You are on page 1of 14

570

IEEE TRANSACTIONS ON AWMATIC

CONTROL,VOL. AC-23, NO. 4, AUGUST

1978

Stable Adaptive Controller Design Direct Control


KUMPATI S. NARENDRA AND LENA S. VALAVANI

A/~srmct--This paper deals roifb the design OE stable adaptive eontrollers


using a model reference approach. A systematic praredure is developed
using the colloept of positive realness and is geoeralized to arbibarlly
located control parameters. For plants a n poles and ( > )(n -2) zeros,
the uniform asymptotic stability in the large of the adaptive loop is
demonstrated. For plants with ( < ) ( n - 3 ) zeros tfae M i problem is
clarified and is stated as a oopljecture. Simulation studies of the adaptive
control of both stable and rmstable p h t s are included towards the end of
the paper.

I. INTRODUCTION
HE theory of adaptive observers,
which
simultaTneously estimate the state and parameters of an unknown linear time-invariant system, is well understood at
the present time. The original observers were proposed in
[l] and [2] and were later generalized in [3]. The extensions to multivariablesystems [4],[5], the analysis of
related stability questions [6], [7] and the study of the
speed of response of the observer [8] have provided valuable insight into the design of such systems. Further, it
has been shown that all the results can be extended in a
straightforward manner to discrete systems [9].
Contrary to the expectations of research workers in the
field, the solution of the control problem did not follow
directly from the solutions obtained for the adaptive observer. The principal difficulty lies in the assumption that
the input and the output of the observed plant are uniformly bounded. As described in [lo], this is not assured
in the control problem, since it is precisely the question of
stability of the overall feedback system that has to be
resolved in this case. Further, the structure of the observer
can be chosenby
the designer so that thenecessary
signals can be fed into any part of the observer to generate the desired estimates. In contrast to this, in the control
problem, only the input to the unknown plant can be
affected and this poses a serious constraint in its resolution.
In this paper, the adaptive control problem for the case
of an unknown plant with a single-input and a single-output is considered. The plant is assumed to be linear and
time-invariant and a differentiator-free controller is to be
Manuscript received June 22, 1977; revised December 30, 1977. Paper
recommended by R. Monopoh, Past Chairman of the Adaptive, Learning Systems, Pattern Recognition Committee. This work was supported
by the Office of Naval Research under Contract N00014-67-C-0017.
The authors are with the Department of Engineering and Applied
Science, Yale University, New Haven, CT 06520.

designed to generate a control input u(t) so thatthe


output of the plant evolves asymptotically towards that of
a given model. Earlier work in this field is due to Monop
oli [ 111 who, following the approach adopted in adaptive
observers [3], suggested an ingeniousschemewhich involves the use of an augmented error model. The approach developed in this paper has the general flavor of
the method used in [ 111 though the actual structure of the
controller is different. The latter provides the necessary
motivation for the changes that have to be introduced in
the controller structure as well as the adaptive laws on the
basis of the prior information regarding the plant transfer
function. In Section VI, the various arguments used in the
design of the controller are explained in terms of a single
arbitrarily located control parameter to indicate the generality of the approach and its potential application in other
control situations.
Perhaps the most crucial question that has to be
answered in the context of the adaptive controller, as in
the adaptive observer, concerns the stability of a set of
nonautonomous error differential equations which assures
that all the parameters are bounded and that the output
s
on which the entire design
error tends to zero. T h ~ point,
is based, has not received the careful attention it deserves
in [ 111 and hence, the conclusions drawn there are not
justified. The principal contributions of this paper are a
systematic procedure for the design of adaptive controllers and the clarification of the related stability problems. For cases where the plant has n poles and (n - 1) or
( n -2) zeros, the controller is shown tobe uniformly
asymptotically stable. In cases where the plant has ( <)TI
- 3 zeros, the necessary modification of the method raises
questions regarding the boundedness of the outputs of the
model and the plant evenwhen
the output error is
bounded. This problem is formulated in a more general
setting in terms of the behavior of signals in a feedback
loop when certain known linear transformations of these
signals are bounded. It is conjectured in the paper that
under the above conditions the closed loop signals will, in
fact, be bounded, thus assuring uniform asymptotic stability of the adaptive system.While the validity of the
conjecture remains at present an open question, its verification will justify theuse of the relativelysimple controller structure suggested here and result in the complete
resolution of the adaptive control problem.

0018-9286/78/0800-0570$00.75 01978 IEEE

NARENDRA AND

57 1

VALAVANI: STABLE ADAF'TIVE CONTROLLER DESIGN

Hence, to have a well defined problem, the model transfer


function should also satisfy inequality (5). For the sake of
simplicity, we shall assume, in the following sections, that
r = m and that the equality is satisfied in (5).
The design method proposed in this paper calls for an
is
operator L(s) such that the transfer function WM(s)L(s)
strictly positive real.' Once again, for the sake of simplicity and with no loss of generality, it is assumed that L(s),
a polynomial of degree (n - m - l), exists so that this
condition is satisfied. [It is obvious that a rational function L,(s) with a numerator polynomial of degree (n - 1)
and denominator polynomial of degree m exists such that
W,(S)L,(S)is positive real. Such a function can be used
in place of the polynomial L(s).]
11. STATEMENT
OF THE PROBLEM
The adaptive control problemmaynow be stated as
follows:
The plant P that is to be controlled iscompletely
Given aplant P withinput-output
pair {u(t),y,(t)}
represented by the input-output pair { u(t),yp(t)} and can
described
by
(1)
and
(2)
and
a
model
M({r(t),yM(t)})
be modeled by a linear time-invariant system described by
described
by
(3)
determine
a
suitable
control
function u ( t )
the differential equations
such that
ip
= Apxp+ bpu(t)

