You are on page 1of 3

TECHNICAL NOTES AND CORRESPONDENCE 77 1

the determination of cascade networks usirig the Root Locus approach


by Ross, Warren, Thaler, and Wakeland [7H9],but for the Bode and
Nichols method none exist. In this correspondence a method is given for
the exact determination of the desired lead and lag-lead networks by the
Bode method on the first effort. This is accomplished by the solution of
a quadratic equation or the use of a set of design curves.
The Bode design objective is to provide a desired crossing frequency
(w,), and phase margin (‘P). Thesecriteria have beenobtainedfrom
time-domain requirements, usually concerning overshoot and settling
time. The frequency response characteristics of the lead or lag-lead
networks are used to modify the response of the open loop so that the
desired crossing frequencyandphase margin areobtained. Fig. 1
illustrates the typical unity feedback circuit. The cascade compensatoris
shown as G,.

LEADCOMPENSATOR
Example: Consider a second-order system with gain aconstant)
The transfer function for lead networks (not including
is given as

The pair ( A , B ) is always controllable, and ( A , Q is observable if If, after examining the open-loop frequency response, it is determined
q1 > 0 irrespective of values of a, and a2. In this case, it can be easily that positive gain and positive phase contributions are needed of G,(s) to
verified that even if B is nonsingular, if either al =O or a2=0, then the obtain the desired wc and \k, thenaleadnetwork may provide the
asymptotic stability of PA is a necessary and sufficient condition for the desired time response. Equations for the gain (dB) and phase contribu-
stability of (1). This shows that rank B = n in the corollary above is only tion (a) of G, are
a strong sufficient condition.

ACKNOWLEDGMEhT

The author would like to thank Prof. M. Aoki for his useful Corn-
ments.

REFERENCES Solving yields


M. Aoki, “Stochastic control theory in economics:Applications and new problems,”
Preprints IFAC Symp. Stochastic Control (invited lecture), SepL 1974.
T. L. Gunkel, 11, and G. F. Franklin, “A generalized solution for h e a r sampled-data
control systems,” Tram. ASME (J. Basic Eng.). ser. D. vol. 85, pp. 19%203, June Since it is clear that multiple roots at zero and 1 are not applicable to the
1963. design, the solution for a must be posi$e, it can easily be shown that for
G . A. Hewer, “Analysis of a discrete matrix Riccati equation of Linear control and
Kalman filtering,” J. Math. A M [ . Appl., vol. 42, no. I, pp. 226236, 1973. onecompensatingnetwork c >p2+1. With this condition met, the
W. M. Wonham, “On a matrixRiccatiequation of stochasticcontrol,” SIAM J . quadratic will have a positive and negative solution. Theapplicable
Contr., vol. 6, no. 4, pp. 681-698, 1968.
value for a is positive. Using w, and a, T can be determined from the
magnitude expression of (2) written as

a2- a2c
(UT)’=-.
c- a2

The above procedure is dqect, exact, and accurate. However, if g a p h


Bode Compensator Design pick-off accuracy is acceptable, the solution can be obtained without
W. R. WAKELAND calculations by using Fig. 2. The curves are entered with the pecessary
gain and phase contribution. The values (JT and a are determined by
Abstruct-Since cascade compensation of feedback control systems bas interpolation of the parametriccurves.
been f i t presented, the determinationof the lead or &-lead network bas
been taughtas a trial and e-r procedure. This correspondence presentsa LAGLEADCOMPENSATOR
method, based upon the solution of a quadratic equation or a set of
The transfer function for thelag-lead network is given as
parametric curves, which will e-te this trial and error. Examples for
each case are given.
G,(s)=
[ -][
1+T2S

l-kaT2.9
~
1+T1S
1+71S/Ci
].
In spite of the advances in control system design techniques concern-
ing optimaland descrete systems, much design is performed using
classical frequencydomain procedures. Since cascade compensation of The desirable characteristics of the lag-lead network are a negative gain
feedback control systems has been first pfesented, the determination of and a positive phase contribution. Consider the transfer function astwo
the lead or lag-lead networkhasbeentaught as a trial anderror parts. Thepart involving T~ a lag, andthat involving T ~ is , the lead

