IEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL AC-24,NO.

2, AFWL 1979 369

G(x) may have in (- r , r )three roots. It has exactly three roots if If s=o then
r
o<-q<y.

In this case for any integer n different from 0 and - 1 there is in the
interval (nr,( n + 1)r) exactly one real root of G(x) =0. Thus fork > 1 we [l] R Bellman and K. L. Cooke, Diflerenrinl-D@ereme Equariom. New York:
have 4k+ 1 roots in the interval [ -2kr,2kr]. If 1- p < 0 then G’(x) has Academic, 1963.
[2] I. S. Poutriagin, “On the zeros of some elementary transcendentalfunction%” Izn.
no roots in (- r / 2 , ~ / 2 ) . Thus G(x) may have at most two roots in any A M . Nuuk SSSR, Ser. Mal., vol. 6, pp. 115-134, 1942.
interval [n2r,(n + 1)2r]. T i s is less than is required by Theorem 1.
Now, we have to show that G’(x)F(x) > 0 for any rootof G(x) if and
only if (3) holds. For any x satisfying the relation G(x)=O the situation Stabilization of Linear Systems with Time-Varying Delay
exists that G’F may take only one of the two following values:
MASAO IJSEDA AND TOMONORI ASHIDA

I
G ‘ ( x ) F ( x ) = E-(I--$$)-$(I+-&)]

[P- +q] (7)

Abstruc-me problem of stabilizing a h ear system with time-varying
delay by means of linear feedback witbout delay is considered A sufficient
mdition on the open loop system is presented for the possibility of
adnalizing the a r b i i y prescribed degree of stabiity m tbe closed loop
where P = 5 1. system. Under the condition, the feedback gain can be constructed from
We denote by xI,x2 two roots of G(x) closest to zero and let lxll < the knowledge of the upper bounds of tbe delay and its derivative; no
Ix21. Note that xl,xzE(-r/2,r/2) and additional informationon the delay is necessary.

I. INTRODUCTION
G’(xl)<O, G‘(x~)>O (9) Timevarying delay is seenin various engineering systems[ 11-[4], and
the existence is frequently a source of instability 151. There are unstable
and time-varying delay systems which are stable for any constant delay less
than the maximum of the varying delay.
This note -considers the problem of stabilizing a linear system with
time-varyingdelaybymeans of h ear feedbackwithoutdelay. The
Sincep > O then e = + 1 for x = x j i = 1,2. Thus, using (9) we have object is to find a sufficient condition on the open loop system for the
possibility of actualizing an arbitrarily prescribed degree of stability in
F(x,)=Vx:+p2-s2 +q<o the closed loop system. Such stabilization problems have been studied
extensively for time-invariant delay systems [6]-[lo].
F(x2)= d m +q>O.
11. SYSTEMDE~CRIP~ON AND STABILI~~ DEGREE
Therefore, condition 3) is necessary.
Now we assume that condition 3) is true. Using the relations (8), (9, Weassume thatthelinear systemwith timevarying delay has a
(9) we have for any root x,of G: representation
x(t)=A,(t)x(t)+A,(t)x(t-h(t))+B(t)u(t) (1)
if C = + 1then F(Xk) >F(XJ>o and G’(Xk) >o for xkx2 <o,
G‘(Xk) G‘(X3 >o for XkX2 > 0 where x(t), which is assumedto be measurableat every t, is a vector,and
u(t) is the input vector. Ao(t), A l ( t ) , and B(t) are matrices with a p
if C = - 1 then F(Xk) <F(X&<o and G’(Xk)<O for XkX2 >o, propriate dimension all having continuous elements. h(t) denotes the
G‘(xk)< - G’(x3 < 0 for xkxz> 0. delay, and it is assumed that there exist positivenumbers h, and hs such
that
Thus, we always have G‘(xk)F(xk)> 0 for any root x, ofG. The above
O<h(t)<h, and h’(t) <hs< 1 (2)
considerationslead to the following.
hold for all t.
Criterion Remark: We do not care about the cause of the variation of the delay.
The result of this note holds independently of the cause.
System i(t)=Alx(t - T ) is exponentially stable with decay rate y if Consider linear feedback of the form
and only if:
u(t)=K(t)x(t) (3)
1) y7< 1;
2) Isl< -y7sinyo+yocosy0 whereyoE(0,r/2) is the root of 1 - y = where K ( t ) is a matrix with appropriate dimension having continuous
xtanx; elements. Then the closed loop system is represented by
3) d x f + ( y ~ ) ~ - < s ~- q < v m where lxll <Ix21; i(t)=(A,(t)+B(t)K(t))x(t)+A,(t)x(t-h(t)). (4)
x l , x 2 € ( - r / 2 , ~ / 2 ) and In this note, we derive a sufficient condition for the feedback gain
-xicosxi+y7sinxi+s=0, i=1,2 K ( - ) to existsuch that (4) hasthe prescribed stability degree. The
definition of the stability degree is due to the following.
where a = q + is is any eigenvalue of A multiplied by .re yT Definition: Equation (4) is said to have the stability degree h if there
exists a positive number c such that any solution x( .) of (4) satisfies
Example
Ilx1*11~cllx,,llexP(-X(~2-tl)) (5)
Consider y=O. Then y,=0.86033351 and we obtainthe necessary
condition ~s~<0.56109642=yocosyo=dfso. If, for instance, s=0.561 <So,
then we have the followingstability area for thereal part q of a complex Manuscript received May 18, 1978.
number a: M. Ikeda is with the Deparrment of Systems Engineerin& Kobe University, Kobe,
Japan.
T. Ashida is with the Production Engineering Division. Matsushita Electric Industrial
0.6394952 < - <0.66491 154. Company Ltd., Kadoma, Japan.

