 0 )
d. o
x= (-2 -JJx +(~)u
0
3.11 1

y= ( - a~
- a )Tx+(l)v, 3.12 a. O
b. 1

2.23 a. A =
(
O
O 1
O ) b=m
1
e ~ ('~)
3.14
3.17
- cTAb,0, 0
a. - 2a(s(s + a) 2(10s + 1))/{[s(s + a)2{10s + 1) + K(s I l)j( s ~ id }
- 6 -11 -6 b. aKs(l + 2s)/{(s 2 (s + IOXI .as)+ ,'."(l t 2s)](l t iu)}
)
b. A =
(
- -6
11
- 6

- O
1 O
1
O
O
1
O

)' -b-(i)
b= L~l c=m
3.18
3.23
3.25
- aK/[(K + l)s + (3K + aK + 1))
(2s 2
a.
+ 2rw,,s)/(s 2
1/ (15kp)
+ 2{w,,s + w,~)

c. A = -2 O b. O, assumng asymplotic stabili1y

( e= ( - ~)
O O -3 3.28 a. (1). Y(s)/R(s) = Ks/(s 2 + (7 + Kk 2 )s + ( 12 + Kk 1)J
- 3 2 5) c. (1). Y(s)/ R(s) = Ks/ [s 2 + (7 + Kk 1 1 Kk!)s 1 (12 + 4K/.. 1 +
d. A =
(
O
O
-2
O -1
5
b=m c=m ~
3Kk 2 ))
24) .
2.26 !vi= (b, Ah, A 2 b) = 41
(
2
5
10)
7 , det M = 18, controllable
3.29 K = ~. k = (2: ,where x 1 = 0,./4, x2 = x1 and x3 = ~1
O l 2
3.30 k* = pTk
Q = (e, Arc,(AT)2c) 1 - 4 =(
O 1 -4
13 , det Q = O, not obscrvabk
-5 15 -45
2.28 Y(s)/ U(s) = (s 3 + 6s 2 + 13s + 3)/(s 3 + 5s 2 + 6s + 1) Chaptcr 4

2.30 a. X = (_~ _~ )x + (~)u, Y = (ir x 4. 1 Curve A: 35%, 0.4 sec, 0.4 sec, 2.5 sec
Curve 8: 0%, 2.1 sec, 0.7 sec, 2.7 sec
b. x* = ( - ~ - ~ )x + ( nu. y = ( - i) \. 4.4 a. Underdamped, 0.1, 1
b. Overdamped, 2, l O
c. p = (- 21 -1)3 c.
d.
Underdamped, 0.5, 10
Cri tcally damped, 1, 2
 500 Answers to Selected Problems Answers to Selected Problems 501

4.5 8. 0.729, 3.16 sec, 72.9% 5.2 a. Integral in (5.3) equats }

c. 0.163, 0.36 sec, 16.3% b. 1/[(s + lXs + 2)), potes at -1 and - 2
4.6 8.
b.
1 - l.OQ5e- O.lr sin(0.995t + 84.3)
1 + l.077e-2.68, _ 0.077e-31.J2,
c. A =( _~ -! ). eigenvalues o A are -1 and -'2