Indirect and direct adaptive control of an unknown


plant were discussed in [lo]. In the former, an observer is
employed to estimate the parameters of the plant and
these estimates, in turn, are used to determine the control
parameters. In direct control, the parameters are adjusted
directly, without an intermediate identification stage. In
this paper, only direct control is discussed; indirect control as well as the relation between the two approaches
will be discussed in a later paper.
Simulation results are included mainly at the end of the
paper for different control situations of varying degrees of
complexity.

y, = h 'xp

(1)

The solution to the aboveproblemmaybebroadly


where A, is an (n X n) matrix and h and bp are n-vectors. divided into two parts. The first part (SectionIV)is
The transfer function W,(s)of the plant may be repre- algebraic in nature and addresses itself to the question of
sented as
realizability of a suitable controller structure. Using such
a structure, constant values of the controller parameters
(2) can be shown to exist such that condition (6) is satisfied
for any arbitrary input r(t). The second part (Sections V
Wp(s)is strictly proper with Zp(s) a monic Hurwitz poly- and VI) is analytical in nature and deals with the adaptive
nomial of degree m( < n - l), $(s) a monic polynomial of scheme for updating the parameters so that the error e,(t)
degree n, and kp a constant gain parameter. We further in (6) evolves asymptotically to zero.
assume that only m, n, and the sign of kp are known for
usein the design of the controller. (Thesign of k, is
PRELIMINARIES
111, MATHEMATICAL
assumed to be positive throughout this paper.)
A model M represents the behavior expected from the
The theory of adaptive observers, as shown in [3], may
plant when it is augmented with a suitable controller. The be developed entirely on the basis of three prototypes
model has a reference input r(t) which is piecewise con- which indicate how the parameters of a systemcanbe
tinuous and uniformly bounded and an output yM(t).The adjusted to match those of a model using only input-outtransfer function of the model, denoted by WM(s),may be put information. We state below several lemmas which are
represented as
useful in the control problem and which are derived from
the third prototype.
A . Lemma1

where Z,(S) is a monicHunvitzpolynomial


of degree
r < m, RM(s)is a monic HuMitz polynomial of degree n
and k , is a constant. The deviation of the plant from the
desired behavior is measured by the absolute value of the
error between plant and model outputs as
lel(t>lk l Y p ( d - Y M ( ~ ) l .

Given a stable (n X n ) matrix A , a symmetric ( m x m)


positive definite matrix r, vectors h, d E R" and
a(?):[0, co)+R" whose elements are bounded and piecewise continuous, the equilibrium state of the set of (n + m)
differential equations

(4)

Since the dynamic controller to be designed should be


free of differentiators itfollows that the plant together
with the controller (with constant parameters) would have
a transfer function with
(number of poles) - (number of zeros) > ( n - m). ( 5 )

e , = h'e

'Throughout the paper both differential equations and transfer functionsare used in the arguments and, depending on thecontext, "s'' is
used as a differential operator or the Laplace transform variable.

512

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. AC-23, NO.

is stable and lel(t)l+O as t+m if the transfer function


h (sl-A)-d is strictly positivereal (SPR). Further, if
w ( t ) is sufficiently rich,

lim ll+(t)Il =o.

r-30

Proof: The proof follows directly by


using
Lyapunov function candidate

4, AUGUST 1978

B. The Operator P,(O)

In the development of the solution to the control problem, an operator P,(@ is found toplay an important role.
If L(s) is a Hurwitz polynomial of degree n, in the
differential operator s and 8(t) is a bounded differentiable function of time, the operator PL(8) is defined as

PL(e) g ~ ( s ) e ( t ) ~ - l ( s ) .

V ( e , + ) = 1Z [ e T P e + + T I - l @ ] with P = P T > O .

From the Kalman-Yacubovich lemma it is known that


if h (SI- A ) - d is strictly positive real, a matrix P exists
such that
A T P + P A = -qqT-EL
Pb= h

(9)

i) It follows directly that PL(6)is h


e
a
r in 8. P,(cB)=
cP,(B); PL(aB,
= olpL(B1) + pPL(8J for all real
numbers a, /3 and c.
ii) When 8(t)= e*, where 8* is a constant, PL(8*)= 8*.
Hence, PL(B)- e* = P,(O - 8*). If O(t) = 8* +HI)

PL(0) - 8- PL(+)

for some vector q, matrix L= LT>O and E >O. This, in


turn, yields a time derivative

-+.

iii-a) If ~ ( s ) = s + a , P , ( e ) = [ e + B ~ - ~ ( s ) l ;
iii-b) If L(S)= s2 + US b, pL(e)=[e + 8(s ~ / 2 ) / y s )
(s a/2)8/L(s)];
P(e,+)= 1
-3eT(qq+~)e<O.
iii-c) If L(s) is any Hurwitz polynomial of degree n,,
PL(@
can be expressed as
Hence, the equilibrium state is uniformly stable and e(t)
and +(f) are bounded if e(to) and +(to) are bounded. Since
n,
w ( t ) is bounded, from [T, lim IIe(t)ll=O. This, in turn,
t+m
pL(e)=e+
[ B , ( ~ ) ~ ( ~ ) A , ( ~ ) ] L (io)
-~(~)
yields from (7)
i= 1

+ +

lim II&t)ll=O

and

t+30

Iim +(t)w(f)=O.
2-30

If w(t) is sufficiently rich as defined in [7], the above


two conditions imply
lim

t-30

where Bi(s) and Ai(s) are polynomials of degree (i - 1) and


(i), respectively. The expressiongiven
in (10)is
not
unique. A useful expansion when L(s)=X;l ,(s + 3) in
terms of B,(s)and A,(s) is