0018-9286/79/0400-0369$00.75 01979 IEEE

370 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. AC-24, NO. 2, APRIL 1979

for all rI and r2 ( >rl), where From these inequahties,it can be shown that the solution of (IO) satisfies

Ix(~)I~~ll~l,IIexP(-~(~-~I)) ( 14)

for all rI and T (> rI). where E ((a2+h u ) / u l ) 1 / 2Hence

.

Ilx,211= SUP Ix(.)I

t2-hOC7<t2
111. STABILJZABILITY CONDITION
SUP fll~l,llexP(--h(~--l))
The result of this note is summarized as follows. t2-hm<r<t2

Theorem: If 1) ( A , ( . ) , B ( - ) ) is auniformlycompletelycontrollable
=cexP(~,)llxl,llexP(-X(r2-t,)) (15)
pair [ I 11, and 2) thecolumns of AI(r) can berepresented as linear
combinations of the columns of B(t) at every t , then for any specified for all r l and r2 ( >II).
stability degree, there exists a feedback gain K(.)which actualms the Thus the feedback gain K(.)defined by (9) actualizes the specified
degree in the closed loop system (4). stability degree X in the
closed
loop
system
(10). Q.E.D.
Remark: The proof of the theorem implies that the feedback gain can
be constructedwithoutthepreciseknowledge of thedelay h ( . ) ; it is IV. CONCLUSION
enough to knowtheupperbounds ha of thedelay and ha of the
derivative. We have obtained a sufficient condition for the possibility of actualiz-
The following lemmas are necessary to show the theorem. ing an arbitrarily specified stability degree in a linear time-varying delay
Lemma I [12]: If ( A , ( - ) , B ( . ) )is a uniformly completely controllable systembymeans of linearfeedbackwithoutdelay. The stabilizing
pair, then ( A o ( . ) + M , B ( . ) ) ,where X isanyrealnumber and I is the feedback stated in this note bas the merit that it can be realized even if
identity matrix, is also a uniformly completely controllable pair. the time-varying delay is not known precisely.
Lemma 2 [12]: If ( A o ( - ) + M , B ( - ) is ) a uniformly completely control-
lable pair, then the matrix Riccati equation
REFERENCES