c. 1 - l.15e- 05' sin(8.66t + 60) S.3 a. Unstable: 2 RHP poles, 2 LHP potes
d. 1 - e - 21 - 2te- 2 ' b. Unstabte: 2 RHP potes, 3 LHP potes
4.7 O :s; KA :s; 0.093 c. Unstable: 2 RHP potes, 3 LHP pot~s (includes a quad)
d. l)nstabte: 1 LHP pole, 4 jw-axis poles (doubles)
4.8 8. 35.1%
b. 2.86 5.4 2, K > 1
c. 3.74,4.81,7.81 5.5 a. No, fals Hurwitz test
4.9 89.5%, 0.068 b. Yes
c. 1< K< 5
4.10 0.0612
5.6 a. K > 3.39
4.12 8. 72.9%
b. - 0.5 < K < 3.31
b. 30.9%
c. 35.6% 5.7 K > 505
4.13 a. IOe -' + 2oe - + 50e-
21 3' 5.8 a. K > 32/3
b. 3 + 2e- 101 - i b. 32/3, /5/3
c. 5e - 21 - 2./5 sin (2t - 63.4) (l, 5.9 K > 1 - 2T, T <
4.14 a. Poles: O, - 1, - 2; residues: 1, - 5, 0.8
5.10 a. K*=30,w*=./5,o 1 =-2
b. Poles: - 3, - 2; residues: 2, 5
c. Poles: O, - 2 + j3, -2 - j3; residues: l , O;l/150,0.1/-150
b. K* = 2, w = ,fi
4.15 b. l - l.21e - 559'sin(8.29t+56.0) c. K* = 30, w = ./14, o, = 1
4.16 a. 1 - l.512e - 7 5 ' sin(6.6It + 41.4) 5.11 For O< K < 6: 3 RHP poles, 6 < K < 11.4: 2 RHP pole.~. K > 11.4: no
h. 1 - J.068e - 75' sin(6.61t + 110.8) RHP poles
4.17 a. 2 5.13 K* = 90/7, w = y20/7, o 1 = 3.5
b. 4,6 S.14 a. No jw-axis crossings
c. /1 1 can be any value except 1.5, /l2 = 6 b. K* = 6, w = .fS, o 1 = -0.5
4.18 a. 4 S.JS b. O
b. 10, 12 c. 3/2
c. Any /1 1 and /12 such that 4/1 1 + 3{J2 = 6 S.16 Sg = 3.15, O= 3.5
4 19 l 3e - 2, - 2 e - 3, e - 2, - e - 3t )
~1
( ) ( -6e - 2 ' + 6e- 3 ' -2e - 21 + 3e- 3'
S.17 K* = 2.5, w = ./5
4.29 6.25 - 16.67e-' + 12.Se - 2 ' - 2.08e- 4 ' S.19 a. .-Basic
b. Modified
4.30 ae 2 ', where a= (4/3)e 2/(e 4 - 1)
c. Modified
4.31 Yes (det M = - 1), yes (det Q = -2) S.20 K* = 2.4, w = v'If
S.21 a. ", = t
b. x = 4
Chapter 5 ' c. "'. = {6
.. - 18.4
5.1 a. Asymptotically stable
S.24 K < 5, w - v'ff .
b. Marginally stable
c. d. Unstable S.26 x - 6, w - /i and x - 8, "'. - O; s a = 1 + .fS, 1 - .fS
 502 Answers to Selected Problems Answers to Selected Problems 503