Il+(t)ll =os

n,-i

B,(s)=

Comment: It isseen from Lemma 1 that IIe(t)ll and


I[+(t)ll arebounded
with no constraints on w(t). The
boundedness of Ilo(t)ll,however, assures the convergence
of Ile(t)ll to zero. Further, the richness of w ( t ) results in
+ ( I ) converging to zero.
In Lemma 1 w ( t ) was an independent input vector. If,
however, it is generated as a feedback signal using a stable
systemwith e l ( t ) as aninput, we obtain thefollowing
corollary.
CoroZZury I : If, inLemma 1, the condition on w ( t ) is
replaced by the differential equation

II (s+z/);
j=2

A,(s)=

II (s+zj)

i=1,2,---,n1

j=1

and Bl(s)=l.
iv) If W(s)is a stable transfer function with n poles and
m zeros and n, < n - m - 1, W(s)P,(B)is a linear bounded
operator. This follows directly from the facts that L(s) is
Hurwitz, 8 is bounded and W(s)L(s) is a stable transfer
function.
C. Lemma2

If Q(s) and T(s)are monic polynomials of degree n and


m( 6 n - l), respectively, which are relatively prime, polynomials P(s) and R(s) of degree ( n - 1) with P(s) monic
where A is an (mX m) stable matrix and b ER then
exist such that P Q + RT can be made equal to any arbitrary
monic polynomial of degree (2n- 1).
lim lel(t)l=O.
2-20
Proof: Since Q(s) and T(s) are relativelyprime,
Proof: The
boundedness
of lel(t)l assures the polynomials A ( s ) and B(s) of degrees (rn- 1) and (n - I),
boundedness of Ilw(t)ll since A is a stable matrix. By where m > 1, exist such that
Lemma 1 this assures that lel(t)l+O as t+m.
A Q + B T = 1.
Corollary 1 implies that the set of equations (a, (8)
If Q(s)= Z;=oqjs and T(s)=Zy==,risi,
(4, = t,,, = l), the
e=Ae+d+Two; =
; -I(hTe)o;
Lj=Aw+bhTe
above equation can be satisfied if the (n + m - 2) X (n + m
is stable and Ile(t)ll-+O as t+m.
- 2) matrix C given by

;J=Aw(t)+ be,

(8)

NARENDRA AND VALAVANI: STABLE ADAPTIVE CONTROLLER DESIGN

0
1

1
4,-

'

4n-I

1
4,-1

40

40
0
0

0
1

1 L - 1

C=

0 1

0
0

0 0 0

4l- I

0
0

0
1

to

to

Il

0
0
0

0
0

tm- 1
'm-2
tm-k

m-1
where 0 < k < m,is nonsingular. This is assured if Q and T
are relatively prime.
Let M be the arbitrary polynomial of degree ( < )2n - 1.
Then

n-1

differential equations

w(I)

w(') = doyp+ dTd2)

We then have
(MA+NT)Q+(MB-NQ)T=M

where MA + N T is a monic polynomial of degree n - 1


and M B - N Q = R is of degree S ( n - 1). Choosing MA +
N T = P , the result follows.
ADAFTIVE

SYSTEM

As mentioned in Section I, we first require the adaptive


controller to possess enough freedom so that the control
problem has a solution. This implies that for some constant values of the controlled parameters the corresponding transfer function of the controlled systemexactly
matches that of the model.

The Controller Structure

The basic structure of the adaptive system is shown in


Fig. 1. This structure is suitably modified for the different
control situations considered in Section V. The controller
consists of a gain ko, and auxdiary signal generators Fl
and F2. F, contains ( n - 1) parameters ci ( i = 1, 2,. . ,n 1) and F2 contains n parameters di (i = 0,1,2, . . ,n - 1).
Together with ko these constitute the 2n adjustable parameters of the controller, which are denoted by the elements
of a parameter vector O(t).
e'(t>'

and br= [O,O;..

where R is of degree ( 9 ) n - 1.

IV. THESTRUCTURE
OF THE

= cTu(')

[ko(t),c,(l), . . . , c n-,(t),do(t),d,(t)...dn-l(t)]

Li [ko(t),cT(t),do(t),dT(f)]. (1 1)

Fl and F2 are described by the ( n - 1)th order vector

(1 2 4

(Fl)
d(2)=Au(2)+byp

If N is the proper part of B M / Q

d(1)

MAQ+MBT=M.

BM
-=N+-

(F,)

(12b)

11.

where A is an ((n- 1) x ( n - 1)) stable matrix, Fl and F2


have transfer functions W,(s)and W2(s),respectively, for
constant values of the parameters ci and 4 (i = 1, 2, * ,n
- l ; j = O , l;..,n-l).

W 2 ( s ) = d O + d ' ( ~ l - A ) - ' b do+-D ( s )


(13)
N(s).
If (A, b) is in companion form, ci and 4. (i = 1,2, . ,n - 1)
represent the coefficients of s n P i - l of the numerator
polynomials of cT(sl--A)-'b and dT(sI-A)-'b.
For constant values of 8, the overall transfer function of
the controlled system is W(s)where

kokpZp(s)Ns)
[ N W + C(s)]%(s)+k,Z,(s)[D(s)+doN(s)I

(14)

W(s) as given in (14) has (3n - 2) poles and (2n m - 2)


zeros of which (n - 1) are common. Further, by Lemma 2
it is known that a constant parameter vector 8* exists such
that W(s)= W,(s), which implies there are (n - 1) additional pole and zero cancellations. From (14) it is seen
that the poles of the auxiliary signal generators are also
zeros of W(s). In order that the overall transfer function
have the same zeros as the model, N(s) is chosen to
contain Z,(s) as a factor.
If the vector w ( t ) is defined as
w '(1)

2 [ r(t),~(')(t),y,(t),u(~)(t)]

IEEE TRANSACTIONS O N AUTOMATICCONTROG VOL AC-23, NO.