procedure [1H6]. For an experienced designer the number of trials is portion. In the analysis, the polynomials can be maintained at the fourth
usually small, but for the student or the inexperienced, success comes order by first assuming the phase contribution ($3of the lag portion (lag
only after many guesses. Analytical methods have been developed for residue) be a small negative value, and second, that the gain contribution
of the lag portion be -20 loga. The first assumption wil be made to hold
exactly by the calculation of T ~ the
, second will produce a very small
Manuscript received March 10, 1975; revised June 2, 1975. error, shown in Fig. 3. Theassumed lag residue is accounted for by
neauthor is with the School of Engineering,TrinityUniversity, San Antonio, Tx
78284. ‘ adding to the phase requirement of the leadportion.
772 IEEE ‘I&WSACTIONSONALTOMATICCONTROL, OCTOBER 1976

Fig. I . Unity feedbacksystem.

Fig. 2. Cascade compensator. (a) For lead nemorks enterwithapositive dB gain and
phase needed to obtain ther.c and a.@) For lag-lead networks enter with a negative
dBandapositivephaseneeded to obtainthe a andavalue for (~7,)’. Thenuse
(31,=U/(L7,)’.

Fig. 3. Percentmagnitudeerror. The curves show, for representativevalues of a , the


percent error in assuming that the lag magnitude contributionis - 20loga.

The gain contribution is given as It is relatively simple to show that the set of parametric curves of Fig.
2 can be used for the lag-lead network if the gain axis is read as negative
dB, and the value for (JT] is taken as wrl=a/(wrl)’ where (07~)’is the
value read from Fig. 2.

where c = The lead phasecontribution is givenby(2a).Solving Examples


from (2a) and (6) yields
The details of two examples are given in Table I. Second-order system
a 2 ( a - 1 ) ~ [ ( p 2 c 2 + c ~ - c ) a ~ + 2 c p 2 a + ( p ~ - c + 1 ) l = 0 . (7) relationships are used to translate time domain requirements into
frequency requirements. The G (s) for both examples is
Just as for the lead network, the quadratic in the brackets contains the
useful solution for a. For a single network and a positive a , l / c > p 2 + 1. G (s)= K
Using a and we in (6) determines T ~ T~. is determined from the expression s(l+0.2s)(l+0.45~)
for the phase contribution of the lag portion of the network, which was
assumed earlier as
CONCLUSIONS
(aa:tanQ2)T:+ac(a- l)r;+tan@,=O. (8)
There is no longer any need for a trial and error procedure for the
The larger root is the desired root. The above procedure is as direct as determination of thelead orlag-lead cascadecompensator using the
the lead procedure. Although a small error is acquired in the magnitude Bode design approach. The use of the digital computer to produce the
assumption, it is sufficiently small for the accuracy needed in the Bode frequency response of the open loop is quite prevalent. With the com-
design. pletely analytical procedures presented in this correspondence, computer
TECHNICAL NOTES A ? CORRESPONDENCE 773

TABLE I I. IXTRODUCTION
Lead Lag-lead A number of design techniques for model-reference adaptive systems
(MRAS) guaranreeing asymproric stability for the continuous case have
Overshoot 10 percent 10 percent
Settling time, 2 percent 2s 9s been presented [1]-[3] but results for the discrete case are few [4], [5]. A
Velocity error constant 2.4 24 recent paper, however, has extended the Lyapunov design approach for
Crossing frequency we 2.4 38 continuous systems to the discrete-data case [6]. In that case, an identifi-
Phase margin 58.6” 58.6” cation procedure which also guarantees state error conver,Oencewas
Gain constant K 2.4 24 developed.
Magnitude G(&) - 4.2 dB 12.22 dB In this technical note a design procedure is developed which results in
Phase G(jw,) - 162.2” - 174.4“ arelationshipbetween adaptation design parameters and the adaptive
CD2 assumed - - 3” error transient response. This is effected by developing a linearized error
Gain needed at a, 4.2 dB - 12.22 dB characteristicequationforthe plant-modelsystemalong the lines of
Phase needed at wc 41.2” 56” continuous case MRAS schemes [2], [7], [&I.
C 2.63 0.06
P 0.88 1.48
11.21 a 6.4
FORMULATION
11. PROBLEM
0.88 71 0.55
72
- 5.79 1
The problem is one of identifymg an unknown plant of known order
Actual response os. 9 percent 9.2 percent
Actual response
settling
time 2s 8.5 s by adjusting a model “observer.” The dynamics of the plant are given by