5.27 a. K* = l + ti.
w* = /iti -
l Chapter 7
b. K* = - 2, w* = O
7.1 Os K s 19.2, K s 200
5.32 No jw-axis crossings (except al origin for b = O), s 8 = -34.1, -5.9 7.3 Os K s 52.7
7.5 For C = 1 .F, R 1 = R 2 = 100 Kl. For C = 10 .F, R1 = R 2 = 1 Ml.
7.6 For C = 1 F, R 1 = R 2 = 51 Kl. For:c = 10 F, R 1 = R 2 = 10 Kl.
Chapter 6 7.7 One design: Steps 1 to 7 in Figure 7.11 yield z = 15 and p = 53.3 using
the 20% overshoot design specification (with a: = 5), but a computer
6.1 a. 14.0 rad/sec, 1.5 simulation in Step 8 indicates an actual overshoot of 27%. Using a 5%
b. 6.3 rad/sec, 1.0 design specification (with a = 8) yields z = 10 and p = 72, which givcs a
6.2 a. 4.0, 63 15% overshoot, 1, < 0.5 sec, and e., = 0.036 rad. For thc latter design,
b. 4 .0, 27 R 1 = 100 Kl and R 2 = 16.1 Kl for C = 1 .F.
6.3 a. 2.97 rad/sec, 2.07, oo, 28.0 7.10 One design: a= 2.7, T = 0.05, R 1 = 132 Kl, R 2 = 77 Kl
6.4 b. a 7.75 7.14 One design: a = 0.175, 1/T = 0.06, R 1 = 140 Kl, R 2 = 30 Kl
6.5 a. O < K < 67.6 7.17 One phase-lead design: Use the root locus procedure in Figure 7. 11 with a
b. O< K < 90.6 10% overshoot design requirement (a = 4) to yicld z = 5 and p = 25. A
c. O< K < 88.7 computer simulation indicates an actual 28% overshoot ami a rise time o f
6.6 a. O < K s 12.5 approximately 0.2 sec. Further calculations show ess for a unit ramp input
b. 7.0 < K < l l.9 equal to 0.03 rad and a phase margin of 47.2 (wcc = 7.52 rad/sec). Ll'l
c. K 8.0, wtch yields 8.0 s K < l l.9 to satisfy ali three. R 1 = 200 Kl, R 2 = 50 Kl, and C = l .Fin Figure 7.9a.
6.9 b. Starting ( w = O) with a magnitude of unity and a phase of 180, the 7.19 Use phase-lag design. So much additional phase is needed al the new gain
magnitude increases somewhat for increasing w and then decreases to crossover frequency that phase-lead design is not applicablc.
zero as the phase decreases from 180 lo - 90. The polar plol swings 7.20 Use phase-lead design. Phase-lag design is not applicable in this case. (Why
from the negative real axis of the GH(s )-plane through the second, not?)
first, and fourth quadfants.
6.11 For K = 2, N = O, P = O, Z ..= N + P = O, stablc. For K = 10, N = + 2,
P = O, Z = N + P = 2, unstable. Nyquist plot has Wpc = V1 and
Chapter 8
Re(G(jwpc)} = -K/5, which implies K* = 5. T hus, stablc for O< K
< 5. 8.1 a, b. k* = oo, J* = O
c. k* = 0.5, J* = 2.5
6. 12 a. For - oo < K < 32/3, N = O, P = 2, Z = 2, uns table. For K > 32/3,
N= - 2, P = 2, Z = O, stable (wpc = /5/3 ). "' 8.2 .\ = -2x* + .\* i* = -x + u u = -A*/2
~ O; u*(J) = -(!. -
b. For -17.9 < K < 34.5, N = O, P = O, Z = O, stable. Otherwise, N = x(O) = 1, x(oo) l)e - /2 1 '
+ 2, P = O, Z = 2, unstable. Valucs of w,,c are 0.62 and 2.94 rad/sec. 8.3 - (ti - 1), same as the ratio in Problem 8.2
6. 13 For ali K > O, N = O, P = 2, Z = 2, unstable 8.5 0.4
6.15 a. For O< K < 90/1, N = O, P = l , Z = 1, unstable. For K > 90/1, 8.7 xj(t) = [/fo /(/fo - l)]e - ' - (1/(/fo - l)J e fw,, x!(1) = tf(I)
N = 2, P = 1, Z = 3, unstable (wrc = /20/7 ).
b. Two closcd-loop poles cross the jw-axis at J/20/7 for K* = 90/7. 8.10 52 28 - p 11 ) ~
Q = ( 28 _ Pu , where 4 < p 11 < 28 + v572
11
6.2 1 3.2
8.13 a/b = 2.4; lf one selects b = 1, then a= 2.4
6.22 b. 2, 32, 4.5
c. 1.3, 70, 4 8.14 w = -4w - 22y + u, z = x2 = w + 3y
6.25 a. l.7 8.20 From Theorem 2.2, Q = (e, A Te) = (1
0
-3)0 , det Q = .
O; plant 1s not
b. 5.0 observable.
 INDEX INST. TE~;. r:r ~-;'.
DC IRWU 95 I"i i