4, AUGUST 1978

,..

Fig. 1. Basic structure of the adaptive system.

Fig. 2.

However, (17) is used only for purposes of analysis of the


error equations and not to generate any adaptive signals.

the overall system can also be represented as

The Error Equation

The error equations between model and plant may now


be expressed as
G=A,e+b,[+'(t)u]

yp = h 'xp.
Denoting 8(t)= 8* +Q(t) the state equations (15) can also
be written as
f=A,x+b,[ k,*r++'(t)u]

where

[ Ap+d,*bphT

bpc*T

e, = h,Te

where e( t ) A x ( t )- xm(t), h, and b, are (3n - 2)-vectors,


h,' = [ 1,0,0,
,01, b,' = [b:, b ', 01. The error transfer
function We(s)
is given by

- -

~ e ( s ) = h ~ ( S ~ - A c ) - ' bkPc =W,(S).


k,

I bpd*T 1

(18)

(19)

Hence, the error equations can be represented as shownin


Fig. 2, with '(t)u(t) as the inputtoa
systemwith
transfer function We(s). It is worth noting that the signals
q ( t ) , ( i = 1,2;
,2n), are derived from r(t), the reference
input, u(t), the control input, and y,(t), the output of the
plant.

+
-

A, =

Mod$ication of Controller Szructure

---

and b T = [O,O, ,1] is an ( n - 1)th order vector. When


+ ( t ) 0 and 8(t)= 8*, (16) also representsthemodel. If
x,, is the state vector of a nonminimal representation of
the model,
fmc
= A,x,

+ b,k,*r

(17)

where x:, = {x:,,~:)', }


'
)
:
~
t
and the (3n - 2) elements of
x,, correspond to those of the plant.
We note that the plant is not known and, hence, the
model cannot berealizedinthe
form shown in (17).

In the representation of the controller that we have used


thus far, if every element Oi(t) of the parameter vector 8(t)
isreplacedby P,(Bi) (with L - a polynomial of degree
( n - m - 1) in "s"), it follows directly that i) B(t)= 8*
would make the overall transfer function W(s) equal to
WM(s),and ii) the error equations can be represented by a
linear time-invariant dynamical system with (3n - 2 ) +
2 n ( n - m - 1) states, a single-output e, and a single-input
ue(t)=Z:l,+i(t)&(t). The transfer function from ue(t) to
e , ( t ) is now k,/k,L(s)W,(s) and the signals &(t)in Fig.

NARENDRA

AND

575

VALAVANI: STABLE ADAPTNE CONTROLLER DESIGN

-$ e, ( t l

w(1)

L-'

L(S)

WM (s)

Fig. 3.

Case ii): Wp(s)with (n - 2) Zeros and n Poles

In this case, a model is chosen with (n - 2) zeros and n


poles such that L(s)W,(s) is strictly positive real with

L ( s )= (s + p).

3 are given by

From Section 111-B with L(s)= (s + p), the operator P,(+)


& ( t ) = L - ' ( s ) w , ( t()i = 1 , 2 , . . - , 2 n )( .2 0 )
= L+L-' is functionally equivalent to { + ( t ) + + ( t ) ~ - ' } .
Since $ ( t ) is determined completely by the adaptive law, it
The proper choice of L(s), the realization of P,(@ in the is a known signal. The adaptive controller structure has
systemwithoutusing
differentiators and the proof of the form Shown in Fig. 4. EveryParameter e;(t) isnow
> the control input u(t)
stability of the modified system form the essence of the replaced by the operator P L ( ~ ;and
solution to the control problem.
bygiven is
2n

u(t)=

V. THECONTROL
PROBLEM

2 [e;(t)+B,~-~]~~(t).
i=
1

The complexity of the solution to the control problem


depends to a large extent on the prior information that is
available regarding the transfer function of the plant-in
particular, the order "n" of the plant and the number m of
zeros.While our aim is to developa controller for the
general case, for convenience of exposition we shall consider the following four cases of increasing generality
where it is known that Wp(s)has:
i) ( n - 1) zeros and n poles.
ii) (n - 2) zeros and n poles.
iii) m( < n - 3) zeros and n poles and a known gain kp.
iv) m( < n - 3) zeros, n poles and an unknown gain

If the parameters are updated using the law

e= - r e , ( t ) t ( t )

r = rr>o

where
L - ' o ( t ) = < ( t ) , the error e,(t)+O as t+m.
Proof: The proof of stability of the overallsystem
and the convergence of le,(t)l to zero follow along lines
very similar to those in case i).
2n

kP.

The first case with (n - 1) zeros and n poles has been


known for alongtime.
It hasbeenexplicitlyused
in
adaptive observer theory by Narendra and Kudva [3] and
in the control context by Monopoli and Gilbart [13]. The
results presented for the second case are new although a
less general version appears in [ll]. The modification of
the controller structure as discussed in the previous section enables the same concepts to beextended to the The error equation (18) may now be written as
general cases iii) and iv).
r 2n
1
Case i): W,(s) with (n - 1 ) Zeros and n Poles

A modelis chosen2 which has a strictly positive real or


transfer function:
2n

t = Ace + bPR2 +&;

e , = A,'.

i=l

zThis is for convenience of analysis only; if the model is not positive


real
the
same
analysis
can be used
by
proper
prefiltering
of the
reference
signal.

input is fed into the plant, it can also befed into the
as described in the
section.
Comment 2: For this
is case it seen that 2n additional

576

E E E TRANSACTIONS ON AUTOMATIC CONTROL,V O L AC-23, NO.

