Citation
A controller design method which applies to time varying linear systems

Material Information

Title:
A controller design method which applies to time varying linear systems
Creator:
Koenig, Kurt Walter, 1967-
Publication Date:

Subjects

Subjects / Keywords:
Acceleration ( jstor )
Aircraft pitch ( jstor )
Automatic pilots ( jstor )
Eigenvalues ( jstor )
Linear systems ( jstor )
Missiles ( jstor )
Parametric models ( jstor )
Simulations ( jstor )
Trajectories ( jstor )
Velocity ( jstor )

Record Information

Rights Management:
Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
20381782 ( ALEPH )
32490049 ( OCLC )

Downloads

This item has the following downloads:


Full Text










A CONTROLLER DESIGN


METHOD WHICH APPLIES TO TIME VARYING
LINEAR SYSTEMS


By

KURT WALTER KOENIG


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


1994













TABLE OF CONTENTS




A B ST R A CT . . . . . . . . . . . . . . . . . .

CHAPTERS

1 INTRODUCTION ...............................

1.1 Earlier W orks . . . . . . . . . . . . . . . .
1.2 P purpose . . . . . . . . . . . . . . . . .

2 THEORY BEHIND THE DESIGN METHOD ..................

2.1 A Geometric Inerpretation of Lyapunov's Linear Stability Theorem .
2.2 Adding Control to the System ......................
2.3 A Linear Feedback Set to Control xTp ................
2.4 One Linear Feedback Matrix to Control xTpk . . . . . . .
3 A TIME VARYING SECOND ORDER EXAMPLE . . . . . . .

4 DERIVATION OF A MODEL OF THE EMRAAT MISSILE ........

4.1 The Nonlinear Model ...........................
4.2 The Linear M odel .............................

5 THE DEPENDENCE OF GAINS ON FLIGHT PARAMETERS .....


Generating p and V from M and Q . .
The Flight Parameter Generator . . .
Initializing the Iterative Lyapunov Design
The Iterative Lyapunov Design Method .
Formulation of a State Tracker . . .
Comparing Gains with Flight Parameters


Method .......


6 COMPUTING LOOK-UP TABLES ......................
6.1 Determining a Grid of Points ......................
6.2 Formulation of :he Design Constraints ...................
6.3 Generating the Look-Up Table ......................









7 GAIN SCHEDULING .....................


7.1 C urve Fitting . . . . . . . . . . . . . . . 83
7.2 Testing the Fit . . . . . . . . . . . . . . .. .. 83

8 NONLINEAR SIMULATIONS . . . . . . . . . . ... .. 88

8.1 The Nonlinear Simulation . . . . . . . . . . ... .. 88
8.2 A Test of State Tracking . . . . . . . . . . . .. .. 90
8.3 Simulation of Flight Scenarios ... ..... . . . . . . . . .. 93

9 CONCLUSION . . . . . . . . . . . . . . . ... .. 104


APPENDICES

A AERODYNAMIC DATA FOR THE EMRAAT AIRFRAME ........ 106


B INERTIAL DATA FOR THE EMRAAT AIRFRAME ........... 117


REFERENCES ................................... 118

BIOGRAPHICAL SKETCH ............................. 120


. . . 83













Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
A CONTROLLER DESIGN METHOD WHICH APPLIES TO TIME VARYING
LINEAR SYSTEMS
By
Kurt Walter Koenig
August 1994
Chairman: Dr. Thomas E. Bullock
Major Department: Electrical Engineering

A feedback controller design method has been formulated which applies to linear

time varying systems. The motivation behind this technique is the design of an

autopilot for an air-to-air missile. The missile under study is the extended medium

'range air-to-air technology (EMRAAT) missile, a theoretical bank-to-turn missile

which is under study by the United States Air Force. Knowledge gained by this

missile will be applied to future bank-to-turn missiles.

Conventional autopilot design techniques use pole placement, linear quadratic

regulators, linear quadratic Gaussian/loop transfer recovery techniques, or eigen-

structure assignment methods. These techniques are applied to a linear model which

depends on time varying flight parameters. In applying these methods, the false as-

sumption is made that the linear model is slowly varying. Since the system is changing

rapidly, no theoretical basis exists to support the success of the resulting controller.

The proposed design method finds state feedback gains which cause the closed loop

linearized system to be stable with respect to a specified Lyapunov function. Even

though flight parameters change rapidly, local stability around the operating point is

achieved.









The design algorithm finds feedback gains which place the eigenvalues of the

derivative of a given Lyapunov function. Limitations on placement of these eigenval-

ues are stated. This concept is applied to a second order linear time varying system

where ordinary pole placement techniques fail. The design method is then applied

to a linearized model of the EMRAAT missile which is a function of time varying

flight parameters. Feedback gains are generated as a function of these flight param-

eters. It is discovered that the gains depend on dynamic pressure, mach number,

and angle-of-attack. A combination of interpolation and polynomial fitting is used

to create a look-up table for the gains. The resulting controller is programmed into

a nonlinear simulation which runs missile and target scenarios. Small miss distances

are achieved.













CHAPTER 1
INTRODUCTION

The original goal of this dissertation was to formulate an autopilot design method

which stabilizes the flight of a bank-to-turn air-to-air missile. The result was the de-

velopment of a controller design method which applies to time varying linear systems.

When used on a nonlinear system, local stability is achieved. The remainder of this

chapter gives a brief description of the missile used in this study, a summary of related

works preceding this study, and the purpose and outline of this dissertation.

An air-to-air missile is launched from a military aircraft with the intent of in-

tercepting an enemy aircraft or target. Bank-to-turn (BTT) missiles have wings

giving them greater maneuverability than the conventional skid-to-turn (STT) mis-

siles which have fins. Thus, the target must work much harder to evade a BTT

missile. The dynamics of BTT airframes are highly unstable, making for a difficult

controls problem.

The missile under study in this paper is the extended medium range air-to-air

technology (EMRAAT) missile as shown in Figure 1.1 [1]. The EMRAAT missile

is a theoretical missile formulated by the Air Force with the intent of studying the

feasibility of the BTT concept. The knowledge gained from the study of this missile

will be used by the Air Force if it ever decides to make a real BTT missile.