A e
Absolute error: Canonical form:
integral, 459 c:ontrollabtlity, 66
integral time, 459 controller, 66
Accieleration error oonstant, 111 observability, 67
Accessible states, 485- 488 observer, 67
AC servomotor, 44 Capacitor, n'Ovable phrte, 92, 93
Actuators, hydraulic, 45 Cayley-Hamilton:
Adaptive control, 489 ihethod, 194--199
AlgebTt1ic reduction of block diagrams, 43 theoreftl, 194
Amplidyne, 44 CharKterltdc,, feedback, 102
Amplification, power, 104 Cloled-loop ty1tema, 7, 310-315, 363-378
Amplifiers, 17, 90 Coefflcieots:
Angles of departun or arrival, 257 dampina, 160, 168
Armature-controlled de motor, 37 aemltlvity, 125
Asymptotes of root loci: Cormnunication -at-a-di sanee, 1O3
finite K. 285-289 Ccmparilooa, design, 476-482
infinite K, 254, 273, 274 Compenaation:
Asymptodc stabllity, 235 lq-lead design, 447- 4SI
Automatic control, 3, 13 phase-lqdesign, 431-447
AlUi.liary polynomia), 244, 245 phase-lag network, 411-413
phase-lead design, 413- 431, 445- 447
8 phase-lead network, 408- 411
series, 395, 408- 413, 48 1,482
Bid elec1romotive fon:,e, 38 Complete time solution, 200-206
Back emf con.stant, 38 Components, control system, 26-46
Bandwidth: Computational aids, 452
caJculation,306,311 , 401-408 Computer simulations, 3, 64
.,, definition. 108
gain tradeotTs, 107
Conditionally stable system, 348
Constraints:
Basic root locus construction ruJes, 250 plant input, 158
Block diagram reductions: state variable, 158
algebraic, 43 Constroction rules:
Mason's rule, 51 basic root locus, 250
transformation rules, 43, 48 modified root locus, 274
Block diagrams: Contours:
reductions, 43 Constant-M, 351 - 354, 364- 367
schematic, 7 Constant-N, 364- 367
transfer-function, 7 mappings, 324-328
transfer matrices from, 82- 84 root, 287- 293
Bode diagrams, 315-324, 423-431, 440- 451 Control :
Boundary conditions, 467, 470 adaptive, 489
Breakaway points, 259 decentralized, 489