Ym

MODEL

SIGN
AA
UL
X.
GENERATOR

Ei

[ i =2,

4, AUGUST 1978

...n)

ti ( i = n + l ,

... 2 n l

Fig. 4. System structure for plant with n - 2 zeros.

transfer functions of the first order are needed in the


controller.
Case iii): With m( < n - 3) Zeros and n Poles and Known
Gain kp

When kp is known, by including a gain kM/kpin series


with the plant, its gain may be adjusted to have any
desired value kM. Hence, without loss of generality, we
can assume that the model gain kM is chosen to be equal
to that of the plant, i.e., kp= kM.
In case ii) the derivative &t) of the parameter vector is
available from the adaptive laws and it is this fact that
allows the use of the procedure outlined. Unfortunately,
thismethod cannot be directly extended to the present
case since higher derivatives of O ( t ) are not available.
Faced with this problem we develop an alternative
approachwhichinvolves
the use of the model in the
generation of the adaptive signals.Since a successful
adaptive scheme wil result in the plant transfer function
evolving to that of the model and in the derivatives of Qr)
( i = 1,2, * ,2n) tending to zero, the new algorithm calls
for an additional input to the model, as in [ll], which
depends on these derivatives. Specifically, the signals fed
into the model are such that the form of the previous error
equation (18) is preserved.
If the state equations of the plant together with the
controller are given by

--

Since +- P~(+)=+-L+L-'=L[L-~+-+L-'],
if L(S) is
chosen to be a polynomial in "s" of degree n - m - 1 so
that

{h,T(sI-A,)-'b,}L(s)= W , ( s ) L ( s ) = {l2:(SI-Ac)-'bpR}
is strictly positive real, the auxiliary signals can berealized
without the use of differentiators.
The model state equations (24) are modified to
~m,=Acxm,+bck~r+b~RIL-'+(t)-+(t)L-l]'w(t).
If
[L-L~;i(t)-+;(t)L-']w,(t)=S;.(t)

(25)

the controller requires two transfer functions of order


(n - m - 1) corresponding to eachparameter
that is
adjusted. Further, since kM= kp in this case, k,,= 1 and
only (2n - 1) parameters need to be adjusted.
The error equations corresponding to (23) and (24) are
(26)

~=A,e+b,PL(+)Tw(t)

e, = h,'.
which are identical to (22) in case ii). From (26) it follows
that the adaptive law
+= -rel(t)E(t)

where L-'(s)o(t) =<(t), as before, results in bounded out,2n).


i=A,~+b,[k,*r++~(t)w(t)]
(23) put error e l ( t ) and parameter errors $+i(t) (i= 1,2;
It is at this point that we encounter a new stability
the auxiliary signals [+- PL(+)]w(t)(or [ O - P,(O)]o(t)) are problemwhich did not arise in the previous two cases.
used as inputs to the model in equation (17) resulting in The boundedness of e , ( t ) does not assure the boundedness of the plant and model outputs. The model output
the model state equations
now consists of two components-the desired bounded
output and the output due to the auxlliary signals. Since
x~=A,x,,+~c[k,*~+{~-P,(+))=~(t)]
the latter signals may be unbounded, the adaptive scheme
T
Y, = 4 - L C .
(24) mayresult in an overall unstable system. This question,

--

577

NAFENDRA AND VALAVANI: STABLE ADAPTIVE CONTROLLER DESIGN

whose resolution is essential to the completion of the suggested in this paper. We discuss this question here and
adaptive control problem, is discussed later in this section make a conjecture whose verification will complete the
solution to the adaptive control problem.
in greater detail.
The proof of boundedness of the output error e l ( t ) is
Care iv): Wp(s) has m( < n - 3) Zeros and n Poles; kp relatively straightforward in all the 4 cases discussed
Unknown
earlier. In i) and ii), since the output of the model is
bounded, the output of the plant yp(t) (=y,(t)+e(t)) is
In iii) the gain kp was assumed to be known. Here, since also bounded, in which case e,(t)+O as t+w. In iii) and
the auxiliary signal corresponding to each parameter Q t ) iv) the introduction of auxiliary signals to the model
is fed into the model (whose gain kM is different from kp), precludessuch an immediate conclusion. If it canbe
additional compensatory gains k,(r) ( i = 1,2; ,2n)are
shown that the plant output is bounded, then the error
needed in the adaptive controller. While the plant equa- el(t) will tend to zero asymptotically and, hence, the
tions remain the same as in iii) the model state equations auxiliary signals also tend to zero as t - x a . The model
are modified to
output then evolves to the desired output and is followed
by
the plant.
r
2n
The plant state equations (29) and the system error
k,*r+k, 2 L ( s ) k , ( t )
i= 1
equations (30) are

x
212

=A,x,+

b,k,*r+ b,

k,ki(t)&(t).

i= 1

The error equations may be expressed as:


where the components of o(r)are all state variables of the
plant. Hence, (29) represents a feedback loop with (2n - 1)
time-varying parameters while (30) is an open-loop system
containing the same parameters. The fundamental stability problem is then to conclude the boundedness of x ( t )
from the boundedness of e(?)and Q(t). Since A , is a stable
matrix and only the boundedness of x ( t ) is of interest, it is
sufficient to treat the homogeneous equation (i.e., r(t)=O
in (29)).
The principal stability problemwhich hastobe
resolved is now stated as a conjecture.
A Conjecture: Let twosystemsbe
described by the
differential equations

Defining [I -(k,/kp)kj(t)] 2 1C;(t),(27a) reduces to

2n

e = ~ c e + b p R k p [i2
=1 ~;(t)S;(t>++;(t>~i(t>

(28)

e, = hTe.