The EMRAAT missile is equipped with a seeker. If the seeker is infrared based,
then it measures the line of sight angles to the target. In the case that the seeker is

radar based, then, in addition to the line of sight angles, the range and range rate

of the target are measured. This information is passed to the guidance law which

determines the desired normal accelerations needed to intercept the target. The



















cross section, constant -


Start -
wedge
fairings


7.5 din


All dimensions are in inches


V A-A
View A-A


Figure 1.1. The EMRAAT missile [1].


ZZ










autopilot attempts to achieve these accelerations by applying the proper control to

the control surfaces.

1.1 Earlier Works

Early autopilot designs separated the missile dynamics into two or three lower

order models. These models were linearized, and classical techniques were used to

stabilize them. Other designs applied more advanced techniques to higher order lin-

earized models. Some of these techniques are pole placement, linear quadratic regula-

tors, linear quadratic Gaussian/loop transfer recovery techniques and eigenstructure

assignment methods [2] Feedback controllers would be designed as a function of the

linearized models. Gains would then be scheduled against the flight parameters on

which the models depend. Gain scheduling is a popular technique used in control

designs and has been studied in depth by Shamma, Athans, and Cloutier [3,4,5,6].

The problem with these design methods is that they assume that the system is
time invariant or slowly varying. The linearized models depend on rapidly changing

flight parameters. Although these methods often work, no theoretical basis exists

to support the success of these designs. More recently, H, and yi synthesis design

techniques have been used to formulate one dynamic controller which would stabilize

the system for all modeled uncertainties [7,8]. A disadvantage of H.. controllers is

that they have very high orders. Also, the existence of a robust H"" controller which

satisfies the performance requirements is not guaranteed. The only way to determine

the stability of a given design is to test it using a nonlinear simulation.

Desoer [9] states if a given time varying system

x= A(t)x (1.1)

is stable for each t when t is frozen, then it is not necessarily stable when t is allowed

to change. However, (1.1) is stable if A(t) changes slowly. A conservative upper












Alpha vs. Time
12, ,,
12o ---i--" --------'- --- ---------
10-

8

S6
12
4

2

0-


0 1 2 3 4 5 6 7 8
Seconds

Figure 1.2. Angle of attack from one trajectory of the EMRAAT missile.

bound on supt>o IIA(t)lI is given which, if enforced, guarantees asymptotic stability

of (1.1). As will now be shown, this limit is too restrictive for the EMRAAT missile.

Figure 1.2 shows the angle of attack of the trajectory of a missile flown using an

already existing autopilot. Desoer proposes the Lyapunov function


V(x, t) = xT(61I + P(t))x (1.2)


where ei > 0 and P(t) is chosen so that


AT(t)P(t) + P(t)A(t) = -31 (1.3)


Desoer gives a bound on V,

V < xTx[-3 + 2elaM + aM3m4 (1.4)


where the following definitions are made:


aM :=sup IA(t)|| < c (1.5)
t>0










aoo is positive and

Re Ai[A(t)] < -2c7o Vi, Vt > 0 (1.6)

m is a constant and depends only on ao and aM so that

|exp(rA(t))|l mexp(-oCor), Vr > 0 Vt > 0 (1.7)

Also,

iaM:=sup IIA(t)l. (1.8)
t>0
If c is allowed to be very small, then from (1.4), stability would result if
4o.2
M < -- (1.9)

For the linear pitch model of the given trajectory, o0 = 15.75. At .04 seconds into
the trajectory the closed loop matrix is

A -3.2533 .8997 (1.10)
-1678.2 -83.50
At this instant in time, m = 22.88 and aM = 2231. 42 = .0036. From this we see

'that the linear system is changing much faster than the limit shown in (1.9). The
system is not slowly varying. Note that m was only found for t = .04 sec. and not
for all time. If a greater m were found for the rest of the trajectory, then the limit in
1.9 would be even more restrictive and the result would be the same. This test does
not show that the system is unstable. In fact, a Lyapunov function is known which
shows that this linear system is stable but Desoer's theory can not support this fact.
Wilson, Cloutier, and Yedavalli [1] give a stability condition for a constant linear
system with time varying uncertainties. Given a system,

x = [A+ E(t)]x (1.11)

if a positive definite P can be found which solves the Lyapunov equation,


PA + ATP +21 = 0


(1.12)











Alpha vs. Time
12


10


8-

a)
6-




4-/
0



& 3



0 1 2 3 4 5 6
Seconds

Figure 1.3. Angle of attack from one trajectory of the EMRAAT missile.

then the system will be asymptotically stable if


,max[E(t)] <,, 1 Vt (1.13)
7maz.(P)

where 0,max is the maximum singular value. This idea is initially interesting because

it requires the error, E(t), to be bounded, but does not constrain the time variation.

The above result can not be easily applied to the EMRAAT missile. This will be

shown by applying the result to a trajectory of the EMRAAT missile whose angle of

attack is shown in figure 1.3. In order to check the stability of the EMRAAT missile,

A must be chosen since it may be any matrix which is stable. It is believed that the

best choice of A should come from a closed loop matrix in the trajectory. With this

in mind, A was taken at t = 2 sec into the trajectory.

A-B '-4.477 .9526 (1.14)
A- BK = -1.684 -79.47 (











Sigma max[E(t)] vs. Time
2 5 0,,


200-


Fiso

E
150-
*a-
E
05100-


50



0 1 2 3 4 5 6
Seconds

Figure 1.4. A plot of axB[E(t)] versus time.

E(t) is the result of subtracting this matrix from the closed loop system from the

rest of the trajectory. Solving for P in (1.12) gives

[ 17.29 -.0454 (1.15)
S -.0454 .012 (115

Computing the upper bound in (1.13) gives

S= .0578 (1.16)
am)axj4P)
Figure 1.4 shows the plot of crmax[E(t)]. Obviously since marn[E(t)] is greater than

.0587 for most of the trajectory then (1.13) is not satisfied. This procedure was

repeated by taking the constant closed loop matrix from every point in the trajectory,

and the result was the same. The time varying linear system from this trajectory is

known to be stable because a Lyapunov function exists which can show this. The

upper bound in (1.13) can not be easily used if at all to support the claim of stability.