505
 !>06 lndex
il
lndex 507
C ontrol (Conti11ued) De motors: cbaracteristics, 102 parabolic, 112- 116
derivative, 136, 172- 177 annature-controlled, 37- 39 controllers, 7 ramp, 112- 116
electronic engine, 9, 11 field-controlled, 34- 37 definition. 7 resistance, 90
lerarcllial, 489 Decentralized control, 489 effects, 102 sinSoidal, 305
integral. 132 Decoupling, 481,482 natural , 3, 7 step, 112- 116
inverse optima!, 476 Delay time , 158, 168, 401 - 408 Feed-forward form, 67 Integral:
linear SVFB, 140,213, 217-219 Departure, angles of, 257 Field-controlled de motors, 34 absolute error, 459
linear quadratic optima!, 469-476 Derivative control, 136, 172-177 Figure of merit, 458 control, 132
microprocessor. 10, 11, 488 Design: Final-value theorem, 112 squared error, 459
multivariable optima!, 452, 482-484 comparisons, 476-482 Flow graph, signal, 52 Integrator:
optima!, 458-476 lag-lead, 447-451 Frequency: comput.er simulations, 64- 76
PID, 130, 213-216 overview, 18 gain crossover, 356, 357 impulsed , 191-193
proponional. 131 pilase-lag, 431 - 447 natural, 160 Intersections:
slp rudder. 104, 107 phase-lead, 413-431, 445- 447 performance specifications, 305-308 imaginary axis, 251 , 274
specific optimal, 460 state variable feedback, 458- 463 phase crossover, 354-356, 361- 363 real axis, 254, 273, 274
state variable feedback, 481, 482 Direct fonns: response plots, 315-324, 423- 451 lnverse optimal control, 476
stochastic optima), 489 feed-forward, 67 Fuel, mnimum, 459 Isolation amplifier. 61 , 90
suboptimal, 460, 462, 475, 476 phase variable, 64
Controllability: Discrete-time systems, 15 G J
canonical form. 66 Disturbanees:
definition, 76, 220-222 noise, 489 Gain: Jacobian, 125
matrix, 76 output, 123 crossover frequency, 356, 357
theorems, 77- 79 Dynamic operating points, 14 margin, 307, 312, 354-356, 398-401 , K
Comrollable :
403-408
output, 76, 99 E vs. bandwid1, 107 Kalman, 476
state, 76
Gear trains, 39 KirchholT's laws, 28, 29
Controllcrs: Electrical systcms, 28 Graphs, signal flow, 52
canonical form, 66 Electronic engine control, 9, 11 L
feedback, 7 Encirclements, 333 H
Control ~ystems: Ending points for root locus, 249, 274 Lag:
applications. 4 Energy, minimwn, 459 Heating system, thermal, 45 design, 431 - 447
components, 26 Engine, electronic control, 9, 11 Hierarchial control, 489 networks , 411 - 413
multivariable, 9, 101 Error(s): Homeostasis, 3 Lag-lead dc:sign, 447- 451
single-input, single-output, 60 integral absolute, 459 Hurwitz test, 237 Lagrange:
Convolution, 201 - 206 integral squared, 459 Hydraulic actuator, 45 Euler-Lagrange equations, 463- 469
Comer frequency, 315-324 propagation, 209-213 multipliers, 465- 469
Cost functional, 459 steady-state, 110, 158, 396- 398 Lagrangian forrnu lation, 33
Critc:rion: Error constants: Laplace transfom1alions, 200, 201, 204- 206
Nyquist, 330, 331 acceleration, 111 ldentification, systems, 27, 489 Large -scale systems, 489
Routh -Hurwitz, 236- 247 positional, 111 Imaginary-axis intersections, 251 L.ead:
Critica) damping, 162 velocity. 111 -1'-
lmpulsed integratoc. 191 - 193 design , 413- 431, 445-447
Crossover frequency: Error detector: Impulse response, 201 - 206 networks, 408- 411
gain,356,357 position, 17, 40 lnaccessible state variables, 485-488 Lead-lag design, 447- 451
pilase, 354- 356, 361- 363 synchro, 44 lnductance, 28- 30 Least significant digits , 209- 213
Estimatioo, 13 lnertia: Linear:
D Euler-Lagrange equations, 463- 469 effective value, 34-40 frequency plots, 31 9-324
moment of, 32-34 quadrat1c comrol, 469- 48 2
Damping: F Initial conditions, 28 SVFB control, 140, 213, 217- 219, 481 ,
coefficient, 160, 168 Input: 482
critical. 162 Feedback: impulse, 192, 193 indt!pendcnce, 60
over, 161 amplifier, 90 multiple, 9, 63, 64 Linearization, 14- 17
under, 163, 309-315 applications, 6 node, 52 Log- log frequency plots, 319- 324
 508 lndex
hlllll)(
M N Output:
controllable, 76, 99
Q
M contours, constan!, 351-354, 364-367 N contours, constant, 364-367
Magnitude, peak frcqucncy response, 307, 312, disturbances, 123
Natural: Quadratic control, linear, 469-482