Hence the adaptive laws for the 4n parameters are


Qi(t)=-eltj(t)

(i=1,2,-.-,2n)

\ii(t)=-elli(t)

(i=1,2;-.,2n)

where
by
(20)
and (25) L - ' o , ( t ) = & ( t ) ; [L-'+;(t)Q j ( f ) L - ' ] w i ( t ) = { i ( tand,
)
hence, 4nparameters and 4n
transfer functions L - ' ( s ) are needed in this case.
The questions of stability raised in iii) also apply here
and are considered in the following.
The StabiliQ of f h e Adaptive Loop

The arguments in iii) and iv) lead us to the central


stability question connected with the adaptive algorithm

where A is an (n X n) stable matrix, b is an n-vector and

+ ( t ) is an m-vector ( m < n) of bounded continuous time

functions. If e ( t ) is bounded, then x ( t ) is also bounded.


A better understanding of the implications of the conjecture canbe
obtained by considering only a single
time-varying parameter as shown in Figs. 5(a) and (b). In
this case, the feedback system has a stable transfer function W ( s ) in the forward path and a single bounded
time-varying gain ~ ( tin) the feedback path. The outputof
W(s) is y ( t ) and is the input to the bounded linear
operator [ W ( s ) L ( s ) + ( t ) L - ' ( s ) ]whose output is e(t) as
shown in Fig. 5(a). The special case when W(s)L(s)=l/(s
a) is shown in Fig.5(b).Given that e(t) is bounded,
then, according to the conjecture,y(t) is also bounded.
If y ( t ) is unconstrained, itsboundedness cannot be
inferred from the boundedness of e(t), where e ( t ) =

578

{>
;{>
;

IEEE TRANSACTIONS
ON

AUTOMATIC CONTROL, VOL.

AC-23,NO. 4,

Yp(t)

AUGUST

19'78

YP(t)

PLANT

w,( f )

6(t)

t(t)

(a)

(b)

Fig. 6.

corresponding points l', 2', 3' and 4' in the model are also
accessible. The open-loop transfer function W,,(s) between points l and 4 is equal to W,.,(s) in the model. The
aim of the adaptive procedure is to determine the rule for
updating $(t)(or equivalently b(t)) so that limt+mle,(t)l=
0. Thefour cases that we considered in the previous
section can now be summarized for this general problem
as shown below. It is worth pointing out that the transfer
function W14(s) (or W,,.(s)) is of direct interest in this
case and not the transfer functions W24(~) (or Wr4(s)).
However, for ease of exposition, we shall assume that
points 1 and 2 coincide (i.e., the output of the adjusted
parameter is f e d back to the input) so that W&)=
W,,(s). In this case all the conditions can be stated in
terms of W2.,(s), or the model transfer function. For the
more general case where points 1 and 2 do not coincide,
W,(s) in the followingdiscussion must be replaced by
{ WL+L- '} y(t). The basis for the conjecture lies in the W,&).
i) W,(s) E { SPR } : The adaptive rule
fact that y ( t ) has to belong to the class of signals generated by the feedback loop shown in Fig. 5. If al is the
&(O = - e , ( t b d t )
class of all unbounded signals defined for all t E[O,00)
such that { WL+(f)L-'}yis bounded foryES2, and a2is where o,(t) is the input to the parameter being adjusted
the set of all solutions of an nth order linear differential assures that lel(t)l+O as t+00 by the direct application of
equation with a single bounded time-varying parameter Lemma 1. The evolution of the parameter toits final
+(t),the conjecture states that a, n !& is the null set.
value and the error e l ( t ) to zero for stable and unstable
The application of the conjecture to the adaptive con- plants is shown in Fig. 7(a), (b).
trol problem may be described using Fig. 5(c). According
ii) PVW(s)(s + p ) E { S P R } : When W&) is such that
to the conjecture if el([) is bounded, so is the output of the W+,(s)(s+p) can be made strictlypositiverealby
the
plant j , ( t ) .
proper choice of a constant p, an additional stabilizing
signal is fed into the plant as shown in Fig. 4. T h e signal is
proportional to $(t) and tends to zero as the adaptation
VI. GENERALIZED
A
D
m
- PROCEDURE
progresses. The adjustment of a single parameter using
The method presented in the previous section can now this scheme is shown in Fig. 8(a), (b) for both stable and
be generalized to arbitrarily located control parameters. unstable plants.
iii) W,(s)L(s) E { SPR }: When a polynomial L(s) (of
Since the same procedure can be used for any control
parameter, we shall merely consider the case of a single degree 2 2) exists such that W,L is strictly positive real,
adjustable parameter in the plant. The following develop- the stabilizing signal ( 6 - PL(8)}wI is fed into the model
ment is motivated in part by a procedure used for the as shown in Fig. 9. The model is suitably modified so that
this can be accomplished without theuse of differentiageneration of sensitivity functions as described in [12].
Let M represent the model and P the plant in Fig. 6(a) tors. The evolution of the feedback parameter, the stabilizand let B(r) be a control parameter. It isassumed that ing signal and the output error is shown in Fig. 10(a), (b)
when 6(t)= 6*, a constant, the transfer function of the for both stable and unstable plants.
iv) W , ( ~ ) = k , / k , ~ W , ( s ) ;W , ( s ) L ( s ) E { S P R } : If the
plant is identical to that of the model. The plant with
8( t ) = 8* is denoted by P * ; if e(r)= 8*+I$([),
the plant can transfer function between points 1 and 4 in the plant is
be represented as shown in Fig.6(b) by P* together with a the same as that of the model except for a constant gain,
parameter +(t). For the purposes of analysis four points 1, an additional gain is needed in the path of the stabilizing
The evolution of the feedback
2, 3, 4 in Fig. 6(b) are of interest and we assume that the signaliniii)(Fig.11).