Desoer [9] gives a stability limit on the rate of variation of a given closed loop

time varying system. Wilson, Cloutier, and Yedavalli [1] give a stability limit on










the size of a time varying uncertainty. Both of these tests were found to be too

conservative when applied to an already existing stable autopilot, and the results
were inconclusive. In addition, these tests are analytic tools. As yet, no design
procedure exists which can give a closed loop linear system that will pass either
test. It would be desirable to formulate a design method that can yield a closed
loop time varying linear system which satisfies a less conservative stability condition.
Vidyasagar [10] gives two important results. The first can also be found in Hahn [11]

and says that given a time varying system


S= A(t)x (1.17)

if a positive definite matrix P can be found so that the matrix


PA(t) + AT(t)P (1.18)

is negative definite for all time, then the system is asymptotically stable.
The second result allows the first result to be applied to a nonlinear system. Given

a nonlinear system of the form

S= f(t,x)

where

f(t,O) = 0

and f is continuously differentiable, then let

A(t) = 8f(t, x)
ox x=O
and assume that
IIf(t,x) A(t)xl
lim sup = 0
llxll-o t>0 ||x










and A is bounded. If 0 is an asymptotically stable equilibrium point of i = A(t)z for
all time, then 0 is a locally stable equilibrium point for the system

5 = f(t,x)

These ideas can be applied in the following way. Given a nonlinear system

x= f(x,w,u) (1.19)

define

A(x,w) = B(x,w)= (1.20)
xw .(XW)nomna (X,W)nominai
where x is a vector containing the states and w is a vector containing additional
system parameters. At regions near the operating point, the system becomes close
to

Ak = A(x, w)Ax + B(x, w)Au (1.21)

where Ax and Au are small perturbations between the states and inputs and the

operating point. We would like to find a feedback control law,

u = -K(x,w)x (1.22)

so that

P[A(x, w) B(x, w)K(x, w)] + [A(x, w) B(x, w)K(x, w)]TP (1.23)

is negative definite for all x and w where P is positive definite. If such a control law
is found, then the perturbations from the nominal trajectory are locally stable for

the system

= f(x,w, -K(x,w)x) (1.24)

Shahruz and Behtash [12] give one control law which places some of the eigenvalues
of (1.23) for the case where P = I. However unnecessary limitations on where the










eigenvalues can be placed exist. Also, while many feedback control laws make (1.23)

negative definite, Shahruz and Behtash give only a small subset of these control laws.

1.2 Purpose

This paper gives an algorithm which can stabilize a linear time varying system by

placing the eigenvalues of (1.23) between desired bounds and has fewer limitations

than the control law in Shahruz and Behtash [12]. Limitations on the placement of

these bounds are stated. This algorithm applies to all systems for which a constant

positive definite P exists so that (1.23) can be made negative definite for all time.

First, theory is presented leading to the formulation of this algorithm. Then, to

demonstrate the usefulness of this algorithm, it is applied to a linear time varying

system where normal pole placement techniques fail. This algorithm is then applied

to the EMRAAT missile. A nonlinear model is made, and from this, a time vary-

ing linearized model is generated which is a function of several flight parameters.

Gains are computed, and their dependence on flight parameters determined. A gain

scheduling scheme is then implemented yielding local stability around the operating

point. The resulting design is tested in a nonlinear simulation. Finally, the results of

this test are given, and the usefulness of this design technique is evaluated.












CHAPTER 2
THEORY BEHIND THE DESIGN METHOD

The material in this section gives the theory leading to the proposed controller
design method. The first section presents a geometric interpretation of Lyapunov's
linear stability theorem. The effects of control on the velocity field of a system is
studied in the next section. The nonemptiness and convexity of the set of all feedback
gains which stabilize a system with respect to a given Lyapunov function are then
discussed. Finally, an iterative procedure that finds one element of this set is given.

2.1 A Geometric Interpretation of Lvapunov's Linear Stability Theorem

To understand why time invariance is a necessary assumption for eigenstructure
design methods the following second order linear time varying system was studied.
The example given now is from Vidyasagar's example 5.3,109 [10] and can also be
found in Khalil [13]. Given the following system

S= A(t)x (2.1)

where
[ -1 + a cos2(t) 1 a sin(t)cos(t)
A(t) -1 a sin(t) cost) -1 + a sin2(t) (2.2)
Vidyasagar [10] notes that the transition matrix is given by
P(t0) e(a-1)tcos(t) e-tsin(t) 1
q^(t,0)-= --e(a-1)sin(t) e-* cos(t) (2.3)

and the characteristic equation is

A2+(2-a)A+(2-a)=0 (2.4)

The roots of (2.4) have negative real parts for 1 < a < 2. The exponents in the first
column of D(t, 0) indicate, however, that the system is unstable for these values of a.
11











b
1.^ *- z Y \\


1r\
.-0.5 _\




-10 0 5 10 41 -1 -. 0 0-5 1
X- X1

Figure 2.1. The trajectory of (a) the system (2.1) and (b) a frozen system.

This system was simulated using MATLAB with a = 1.5 and xo = [1, 0]T. The

eigenvalues of A(t) are -!+ s7j and -- j Figure 2.la shows the state trajectory

of the system 2.1 where X, is assigned to the horizontal axis, and x2 is assigned to

the vertical axis. This plot demonstrates the instability of the system. If the system

were frozen, i.e. A(t) = A(0), then the stable trajectory of Figure 2.1b would result.

This trajectory is shown with the velocity vector field of the frozen system.

To explain the instability in Figure 2.la, we will now discuss the time varying

velocity field of equation (2.1). The plots in Figure 2.2 show the trajectory of the

time varying system in the state plane for eight instances in time. For each plot, the

velocity vector field for that instant in time is superimposed on to the trajectory. Each

plot in Figure 2.2 shows the boundaries of four pie-shaped regions which this paper

refers to as positive and negative regions. Two positive regions exist that are defined

as the set of all points whose velocities have an outer radial component, i.e. each of

these velocities have a component pointing directly away from the origin. Likewise,

there are two negative regions containing velocities with inner radial components.