357- 359, 398- 408 Overdamped time response, 161
feedback, 3, 7
Margin: Overshoot, percent, 156, 164
frequency, 160 R
gain. 307, 312, 354-356, 398-408 systems, 7
phasc, 307, 312, 356, 357, 398- 408 p
NcgativeK: Radar tracking systems, 104, 107
Mappings, contour, 324- 328 RaAge of K for stability, 245-257
Nyquist criterion, 344-349
Mason's: Parameter optimiution, 459-464 Real axis:
root locus, 272- 287
rnle , 51 Peak magnitude, frequency response, 307, 312,
Networks, series compensation: intersections of root loci , 249, 274
signal low graph, 51 - 56 357-359, 398-408
lag-lead , 447-451 root locus segments on, 249, 274
Ma tri x: Pendulum, 24, 101
phase-lag, 408-411 Reduced-order observer. 485-488
canonical form, 66, 67 Percent overshoot, 156, 164
phase-lead, 411 - 413 Re-entry points, 249, 259, 274
diagonal, 69, 70 Performance:
Newtonian laws of motion, 30 Regulators, 12
inversc, 83, 84 Nichols chart, 368, 369, 373- 376
frequency specifications, 305-308, 396-408 Regulatory processes, 3
Jacobian , 125 measures, 158,458, 459. 463
Noise disturbances, 489 Relative stability:
plant. 60. 63, 64 time specifications, 155- 159, 396-408
Nonlinear systems, 15 Phase: constant-M contours, 351 - 354, 364- 367
rcsolvent , 191- 193 constant -N contours, 364- 367
Nonminimum phase, 323
Riccati, 470 crossover frequency, 354-356, 361-363
Nyquist criterion: gain margin, 354- 356
statc transition, 189-199 lag design, 431 - 447
definitions, 330 measures of, 350, 351
transformation, 84- 88 lag networks, 4ll-413
encirclements, 333 Nichols chart, 368, 369, 373- 376
Measurc. performance, 458, 459, 463 lead design, 413- 431, 445- 447
examples, 331- 338, 340-349 phase margin. 356, 357
Mechanical systems : Jead networks, 408-411
ncgative K applications, 344-349 Resic:h1es, 182- 188
mtational. 32 margin, 307,312, 356,357. 398-408
open -loop poles on imaginary axis, Resolvent matrix method , 191-193, 196-199
t ransl:itional, 30 Phasc variables:
337-340 Response:
Merit, figure of. 458 dcfinition, 58
steps of criterion, 3 31 impulse, 201 - 206
Microprocessor control, 10, 11,488 direct forms, 67
overdamped time, 161
Min inmm: PJD control, 130, 213-216, 481, 482 sensitivity, 158, 17 l
cncri.y, 459
o P lant:
underdamped time, 163, 309-315
fuel. 459 definition, 7
Observability: Riccati equation, 470--483
phase, 323 input constraints, 158
canonical forms. 67 Rise time, 156, 168, 401 - 408
time. 459 m ultivariable, 87
definition, 76, 223-225 Robotics, 488
Modal coordinate representation, 162, 188, 189 single-input, single-output, 82-84
matrix, 76 Root contours, 287-293
Modcl : Plots, frcquency response:
theorcrns, 77- 79 Root locus:
dcvclopmcnt. 27 Bode diagrams, 315- 324, 4V - 431 ,
Observer: angles of departure or arrival, 257
ovcrvicw. 17 440-451
canonical form, 67 asymptotes for finite K, 285-287
validation. 3 linear plots, 319-324
reduced-order, 485-488 asymptotes for infinite K, 254, 273, 274
Modi ficd root locus rules , 274 log-log plots, 319-324
Oil lamp of Philon, 12 basic construction rules, 250
Momcnt of incrtia, J2 polar plots, 319-3 24
Open-loop systerns, 7, 310-315 breakaway and re-entry points, 259
Motor: ~ Pneumatic system, 45
Operating points: contours, 287-293
arm:iture -controllcd de, 3 7 Polar plots, 319-324
dynarnic, 14 imaginary axis intersections, 25 1,274
field -controlled de. 34 Position error detector, 17, 40
static, 14 method, 247, 271
overvicw, 17-19 Positional:
Optirnal: modified rules, 274
Movable platc capacitor, 92, 93 error constant, 111
control, 458-476 phase-lag design, 431-447
Multivariable : servomechanism, 17, 41
inverse optima! control, 476 phase-lead design, 413-431, 445-447
control systems, 9, 101,452 Positive feedback, 272- 287
specific optima! control, 460 segments on the real axis, 249, 274
optima! control, 482- 484 Power amplilication, 104
stochastic optima( control, 489 starting and ending points, 249, 274
plants, 87 Propagation, error, 209-213
Optimization, parameter, 459- 464 Rotational mechanical systems, 32
Proportional control, 131
Routh-Hurwitz eriterion, 236-247

