-
. . .. - .. .
--..
..
. .
, . .>
830 IEEE TRANS.4CTIONS ON A T M Y b T I C CONTROL, DECEMBER 1972
r-T
S control
I
Fig. 1. Proposedidentilicationscheme.
situat,ions where one would be unable to generate a better impulse Fig. 1. Error-sampled feedback control system with adaptive sampling.
function in an easier manner would be rare indeed. However, the
idea of using a slowly changing time delay could be useful in the Ti = ti+l - ti.
conventional cro-ss-correlationscheme to obtaina continuous est,imate
of [Link] unknown impulse response. By letting e ( t ) be approximatedby a truncated Taylorseries
e(t) e(ti) + e(ti)(t- t i ) , for t 2 ti,
the criterion (1) can be approximately writ,ten as
Comparisons of Adaptive Sampling Control Laws
T. C. HSIA Upon [Link], ID has the expression
Abstract-Three adaptivesampling control laws are examined
and their properties compared. First, a simpler control law from the
constant difference criterion of Mitchell and McDaniel is derived. This yields the control law
Then, it iscompared with the two similar controllaws given by Dorf,
Farren, and Phillips. The results show that they are close approx-
imations to one another and, furthermore, that they are derivable
from the same integral-difference criterion of different thresholds.
This is the first control law to be considered.
The second control law is
INTRODUCTIOK
In recent years, various results have been reported in the lit,erature
on the subject of adaptive sampling. The problem is concerned with = Tmin, otherwise, (6)
the derivation of a sampling interval algorithm (or control law) in
an aperiodicsampled dat,a system for the purpose of achieving which a a s also proposed by Dorf et al. as an approximation (5).
to
sampling efficiency. A block diagram depictingsuch a systemis The third cont,rol law is concerned with the constant difference
shown in Fig. 1. [Link]
In deriving these control laws, many different. kinds of criteria
le(ti+l) - e(ti)l = Y. (7 1
(mostly [Link]) were used. Clearly, there is a lack of a unified
approach to t,he problem. [Link], these results may not, be Again, Y is a constant threshold. This criterion has been con-
entirely independent from one another. It is then very important to sidered by ;\[Link] and LlcDaniel [Z] t o obtaintheir solution
examine whatever interrelationships might exist, among the control of T , as
laws. Once [Link] interrelationships could be established, one would
be able t.0 classify and compare these control laws on some meaning-
ful basis. The immediate [Link] of [Link] [Link] would
be the gaining of further insight into the charact,eritics and per-
formances of [Link] control laws. The long-term [Link] would be They used the approximation
to lay [Link] [Link] for the development of a unified analytic
approach t.0 control law design.
In this correspondence, three control laws are examined and
compared. The results show that. t,hey are all derivable from the The result (8) is rather complicated to implement.. Also, it is not
same integral-difference criterion with different. thresholds, and compatible with the control laws (5) and (6) due to the inclusion of
hence they are close approximations to one another. <(ti).
I n order to get a closer match t o [Link] control laws (5) and (6),
The CONTROL
Laws a simpler controllawincludingonly i ( t i ) is now derived for the
criterion (7) as follows. Take the [Link] for e ( t ; ) as in
Two of these control l a a s were proposed by Dorf et al. [ 11. They
(2):
considered the constant integral-difference (ID) criterion
+ e'(ti)Ti.
ID = IllG1 [Link] (7) becomes
t(ti+l) E e ( t i )
Ii(ti)lTi = Y ,
where c is a constant [Link]. Now define Ti as
which yields the desired control law
Manuscript received Febnlarv 15 1 9 i 2 : revised July 18, 19i2.
The author was on sabl>atical'lea;e at Siemens .\G, Karlsruhe, IVest Germany,
from the Univemity of California, Davis. He is no? with, the Department of
Electrical Engineering, University of Califorma. Davls, Callf. 95616.
TECHNICAL NOTES AND CORRESPONDENCE 831
t
ID
/
/ Ti - Tnax
Ti t
0'
Fig. 2. Comparisonoftheintegraldiiierence (ID) for [Link] control laws.