The plots show that the boundaries separating the positive and negative regions

rotate in a clockwise direction. Also, the current position in the trajectory remains










inside one of the positive regions. Since the velocity of the system always has an

outer radial component, the magnitude of the state vector is always increasing so

that the system (2.1) is unstable.

A direct relationship exists between the positive and negative region boundaries

and the eigenvectors of the system. In this case, the eigenvectors are rotating clock-

wise at the same angular velocity as the region boundaries. It can be shown that if a

system has constant eigenvectors with time varying eigenvalues which have negative

real parts for all time then the system is stable. The stability question, however, is

not as easy to answer for the case with moving eigenvectors.

If the system has no positive regions, then it will be stable regardless of how

quickly it changes. However the converse is not true in general. Figure 2.3 shows a

state vector and its corresponding velocity for some dynamic system. The angle 9

can be computed in the following way.


XTk
cos(0) = xl (2.5)

For k to have an inner radial component then

S< 0 < (2.6)
22

This is true when

xT < 0 (2.7)

where

k = Ax (2.8)

So if xTAx is negative everywhere for all time, then the system is stable. Since the

system is linear, it is sufficient to check the sign of xTAx on the unit circle. If all

velocities on the unit spheroid point inside that boundary, then the same is true for





















10 S \


4 ) 4 ^ '\

0 F. -' >^ \




-5o
-10 \ \ 0 10 is



151
-5 -10 .5 0 5 10 1I

XI



is 7 -A 1-
107





15 -1 0 5 1
.S ,' />




10 7 Ic j / /

-1 -10 *I 0 5 10 15
XI



10 -, ^ '- N


4 -4 -4 - ,- *-i *^





s-10


-s.. " 4-^ f 4-




-1 -10 -3 0 5 10 1S


4
15 -



















-1 -0 5 4 O 1
10

--^ -4 --
5^ -^ 'p x S.

0 '- 05 $ 0 i$

,'F -~ F? 4\ 4 ^


-10
4* 4- -4 - <\ <- -' /
-15 -
-15 .10 -5 0 S 10 15
xi

is

'F 4 ^

-0' *, 4 ^i \ \ .




"' \ ^ C


S 4L
-,.~I ^. >^-v

-10. \ I\ t 4
\ \ I\ \ I s

-15 -10 .5 0 5 10 15
XI

4
15 1


10

*^ ^ 4 4 4 ,






.5-
C*y 4- 4-^


-151 is/


XI


Figure 2.2. The trajectory and velocity field of (2.1)








15



2

1.5



0.5
0/


-0.5
-I

-1.5

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
xl

Figure 2.3. x and x

all spheroids centered on the origin. This leads to the following stability condition

which is given without proof.


Stability Condition 1: If xTic = xTAx < 0 Vx : XTX = 1,V t > 0 for

the system, 5 = A(t)x, then the origin is a global asymptotically stable

equilibrium point.


Figure 2.4 illustrates an example of a second order system which meets Stability

Condition 1. Note, that while all figures given so far represent second order linear

systems, the above stability condition applies to linear systems of any order.

The above stability condition can be made less conservative by expanding the

class of shapes to ellipsoids defined in the following way.


xTPx = 1 (2.9)

where P is a symmetric, positive definite matrix. Figure 2.5 shows an ellipse of the

form (2.9) defined for a second order system. The state vector, x, is drawn to some
































-1 -0.5 0 0.5 1


Figure 2.4. An example of a
ponents


second order system with negative radial velocity corn-


Figure 2.5. One velocity vector on the ellipse xTpx = 1


0










point on the ellipse, and its velocity, :, is also shown. The normal to the ellipse at x
is Px. To ensure stability, it is sufficient to require that the projection of k onto the
normal of the ellipse, Px, is negative everywhere. The resulting term, xTpk is the
normal velocity component on the ellipse. The main thrust of this paper is to be able

to control xTP* and to make this normal velocity component negative everywhere
for all time in order to achieve stability. As before, the linearity of the system allows

one to only check xTpk on the unit circle. This generalizes the previous stability
condition to

Stability Condition 2: For the system k = A(t)x, ifxTpk = xTPA(t)x <
0 V x : XTX = 1, V t > 0 for some constant positive definite matrix P,
then the origin is a global asymptotically stable equilibrium point.

The above condition applies to linear systems of any order.

Lyapunov [14] stated that if any positive definite function of the system states
V(x) is always decreasing, i.e. V(x) < 0, then the system is stable asymptoticallyy
stable for linear systems) [10]. For the linear case, let

V(x) = xTPx (2.10)

V(x) is positive definite if and only if P is a positive definite matrix. Taking the

derivative gives

V(x) = xTPx + xTp (2.11)

= xTATPx + xTPAx (2.12)

= xT[ATp + PA]x (2.13)

Lyapunov's stability condition becomes


xT[ATP+PA]x<0 Vx,Vt > 0


(2.14)










Since

XTpx = XTPAx = xT[ATP + PA]x (2.15)
2
Stability Condition 2 is equivalent to Lyapunov's linear stability condition.

It would be useful to compute the lower and upper bound of the rate of decay of
a given positive definite function,


V(x) = xTPx (2.16)

of the states of a time varying linear system by evaluating,

c := min [AminI PA(t) + AT(t)P] (2.17)
-- 1

c := max [Ama [PA(t) + AT(t)P]] (2.18)

So,

c < xT[PA(t) + AT(t)P]x < V x : xTx = 1, V t (2.19)
_2
Then c and Z are the lower and upper bounds respectively of the normal velocity
components, xTP*, on the unit spheroid. If c < 0, then the system is stable.

2.2 Adding Control to the System

The result in section 2.1 is an analytical tool only. A design procedure is needed
to control stability for the system

x= A(t)x + B(t)u (2.20)

where u is the control vector and x is the state vector. This section focuses on two

questions:

1) What effect does u have on the velocity field of (2.20)?
2) Can the normal velocity component xTP on the spheroid xTx = 1
be controlled?
























Figure 2.6. A velocity vector (a) without control and (b) with control.

To answer question 1), consider a second order single input system frozen at some

instant in time.