f
 ---------- -- - - - - - - - - - - - - - - - - - - - - - - - - -

'l..
5 10 lndex lndex 511
ll s State variables: .,. Translational mechanicaJ system, 16, 30 Vector-matrix notation, 60, 85
l Search algorithm, structured, 56
definition and properties, 56
forms, 64
' T)'Pes of systems, 111 Velocity error constant, 111
~iscous damping, 30.
Second-order systems, 159, 309-315 inaccessible, 485-488 u
Scnstivity: linear SVFB control, 140,213, 217-219 w
coefficients, 125 Static operating points, 14 Uncoupled form, 69, 70, 74
cquatons, 206-208 Steady-state errors, 110, 158, 396-398 Underdamped time response, 163, Waterclock ofKte;ibios, 12
to plant variations, 372 Stochastic: 309-315 Weighting matrices, 459

l Sensor, 7, 42 optima! control, 489 Unstable systems, 235

Series compensation: systems, 15 z
lag-lead design, 447- 451 Structwed search algorithm, 56 V
phase-lag design, 431- 447 Synchro error detector, 44 Zeros:
phase-lag nctwork, 411 - 4 13 Suboptimal control, 460,462, 475, 476 Validation, model, 3 effect on time response, 172- 178
phase-lead design, 413- 43 1, 445-447 Systems: Vector(s), s -plane, 182- 188 location in s-plane, 185- 188
closed-loop, 7, 310- 315, 363- 378
phase-lead network, 408- 411
Servomechansm: discrete-timc, 15 .
historical, 12, 13 human-deviscd, 7

l postional, 17, 41
Se1vomotor, 44
Seuling time, 156, 168, 401-408
natural, 7
nonlinear, 15
open-loop, 7, 310- 315
Ship rudder control, 104, 107 radar, 104

t,_ Significant dgits, 209-213

Simulations:
computer, 3
sensitivity, 119
stochastic, 15
time -varying, 15
tracking, 104
compu tc:r diagrams, 64
forms, 64- 74 translationa l mcchanical, 16
Single- input, single-output plants, 64
Solenoid , 92 T
Specific optima! control, 460
Tachometer, 17, 40, 172- 177

~
Specifications:
frequency performance, 305- 308, 396-408 Tandem fonn , cascade, 71-76
time pe,fonnance, 155-159, 396-408 Theorem:
S -plane, 19, 178 - 188 Cayley-Hamilton, 194
S plane vectors, 182- 188 controllability and observabilily, 77-79
" Spring constant, 30 final-value, 112
Stability: Time:
asymptotic, 235 dclay, 156, 168, 401-408
condi tional, 348 mnimum, 459
definitions, 234- 236 rise, 156, 168, 401-408
marginal, 235 seu ling, 156, 168, 401- 408
properties, 234-236 Time-varying system, 15
range of K for, 245-247 Tracking:
relative, 349-359 radar systems, 104, 107
unstable, 235 systems, 104
Stale constraints, 158 Transfer:
State controllable, 76 functions, 28
State transition matrix: matrices, 82

\ Cayley-Hamilton method, 194-199

definition and properties, 189
resolvent matrix method, 191 - 199
Transforrnation:
Laplace, 200-206
linear, 84
series method, 193-199 rules, 43, 48

,. )
r
J