579

NARENDRA AND VALAVANI: STABLE ADAPTIVE CONTROLLER DESIGN

52 + 4 s + 3.75
s3+ 5 s 2 + 7 s +2 25

[PLANT) Wp(s) =

INPUT
Square
wave
Amplitude 5 units
Frequency = 1 hz

-\

INPUT: Squarewave
Amplitude . 5 units
Frequency= I hz

\A

CONTROL OF UNSTABLEPLANT

12

IO

-8
t in seconds

I\

IO

12

14

.A

Fig. 7.

(PLANT) Wp(s) =

s+ 2.5
s3+6s2+9s+ I

4l

( M O D E L )sW
+ 2y 5( s ) =

s 3 + 6 s 2 + 11s + 6

INPUT: Square wave


Amplitude. 5 untt5
Frequency = I hr
L(s) = s+1.5

INPUT: Square wave


Ampllfude : 5 unlts
Frequency = 2 hz
L ( 5 ) = s+1.5

CONTROL OF UNSTABLEPLANT
ux. signal

IO

I2

14

nl

'

-11 I

t in seconds

Fig. 8.

IO

14
t in seconds

ta

22

~EEETRANSACTIONS ON AUTOMATIC CONTROL, VOL. AC-23, NO. 4, AUGUST

580

1978

MODEL

r(t)

MODEL*

Fig. 9.

INPUT: Sqmare
wave
Amplitude : 5 units
Frequency = I hz

:I
3

L ( s ) =(s+1.5)(r+2.5)

INPUT: Square wove


Arnplltude . 5 units
Frequency = 1.5 hz
L ( S ) = (s+1.51ls+2.51

CONTROL OF UNSTABLE PLANT

t in seconds

Fig. 10.

(b)

parameter 8(t) to the value 8*, the stabilizing signal { ( t ) the output error Ie,(t)l, as well as thestabilizingsignal
for both IS(f>I, tending to zero at rates which are Practical. In View
and the output error is shown in Fig. 12(a),
of the fundamental theoretical questions that arise in the
stable and unstable plants.
stability analysis of the adaptive system, these simulation
VII. FURTHER
SIMULATION RESULTS
studiesprove
invaluable
pro&ing insight into the
nature of the adaptive controller. In this section we preExtensive simulations of the control problem on thesent
simulation results for a third order system for the
digital computer indicate that the procedures outlined in fourcasesdescribed in Section V. The plant and model
the previous sections do result in a robust controller with transfer functions for the four cases are given in Table I.

e)

in

58 1

NARENDRA AND VALAVANI: STABLE AL? CONTROLLER DESIGN

MODEL

r(t)

__

P*

L-

L-9

Fig. 11.

-1

- 1.5
-2

-3

(b)

Fig. 12.

TABLE I
Case

Plant W,(s)

WdS)

i)

s2+5s+4
S3+9S2+24S+20

2(s2+4s+3.75)
s3+6s2+11s+6

+ 9 2 +24s +20

s+4

2(s+ 1.5)
s3+6s2+11s+6

1
s+9s2+24s+20

1
s3+6s2+ lls+6

2
s3+6s2+IIs+6

i
i
)
s3

iii)
iv)

+ +24s +20

s3 9 2

582

IEEE TRANSACTIONS ON AUTOMATIC C O ~ O L VOL.


,

AC-23, NO. 4, AUGUST

(PLANTIWp I s ) =
(MODELIWyIs)-

1978

r+4

s3+

9~2+
24s+ 20

r3+6r2+llr+6

INPUT : S w a m m e
Amplitude: 5 units
FrEqYency
I hz
Llsl :( s t 2 . 5 )

dA-l.--I-r.-I_CA-.

14

IO
12
t i n seconds

16

18

20
-I

1:
-ow

( P L A N T ) Wp

(5

1=

-2IO

12

14

t in seconds

(4

2.6

P1
a

2
S3+9S2+24S+20

II

INPUT: Square wave


Amplitude : 5 units
Frequency = I hz

IO

L ( s ) =(s+1.5Ks+2.5)

signal 2

(PLANT) Wp

(5)

(MODEL)WM(s)=

i
s3+952+245+20

s3+6s2+lls+6

I N P U T : Sum of twelve sinusoids


with frequenciesranging
from . 5 hz to .6 hz
Max. Amplitude : 5 units

L ( s ) = IS+1.5)IS+2.5)

7
6
0

v$

IO
t in seconds

12

14

16

aux.signal I

-I

2
1

-2

t in seconds

(4

Fig. 13.

In cases i) and ii), six parameters are adjusted while in


case iii) only five control parameters are needed. However, in the last case,twelve parameters have to be
adjusted, six of which represent the extra gains discussed
in Section V. To generate the auxiliary signals, F, and F2
are used with characteristic polynomials

REFERENCES

[2]
[3]

h(s)=s2+4s+3.75.
The reference input r(t) in each case was a square wave
with an amplitude of 5 units and frequencies ranging from
0.5 to 6 Hz (Fig. 13).

(4

[4]
[SI
A.

R. L. Carroll and D. P. Lindorff, An adaptive observer for


single-input-single-output linear systems, IEEE Trans. Automar.
Contr., vol. AC-18, pp. 428-435, OCL 1973.
G . Liiders and K. S. Narendra, An adaptive observer and identifier for a linear system, IEEE Trans. Auromur. Conrr., vol. AC-18,
pp. 496-499, Oct. 1973.
K. S. Narendra and P. Kudva, Stable adaptive schemes for
system identification and control-Parts I, 11, IEEE Tram. Sysfs.,
Man, Cybern., vol. SMC4, pp. 542-560, Nov. 1974.
B. D. 0.Anderson, Multivariable adaptive identification, Univ.
Newcastle, New South Wales,Australia, Tech.Rep., June 1974.
S. Morse, Representation and parameter identification of
multi-output linear systems, presented at the 1974 Decision and
Control Cod.