Trossing frequency wc was chosen as 3 instead of about 0.5 because of


experience with lag networks and second-order approximations.
where A is n X n, B is n X r, and some or all of the elements of A and B
programs can now be written to determine the compensator and check are unknown. The observer model is
the time response of the closed-loop system. Such a program has been
developed by the author. x,(k+l)=A^(k+l)x,(k)+~(k+l)u(k). (2)
The use of theparametric curves of Fig.2 enablethe design to
proceed even more quickly if curve pick-off accuracy is sufficient. Some Defining the error vector as
accuracy improvement canbe achieved by generating tables with decibel
and phase inputs yielding ~ r ’ 7and a . Such tables have been generated,
and one-place accuracy is easily obtained. from ( 1 x 3 ) the error equation is
Althoughthe Bode approachdoesnotguaranteethe exactdesired
timeresponsesince it is basedupontheassumption of second-order e(k+l)=Q(k+l)x,(k)+#(k+I)u(k).
predominant roots, the ability to determinethecompensatoralmost
immediately enhances the attractiveness of the Bode approach.

REFERENCES

H. James, N . Michols, and R. Phillips, Theory of Semrnechnnisms. New York:


McGraw-Hill, 1947.
G. Thaler and R Brown, Sertumechisms Analysis. New York: McGraw-Hill,
1953.
V. Del Toro and S. Parker, Principles of Control Systems Engineering. New York:
McGraw-Hill, 1960.
J. D A z w and C. Houpis, Feedback Control %stern Analysis andSynthesis. New
York: McGraw-Hill, 1966.
B. Watkins, Infroduction io Control Systems. New York: Macmillan, 1969. where ( denotesthe 0th element of (.), P is a positive-definite,
R. Dorf, Modern Conrrol Systerns. Reading, MA: Addison-Wesley, 1974.
E. Ross, T. Warren, and G. Thaler, “Design of servo compensation based on the root symmetric n x n matrix, A, the largesteigenvalue of P, and 01 an
locus approach,” AIEE Tram. Application lo I n d u t r y , vol. 79, pp. 272-277, 1960. arbitrary design parameter, 0 < a <2. It has been observed [6], [7] that
W. Wakeland, “Analytic techniqueforroot locus compensation withquadratic
factors,” IEEE Tram. Auromat. Contr. (Corresp.), vol. AC-12. pp. 631-532Oct
the choice of a and P determine the convergence rate, and this is the
1967. subject under study. In the following, it is assumed for-analysis purposes
-, “New analytic techniques for root locus compensation,” in Proc. First Asilomr that at some prior time the elements iv= uv a n d bq= bv and a dis-
Conl. Circuirs and Systems, 1967, pp. 562-572.
turbance occurs at k = 0 such that iv # Q~ and bij# bij and adaptation is
then used to return to the parameter identification stage. However, the
Lyapunov approach guarantees e(k)+O as k+m independent of i v ( k )
+uij(k) and/or bij(k)-tbij(k).
Analysis of Error Convergence Rate for a
Discrete Model-Reference Adaptive System
111. ERRORCONVERGENCE
RATE DERIVATION
B. K. COLBURN A N J.~ S. BOLAND, I11
A secondsrder system will be studied and then results stated for the
Abstract-A technique is presented for the analysis of a class of nth-order multivariable case. This will simplify the presentation.
discrete,adaptivemodel-reference control systems via a linearizederror Although not required for stabilitypurposes, straightforward results
characteristic equation. The approach is used to determine a relationship occur if A , is in the phase-variable form
between system design parametersandtheadaptive m o r convergence
rate. A simulation example is used to illustrate the design approach.

Manuscript received April 25, 1975; revised May 5, 1976.


B. K. Colbum is withtheDepartment of Electrical Enginemin& Texas A & M It will be assumed thatsteadystate in (4) hasbeenreachedand a
University, College Station, TX 77843. perturbation occurs which gives rise to Ae, A@, and A+ Expanding (5)
J. S. Boland, In, is with the Department of Electrical Engineering, Auburn University,
Auburn, AL 36830. and (6) about the operating point %=x, = constant, e = 0, u =constant,

You might also like