S= Ax + Bu (2.21)

A is a 2 x 2 matrix, and B is a two dimensional column vector, u is a scalar input

and can take on any real value. If u = 0 then the velocity field of the system is

k = Ax. Figure 2.6a shows the velocity, k = Ax, of some state vector, x, in the

system. The control vector, B, is also shown. When u : 0, 5 has the extra term,

Bu. The direction of Bu is constant, but its magnitude is directly proportional to u.

Figure 2.6b shows many possible values for by sweeping u through a wide range of

values in small increments. As this diagram shows, the arrow head of each velocity

vector can be placed on the line drawn parallel to B and intersecting the arrow head

of Ax. This demonstrates that velocities can be controlled along the space spanned

by the columns of B. The following theorem answers question 2).













B
0-5
0.5



-1
1.5 -
~2 -1.5 1 -0 .5 o o .5 1 '.5 2


Figure 2.7. A case in which the normal velocity component cannot be controlled.

Theorem 1
Consider the system, k = A(t)x+B(t)u, where dim(x) = n, dim(u) = m
and A(t) and B(t) have compatible dimensions. Let P be a constant
positive definite matrix. The normal velocity component, xTpk, can be
arbitrarily set to any value with the right choice of u at a given time t,
at any point x, on the spheroid, xTx = 1, unless x is contained in the set


S(t) := P-'span(B_(t)) n {x: xTx = 1}


(2.22)


where Bj(t) is a basis of column vectors orthogonal to span(B(t)). If
x E S(t) then


xTpk = xTPA(t)x


(2.23)


and this velocity component cannot be controlled at time t.


Theorem 1 gives the parts of the unit spheroid for which the velocity component
xTP* is uncontrollable and can therefore be used to determine if a linear time varying










system is stabilizable with an appropriate choice of constant positive definite P. If
there exists a constant positive definite P so that

max max xTPA(t)x < 0 (2.24)
t xES(t)
the system is stabilizable. The above condition is equivalent to requiring the following
matrix to be negative definite for all time.

IBT(t)[A(t)P-1 + P-1AT(t)]BL(t) (2.25)
21
This fact was established by Fields [15] and is true for the following reasons. Condi-
tion (2.24) is true if and only if

(P-l B (t),L)T(PA(t))(P-lB (t),t) <0VI' 0, Vt (2.26)
(P-1B (t)'U)T(P- (B (t)j')

Since the denominator of the above fraction is positive, with some simplification the
above statement is equivalent to

21T BT(t)[A(t)P- -1 <0 # 0, + t (2.27)
2 PA P- _L P^At)] y < 0 V IL k 0, V t (2.7

which is equivalent to

BT(t)[A(t)P-1 + P-lAT(t)]BI(t) <0 V t (2.28)

If a constant positive definite P exists satisfying condition (2.28), then the given time
varying system can be stabilized. This result is more general than a similar one given
by Shahruz [12], and the two are equivalent when P = I.
Before giving a proof of Theorem 1, we give an intuitive explanation of Theorem
1 for the second order case with a single input. Figure 2.7 shows the B vector for
such a system along with some ellipsoid xTpx = 1. If a state vector, x, is drawn
to some point on this ellipse whose normal is not perpendicular to B. then there is










always a control, u, which can place x inside the ellipsoid. This is true because the
velocity field can always be controlled in the B direction. This is not true, however,
for points on the ellipsoid whose normal is parallel to span(Bi.) as shown in figure
2.7. These points on the ellipsoid can be found by computing


P-'span(BI) (2.29)

where BI is a basis of column vectors orthogonal to span(B). We now give the proof
of Theorem 1.

Proof:
Case 1: x E S(t)

Substituting 2.20 into xTp*, we have

xTpk = xTPA(t)x + xTPB(t)u (2.30)

Since x E S then

x = P-'1(B.(t)) (2.31)

where l(B-(t)) is any linear combination of Bj1(t). So

xTPB(t)u = l(B_(t))TP-lPB(t)u (2.32)

= l(B(t))TB(t)u (2.33)

= 0 (2.34)

and therefore from 2.30 we have,


xTPk = xTPA(t)x


(2.35)










Case 2: x 0 S(t)
Suppose we want to set xTpik = c where c is some arbitrarily chosen
value. Then we want

c = xTPA(t)x + xTPB(t)u (2.36)

Since x S(t) then xTPB(t)u : 0 and there exists at least one u such
that 2.36 is true.

One could solve (2.36) for u to use as a control law, but this would not be practical
to implement because A and B can not easily be computed. Linear state feedback is
more desirable. The next section will develop a procedure for computing the set of
feedback gains that will implement a feedback control law keeping the normal velocity
component, xTPk, within some specified limit.

2.3 A Linear Feedback Set to Control xTP*

Recall from Section 2.1 that the stability of a time varying linear system

x = A(t)x (2.37)

could be analyzed by evaluating the bounds of xTPA(t)x on the spheroid

xTx = 1 (2.38)

Now we want to find the set of all linear state feedback gains for the control law,

u = -K(t)x (2.39)

for the system

x= A(t)x + B(t)u (2.40)

so that condition (2.19) holds for the closed loop system where c and are now
specified. This section will give conditions for the nonemptiness of such a set followed










by a discussion of its convexity. Finally this section will present an exhaustive search

method for finding the boundary of this set.

For the remainder of this chapter, A and B will be frozen at one instant in
time. A discussion follows on how to find a controller or set of controllers which

satisfies (2.19) at one given instant. If a given system is time varying then the results
which follow must be applied for all time with P being constant and positive definite.

These results apply to all systems which can be stabilized with respect to a Lyapunov
function using a constant P. The success of these methods depends on the existence
of a constant positive definite P so that

2BT(tt)[A(t)P-1 + P-AT(t)]B.(t) < 0 V t (2.41)

Since we require P to be zero, these results are conservative.

Let KA and TC be the set of all K which satisfy the left and right inequalities

respectively of the following expression.

c < xT[PA + ATP PBK KTBTPx V x : xTx = 1 (2.42)

where P is positive definite. The objective is to find


/C := k nC (2.43)

The following theorem gives conditions for the nonemptiness of _C and K?