This resultispracticallymore [Link] [Link] (8) duetoits
simplicity and better immunity from noise disturbances. Now the Fig. 3. Anot,her comparison of the control laws for their Ti versus ( . ( W
relationships.
control laws (5), ( 6 ) ,and (9) are examined below.
OF THE CONTROL Laws
COMPARISONS REFEREXCES
Dorf et al. have shown that control law (6) is an approximation
of (5) because the ID corresponding t o ( 6 ) ,
. . :I1 and JV. L. McDaniel. Jr., "Adaptivesamplingtechnique,"
I E B E Trans. Automat. Contr. (Corresp.), vol. AC-11,pp. 200-201, Apr.
1969.
has similar [Link] the IDcorresponding t o (5)
On the Reconnaissance Game
Both I D funct,ions are shown in Fig. 2. DONALD E. PL$NTE AND ALEXANDER H. LEVIS
T o comparecontrol law (9) with (5) and ( 6 ) , onecan simply
evaluate the I D function for (9) as
Abstract-The reconnaissance game is apursuit-evasion game
with state constraints in which the evader tries to reconnoiter a
targetfromas close a distance as possible, while guaranteeing
safe escape from a [Link] capability
ID9 is also [Link] in Fig. 2. but he also has a territorial constraint. The case of complete ma-
The curves in Fig. 2 show t,hat.ID6 and ID, have approximately neuverability and speed advantage with the pursuer is analyzed
the same general [Link] as IDS. Within t,he range T,i, 5 and optimal open-loop controls are found underthe assumption
Ti 5 T-,, ID6 and ID9 areperturbations of the [Link] IDS. thatthereconnaissancegame can be decomposedintoa target
While ID, tends toincrease its value as i ( t ) increases, ID, decreases game and an escape game. The cases of equal speed and speed
monotonically. The overall similarit.y of the three curves indicates advantage with the evader arediscussed briefly.
that. the three cont.ro1 laws can be regarded as [Link] of
one [Link]. Like control law (6), the nerr control law (9) is simpler
than (5) in instrumentation. For furt,her comparison, the control INTRODUCTION
laws (5), ( 6 ) , and (9) are plott,ed against. e ( t ) as shown in Fig. 3.
The reconnaissance game is played in the plane by [Link] players
Their similarity is again clearly evident.
with a [Link]. a t t h ecenter of a rect,angular coordinate [Link] (sy).
The resultsin ( l o ) , ( l l ) , and (12) reveal anotherinteresting
Each player has a single, bounded, not necessarily continuous con-
property for the control laws. These equations show that one can
trol. We assume that. both players have complete maneuverability
derive thethree controllawsfrom thesame integral-difference
and theevader travels a t unit speed, and [Link] pursuer a t speed cy.
criterion by using different thresholds. That is, a common design
The pursuer dynamics are given by
criterion for all threecontrol laws exists as
2 p = a cos 11; z&) = Xpo
LIl: [e(t) - e(k)] dtlI = g YP = 01 sin u ; YP(t8) = Ypa. (1)
where g isathresholdfunction. When g assumes the [Link] The evader dynamics are given by
g = IDS, IDS, ID,, as shown in Fig. 2, the cont.ro1 laws (5), (6),
and (9) can be [Link], respect,ively, by way of the approximation & = cos v ; x*(ta) = r#l
(2 ). ye = sin o; y.(to) = ye@. (2)
CONCLUSION
The study has shorn that,among various control laws, there are Manuscript received June 13. 1972; revised Aug. 7. 19i2. This note is part. o f a
dissertation submitted by D . E. Plante t o the Departmeut of Ele.ctrica1 Engl-
t,hree t,hat. are closely relatedin terms of the similarity of [Link] neering, Polytechnic Institute of Brooklyn. Brooklyn, N. Y., in partlal fulfillment
design [Link] significant resultis that the int.egra1-dif- of the requirements for the Ph.D. degree, June 1971.
D. E. Planteiswith Aerospace Electronics Systems, General Electric Company,
ference criterion is shown to be more general and useful [Link] the Utica, N . Y. 13503.
A . H. Levis is with the Department of ElectricalEngineeringand Electro-
difference criterion. physics, Polytechnic Institute of Brooklyn, Brooklyn. h'. Y. 11201.