[I] L A. Zadeb, “Operational analysis of variable4elay system” Proc. IRE, vol. 40,
pp. 564-568. May 1952.
121 S. G . Margolis and J. J. O’DonneU. -Rigorous treatments of variable time delays,”
IEEE Tram. Electron. Cornput.. vol. EC-12. pp. 307-309, June 1963.
has a solution which satisfies [3] 1. P. Seddon and R. A. Johnson, T h e simulation of variable delay,” IEEE T r m .
Compur. (Short Notes), vol. C-17, pp. 89-94, Jan. 1968.
[4] R. A. Johnson. “Functional equations approximations, and dynamic response of
n,Z<P(r)<azI systems with variable time delay,” IEEE Tram. Auioma~.Conlr. vech. Notes and
Corresp.). vol. AC-17, pp. 398-401, June 1972.
for all r, where al and a2 are positive numbers. [SI Y.Sato, “A study on stability and stabilization of a time or state-variable lime lag
system,” M. Eng. thesis (in Japanese), Dep. Syst. En& Kobe Univ., Kobe, Japan,
Proof of Theorem: Assumethattheconditions I) and 2) of the 1978.
theorem are satisfied. Then there exists a matrix Q( .) such that [6] Y. S . Osipov, ‘Stabilization of controlled systems uith delays,” D~flerenrial€pa-
tiom. vol. I, pp. 463473, May 1965.
[7] D. W. Ross and I. Flugge-Lux “An optimal control problem for systems with
differential-difference equation dynamics,” SIAM 1.Comr, vol. 7, pp. 609-623,
Nov. 1969.
and, from Lemmas 1 and 2, a matrix P ( - ) exists which satisfies (6) and [SI G. J. Nazaroff, “Stabiliiry and stabilization of linear differential delay systems,”
(7). IEEE T r m . Automar. Comr. pech. Notes and Corresp.), vol. AC-IS, pp. 317-318,
June 1973.
Define the feedback gain K(.)for any specified stability degree X by (91 G . J. Nazaroff and G . A. Hewer. “Stabilization of linear, autonomous differential
delay systems.’’ IEEE T r m . Auromr. Concr. vech. Notes and Corresp.), vol.

;(
AC-18. pp. 673-674, Dec. 1973.
K(t)= - - I + [IO] K. Ryo. M.Ikeda and S . Kitamura, “Stabilization of linear systems with delay: An
exp(mu)
I-hp Q(r)QT(r))BT(r)P(r). (9)
~

extended stabilizability condition” (in Japanese), Tram.SICE, vol. 13, pp. 72-226,
June 1977.
Then the closed loop system (4) is described by [ I l l R. E Kalman, “Contributions to the theory of optimal control,” Bo[. Soc. Mac.
Mexicma, vol. 5, pp. 102-119, 1960.
[I21 M. Ikeda, H.Ma& and S. Kodama, -Stabilization of linear systems,” SIAM J.
Conrr, vol. IO, pp. 716-729, Nov. 1972.

Proportional Minus Delay Controller

I. H. SUH AND 2.BIEN

Abstmcr--A new type of controller, which utilizes the timedelay effeet,

is proposed It is shown that the conventional Peontroller quipped with
no appropriate time-delay performs an averaged derivative action and thus
can replace the PDeontroller, showing quick responses to input cbmges
but b e i i insensitive to high-frequency noise.

I. INTRODUCTION

In many industrial processes, the presence of time delay often causes

various practical difficulties in the system control and a great deal of
research efforts have been devotedto solving the problem of controlling

Manuscript received October 3. 1978.

The authors are with the Department of Elecmcal Science, Korea Advanced Institute
of Science, Seoul, Korea.

70S00.75 81979 IEEE