Theorem 2 Given the 'nth order m-input system,

S= Ax+ Bu (2.44)

k and KT are nonempty if and only if there exists a positive definite P so
that the following conditions are true:

c < minxTPAx = min -xT[PA + ATP]x (2.45)
X ES xES 2










c> maxxT PAx = max 2xT[PA + ATp]x (2.46)
xES xeS 2

Comments:
Conditions (2.45) and (2.46) are respectively equivalent to

c < A,.m[I(vH/)-'B [AP-1 + P-1AT]BH(v')1- (2.47)

and

>_ Anax[(V'-H)-'BT[AP-1 + P-1AT]B(-vH)1] (2.48)

where

H = (P-1B)Tp-1B (2.49)

The matrix H is square, full rank, and has dimension n m. The above is true for
the following reasons. Equation (2.46) can be rewritten as


S> max 1xT[PA + ATp]x (2.50)
x=P-lBBL,,xTx=l 2
1 T T -1p1T1Rt.
= max -yB[APl +P-lATBI (2.51)
UTHL=I 2

Since H is positive definite then ~ exists, is square, and has an inverse. By making
the substitution, y = (v/H)-lz, (2.51) becomes

c > max zT(v/HB)-IBT[AP-1 + P-1A]B(v/H")-1z (2.52)
-zTz= 21
which is equivalent to

Am [vH)-1IT[AP-1 + P-1'A]B(vH)-] (2.53)


Equation (2.45) is equivalent to (2.47) for similar reasons. We now give a proof of
Theorem 2.










Proof:

We will show the nonemptiness of KX. KI can be shown to be nonempty in

a similar manner.

Case 1:

maxxxTPAx < (2.54)
xES
then

xTPAx < V x E S (2.55)



Let K = kBTP where k is a scalar value. Then


x = lxT[PA + ATP PBK KTBTP]x (2.56)

I XT[PA + ATP]x- k[xTPB][BTPx] (2.57)
2i 2

Since xTPB [BTPx]T then


xTPBBTPx > 0 V x S, V xTx = 1 (2.58)

If (2.55) is true and if k is made large enough, then the second term in

(2.57) will dominate V x S and x'p5c < V x : xTx = 1. Since

kBTP E E then E is nonempty.

Case 2:

maxxTPAx > (2.59)
xr=S
Then

XTPAx > for some x E S. (2.60)


By Theorem 1, xPAx can not be controlled V x E S; so TC is empty.










Remarks:
The nonemptiness of the intersection K: = 4c n is not guaranteed. If the designer
discovers that no intersection exists, then the upper and lower velocity bounds will
have to be adjusted.

Another useful property of CK: is convexity. This property is valuable in formulating
iterative search techniques to be described in the next section. K is convex if when

x and y are elements of K, then cax + (1 a)y is also an element of )C for 0
[16].

Theorem 3 Let K: be the set of all K so that

Amax [I[P(A BK) + (A BK)TP]] < (2.61)

Let IC be the set of all K so that

Amin[I[P(A BK) + (A BK)Tp]] > c (2.62)
2
Then /C, CK:, and : = K: fn kC are convex.

Proof:

If K: and K)C are convex, then the intersection K: is convex. Convexity of
kC will be proven here. In a similar manner the proof for the convexity of
KC can be written.

Let K1 and K2 be elements of T. We must show that aK1 + (1 ca)K2
is contained in KT when 0 < a < 1. We know that

IxT[PA+ATP-PBK1 -K KTBTP]x < V xTx = 1 (2.63)

and

lxT[PA + ATP- PBK2 KTBTP]x < V x = 1 (2.64)
2~ ~










Then

ax T[PA + ATP PBK1 KBTP]x < acE V x = 1 (2.65)

and

I(1-aQ)xT[PA+ATP-PBK2-KTBTp]x < (1-_a) VxTx 1 (2.66)
2

where 0 < a < 1. Adding (2.65) to (2.66) gives
xT[PA + ATP]x- _QXT[PBKl + KTBTP]x
(2.67)
-( a)xT[PBK2 + Kf2BTp]x < Z V xTx = 1
which becomes

1xT[PA+ATp-PB(aK+(1-a)K2)-(oaK+(1-a)K2)TBTP] < Vxx =1
2
(2.68)
So caK1 + (1 a)K2 E C for 0 < a c < 1. /C is convex.

Now that we have conditions on the nonemptiness and convexity of KC and k_ it
would be desirable to find the boundary of these sets. We know that if one of the
eigenvalues of the square matrix Q is c then

det[Q cI] = 0. (2.69)

This is helpful in understanding the following exhaustive search method for comput-
ing the boundary of T. A similar procedure exists for finding Q9_.

1) All eigenvalues of

I-[PA + ATP- PBK KTBTP] -I (2.70)
2
need to be less than or equal to zero. Fix all but one of the entries of K.
Let the i,j'th entry be the free entry and be represented by k. Let Koi,j










contain all the fixed entries of K and

Skl11


be defined as follows.
klj kin 1


kil 0 kin


kml k77j km,


Then


K = kQij + Koi,


1 j
0 0

0
0 1
0


0 0 0


(2.73)


2) Find all values of the free entry which make


det[PA + ATP 2cI PBK KTBTp] = 0


(2.74)


This problem reduces to finding the roots of a polynomial. After sub-

stituting equation (2.72) into the argument of the above determinant, it

becomes

k(-PBQi,j QJBTP) + (PA + ATP PBKoi,j KrjBTP) (2.75)

Let


Pl(i,j) := -(PBQi,j + QT.B TP)

Po(Koi,j) := PA + ATP PBKo,j KTojBTP


(2.76)

(2.77)


(2.71)


where


(2.72)


Koi,j :=


Qij "=










Then (2.74) becomes

det[kP1 + Po] = 0 (2.78)

Solve (2.78) for k.

3) Reject all complex roots. If all the roots are complex then skip the

next step.
4) Test the intervals between the real roots by checking to see if


Ama[PA + ATp 2-cI PBK KTBTp] < 0 (2.79)

The K's that bound the interval which satisfies (2.79) lie on the boundary

of T. Convexity of K implies that no more than two K's bound this

interval.

5) Repeat the process for all possible values for the fixed entries in K.