IEEE TRANSACTIONS ON AUTOMATIC WNTFZO~,VOL AC-23, NO.

4, AUGUST
583 1978

J. S. C.Yuan and W. M. Wonham,Asymptotic identification


using stability criteria, University of Toronto, Toronto, Canada,
Contr. Syst. Rep. no. 7422, Nov. 1974.
A. P. Morgan and K. S. Narendra, On the stability of nonautonomous differential equations ; = [ A + B(t)]x, with skew symmetric
matrix B(t), SIAM J. Contr. Optimization, vol.15,pp.163-176,
Jan. 1977.
C. Kim, Convergence studies foran improved adaptive observer,
Ph.D. dissertation, Univ. Connecticut, Storrs, C T , 1975.
P. Kudva and K. S. Narendra, An identification procedure for
discrete multivariable systems, IEEE Tram. Automat. Contr., vol.
AC-19, pp. 549-552, Oct. 1974.
K.S. Narendra and L. S. Valavani, Stable adaptive observers and
controllers, Proc, IEEE, vol. 64, pp. 1198-1208, Aug. 1976.
R. V. Monopoli, Model reference adaptive control with an augmented error signal, IEEE Trans. Automat. Contr., vol.AC-19,
pp, 474-484, Oct. 1974.
K. S . Narendra, M. A. L. Thathachar, and T. S. Baker, Adaptive
control usingtimedelay and correlation, presented at the 3rd
Ann Allerton Con$ Circuit and @stem Theoy, Oct. 20-22, 1965.
R. V. Monopoli and J. W. Gilbart, Modelreference adaptive
control systems with feedback and prefilter adjustable gains, in
Proc. 4th Annu.Princeton Con$ Inform Sciences and Systems, pp.
452-456, 1970.

Sikorsky Aircraft, and Raytheon. Currently, he


is a consultant to AT & T Long Lines Division
on telephone traffic routing using learning thea
entitled
ory. He is the author of book
Frequency Domain Criteria for Absolute Stabiliq
(Academic Press, 1973). His present interests are
in the area of stability theory, adaptive control,
and learning automata
Dr. Narendra is the Chairman of the Adaptation, Learning, and Pattern RecognitionCommittee of the Control Systems
Society,
an
Associate Editor of the International Journal of Kybernetes and the Editor
of the Journal of Cybernetics and Information Sciences. Hewas the
recipient of the 1972 Franklin V. Taylor award of the IEEE Systems,
Man, and Cybernetics Society.

Kmpati S.Narendra received the Ph.D. degree in applied physics from


Harvard University, Boston, MA in January 1959.

.
,

Atpresent,he is Professor of Engineering and AppliedScience at


CT. Hehasconsulted
for numerous
YaleUniversity,NewHaven,
industries during the past fifteen years, including Honeywell,
Inc., Sperry-Rand Research Center, Bell AeroSystems,Schlumberger,

_1

*,

- A

,&
2!j

Lena S. Valavad was born inThessaloniki,


Greece, on February 23, 1952. She received the
B.S. degree in physics (cum laude) from Bamard
NY
College,ColumbiaUniversity,NewYork,
in1972, and the M.S. and M. Phil.degrees in
engineering and appliedsciencefromYaleUniversity, New Haven, CT, in
1974
and 1976,
respectively. Currently, she
completing
is
her
Ph.D. in the area of adaptive control in the
Department of Engineering and Applied
Science, Yale University.

Stationary Linear and NonlinearSystem


IdentificationandPredictorSetCompleteness
PETER E. CAINES, ~

Abstmcr-A general consistency theorem for stationary nonliwar predictlon error estimatom is presented. Since this theorem does not requjre
the existence of a parameterized systemgenerating the observationq it
applies to the practical problem of modeling complex systenrs with simple
parameterizedmodels.Inordertomeasurethequalityofftbetweenaset
of o h e d processes and a given candidate set of predictors, ttbe notion of
predictor set completeness is introduced. Several examples are given to
illushate this idea; in particular, a negative resultconcerning ttae completeness of certain sets of linear predictors is presented. Ibe relationship of
Y W s definitions of identifiability to various notions ofpredictor set
completeness is examioed, and the strong coosistency of mrurimum likelihood estimators for GaussianautoregresSme moving average s
y
s
t
e
m is
obtained via an application of our techniques. Finally,problems for fnture
research ate described.
Manuscript receivedApril15,1977;revisedMarch
12,1978. Paper
recommendedby R. K. Mehra,Past Chairman of the Identifkabon
Committee. This work was supported by the Joint Services Electronics
Pro am under Contract N00014-75-C-0648.
#e author is with the Division of Applied Sciences,Harvard University, Cambridge, MA 02138.

R EEE,

I. INTRODUCTION
N MANY identification procedures, observations are
Itaken on a complex stochastic system such as a paper
mill, a refinery, a brain, or an economy. These observations are then used to build approximate or simple
systemmodelssuch as low order autoregressive moving
average (ARMA)models, or linear state space models
with state spaces of low dimension.
The implications of the use of approximate system models for regulation havebeen
studied byChang
and
Rissanen [ 11. They consider a general least squares control
problem. In thisproblem, linear control laws are permitted to utilize only a limited set of past observations on
the input and output processes of a given linear system. It
[l] that the optimal control laws in any
is shownin
specified class can be characterized by the minimization

0018-9286/78/0800-0583$00.75 01978 IEEE

You might also like