The result is 9K.

9KC can be computed in a similar way by reversing the inequalities in the above
procedure and by replacing max with Aiin. To find C the intersection of KC and TC

can be found in step 4).

The fixed entries in step 5) must be assigned to a finite number of grid points if

the above procedure is to be executed on a real system. The spacing of these grid

points must be smaller than the size of the feedback set. If the system has a high

order or a multiple number of inputs, the number of grid points will become too large,

and it will not be practical to implement this method. Understanding this procedure,

however, leads to the formulation of an iterative method that can be used on high

order, multiple input systems and will be presented in the following section. First,

an example is given applying the exhaustive search method to a second order, single

input system.











Example 1

Given the system


x= Ax + Bu


(2.80)


where

A '1 12 ]
0 -1 1
and
B = cos(75) ]
S sin(75o)
let

P=I

We want to find all feedback gains which satisfy the following constraint.


c

(2.81)


Before specifying c and c, we must check the normal velocity components
for x E S as Theorem 1 requires. Let


x=B = [ -sin75]

xAxo = -12.4075

1x TAxi =-12.40


(2.82)



(2.83)


From Theorem 2 we must have


c < -12.40 <


(2.84)


From this we choose


c = -14

-- -9


(2.85)

(2.86)


Then









In this example n = 2 and m = 1. It was decided to set i = j = 1 so
that

Q,j=Qi,i= [li 0] (2.87)

Koj = K 01,1 =[ 0 /i2 ] (2.88)

and
K = kQi,i + Koi,i = k K2 ] (2.89)

From equations (2.76) and (2.77) come

Pi_ = -[BQ,1 + Q,1B T] (2.90)
Pl = -[BQ1,, + QIBT] (2.91)

PoC = A + AT 2cI BKo0,1 KTBT (2.92)
Po- = A + AT 2I BKoi,1 KT' BT (2.93)

The roots of the following polynomials are computed in terms of k while
incrementing K2 through a wide range of values.

det[kP + Po] = 0 (2.94)

det[kP +Po0] = 0 (2.95)

Rejecting complex roots and checking the regions separated by the real
roots give

k!(K2), kc(K2) (2.96)
ku(K2), kj(g2) (2.97)

The intersection of these regions are found.

k(K2) = max[k,k-E] (2.98)
k(K2) = mink[,K] (2.99)












14

13 -

12

11 .[7,11]

10

9

81
4 5 6 7 8 9 10 11 12 13 14
ki

Figure 2.8. A plot of the boundary of IC which guarantees satisfaction of the design
constraints of Example 1.

Finally,


C C={[k(K2),K2] V K2 }u{[k(K2),K2] V K2} (2.100)


A plot of 9aC is shown in Figure 2.8. Let K = [7 11]. We can check to see

if K E KC by evaluating


A1 = Amin,(A+AT-BK KTBT) (2.101)

= -12.91 (2.102)


A2 = Am(A + AT BK KTBT) (2.103)

= -10.52 (2.104)


As the following shows,


c < A1 (2.105)

A2 < Z (2.106)










2.4 One Linear Feedback Matrix to Control xTP*

The search procedure given in section 2.3 becomes impractical for high order
systems with multiple inputs because the number of grid points for the fixed entries
of K becomes very large. This section discusses an iterative Lyapunov design method
which saves on computations and finds one element K E K, if it exists, where KC =
KC n KC. By applying this procedure at every instant in time to a time varying linear
system, one can find a control law stabilizing the system if a constant positive definite

P exists which satisfying the following condition.

2BT(t) [A(t)P-1 + P-1AT(t)]B(t) <0 V t (2.107)

By Theorem 2, this condition guarantees the existence of a Z < 0 such that C(t)
is nonempty for every instant in time. This procedure applies to all systems which
can be stabilized with respect to a Lyapunov function given by a constant positive
definite P. The following is a discussion of the iterative Lyapunov method followed
by the algorithm itself. An example is then given applying this procedure to one

operating point of a fifth order linearized model of the EMRAAT missile.
Figure 2.9 illustrates the iterative procedure. Kf is the feedback set which satisfies
the designer's predetermined constraints for some specified P. The constraints are

Cf < xT[PA + ATP PBK KTBTp]x < Zcf V x xTx = 1 (2.108)

Let Ki be defined as the set of all K which satisfies the following constraints.

ci < xT[PA + ATp PBK TBTp]x < cZ V x : xTx = 1 (2.109)

Given Ki, we would like to find c, and ci so that if Ki 0 kf then Ki E MKi. We also
require that Kf C Ki. If Ki E Kf, then we want K, = KC. The following definitions
for c, and ci meet these requirements. Let

i := max xT[P(A BKi) + (A BK)TP]x (2.110)
Ilx||=1 2













































Figure 2.9. A geometric view of the iterative Lyapunov design method.










and

Ai:= min Ix[P(A BKi) + (A BKi)Tp]x (2.111)
||x||=i 2 'J'
Then let

S= max[cf, \i (2.112)

and
ci = min[cf,A ] (2.113)

K0 is the initial guess in the search for Kf E kfA. Co and co are computed using
(2.112) and (2.113) so that Ko is a member of 9ko and /Cf C PC0. Then a new
feedback matrix, K1 is found which lies inside of K0, but not on the boundary. New
constraining values are found in the same way as before so that the boundary of the
next feedback set contains K1. K2 is then found so that it lies inside of the present
feedback set, but not on its boundary. This process is continued until Ki = Kf e ACf.

The success of finding Kf depends on the following conditions.

1. Given that Ki E a9)C we must be able to find Ki1 such that K++i G Ki.

2. We must show that 4i+1 < c, when -, > Zf and ci+1 > ci when ci < cf.

3. We must show that )Cf C kCi+l C Ki.

4. 'Cf must be nonempty.

We now address these four points.


1. Given Ki E aki, we need to find a second feedback matrix Ki2 C dKci where
Ki2 7 Ki. Then, due to convexity, 1 Ki + 1Ki2 is a member of kC{. Figure 2.10Oa shows
a second order example of a procedure for finding Ki+A The algorithm will be given
shortly. The horizontal and vertical axis are assigned to k1l and k12 respectively. The
region PCi is enclosed by aK&L and 9Ci. Kj is known. 'i2 is found by searching along










the line that passes through K, and is parallel to the k1l axis. Point c is found by
computing Ki + -Ki2. Since K;i is convex, then c E kC. This step is repeated again
by searching along the k12 axis. Using similar arguments, point f is also in KCi. KJ+i

is set equal to point f. For higher order systems, additional boundary points are
found and interior points computed by searching along directions which are parallel
to the axis of the coordinate system.

Under normal conditions, this procedure works well. However numerical problems
do occur. These problems will now be discussed along with a cure. Figure 2.10b shows
a second order example of when the above method fails. The boundaries are shaped

in such a way that searching along line 11 and 12 yields no new boundary points. A
solution to this problem is to relax one of the constraining values by setting, for ex-
ample, zi equal to + 6. 9Ad will then move so that point b can be found as shown in
figure 2.10c. Later, the relaxed constraining value can take on its original assignment.


2. We need to show that c, > Zi+l when c, > 5f. In this case


Ti= Ai > Zf (2.114)

Since Ki+j e KCU and Ki+j OQCi, then the following condition holds with strict
inequality.

xT[P(A BKi+1) + (A BKi+I)TP]x < a Vx : xTX = 1 (2.115)


Taking (2.110) for i + 1 then

IxT[P(A BKi+1) + (A BK<+)TP x V : xT = 1 (2.116)


and there exists an x which satisfies the above condition for equality. Therefore
Zi > AX+i and Zi > maxc.f,\i+l] = -i+n. Using similar arguments, it can be shown







38




a b


OKi





S< "/ /
*// Ii0o-/ /
K. a'ttily K ^2 / / /

/ K,
a' l1
/ a1 / h

k12
c


--I-

/
f' /t9

^i+i
S b

d ki


Figure 2.10. Step 5 of the iterative Lyapunov design method for (a) a second order
example, (b) how it sometimes fails, and (c) how this problem is corrected.











that c9 < cj+1 when ci < cf.


3. We now show that Cf C kCi+i C Ki. Since c, = max[f, Aj], then ci > c. We have

already seen that ci > >Ai+. So ci > max[f, AIi+] = ci+i. Similarly c ci+. Cki+l is

the set of all K so that

c I+ < xT[P(A BK) + (A BK)TP]x < c6+1 V x xTx = 1 (2.117)

Since T, > Ei+l and ci <_ 4i+1, then for every element of Kji+ the following holds.

c < xT[P(A BK) + (A BK)TP]x < Z, V x: xTx = 1 (2.118)

Therefore, .Ci+l C K;. Since c-i+ > cf and ci+l < cf, then using a similar argument

kCf C kA+1.


4. In using this design procedure, P is chosen so that the maximum uncontrol-

lable normal velocity component is negative. Then from Theorem 2, cf can be made

negative in an attempt to achieve stability. -f must be greater than the maximum

uncontrollable normal velocity component, and cf must be less than the minimum

uncontrollable normal velocity component. From Theorem 2 this will guarantee the

nonemptiness of Cf and kf. The nonemptiness of the intersection of these two sets,

however, is unknown. If Cf is nonempty, then, as i becomes large, Ki G Kfj. If Cf

is empty then ci and -. will converge to values which do not match the desired con-

straints and Ki will yield a closed loop system that meets the constraints given by cj
and ci. The designer will either have to accept this result or try again with a different

P or different constraining values or both. Since stability is desired, one approach

would be to keep P and Cf and lower Cf until Kf becomes large enough to intersect kf.


The outline of the iterative Lyapunov design method is as follows.










1) Choose P, Cjf, and Zf. This selection must obey Theorem 2. It should

be noted that Theorem 2 guarantees the nonemptiness of Kf and )Cf, but
not their intersections. If the system is time varying, then, in order to
use this algorithm, P must be found so that

2B T(t)[A(t)P-1 + P-lAT(t)]B(t) < 0 V t (2.119)

Otherwise this algorithm cannot guarantee stability.

2) Compute co and Co so that K0 = 0 E &aC0 and )Cf C Co. K0 will be

the initial guess in the search for Kf G ACf.

3) Let i = 0

4) Let i = Z+ 1

5) Find Ki so that KA E KC1I and KJi 0 M9Ai-.

6) Compute ci and c, so that Ki E 9C)i and KCf C /C, or so that Ki E kCf.

7) Repeat steps 4) 5) and 6) until one of two events occur.

1. ci = c and i = f .

2. (ci ci-1) and (i c,_i) become very small.

Remarks:
If event 1 occurs, then kCf is nonempty and Ki E Cf. If event 2 occurs, then kf is

empty and Ki yields a closed loop system that meets the constraints corresponding

to ci and ,i.
We now explain how to perform steps 2), 5), and 6).


Step 2)










The lower and upper bound of xTP* for the open loop system is respec-
tively,

Ao = Amird (PA + ATP)] (2.120)

Ao = Am,[ 2(PA + ATp)] (2.121)

Then,

o= min[ff,Ao] (2.122)

o = max[[cf ,Ao] (2.123)


Step 5)


The algorithm is now given.


K = K,
For j = 1 to m
For k = 1 to n
Koj,k = K kJ,kQj,k
Pi = -(PBQj,k + Q7kBTP)

Po = P(A BKOj,k) + (A BKoj,k)TP 2cl
Solve det[kP1 + Po0] = 0 in terms of k for c = ci, and Zi.
Reject all values which do not meet the constraints from
steps 2) and 6). The result is two intervals whose
lower bound is k_ and k2 and whose upper bound is
k and k2.
Find the intersection of these intervals by evaluating










k max[k, k2]

k = mn[Ti, -k2.
Find the midpoint of the interval by computing

kj,k = (k +).
Let K = ki,kQj,k + Koj,k
Next k

Next j
Ki+1 = K




Step 6)



Let

AiA = min[(PA + ATP- PBKi KTBTP)] (2.124)

SAmaxI(PA + ATp PBKi KTBTp)] (2.125)
2

So,

-* = min[-Cf,A] (2.126)

"i =max[-Cf, ] (2.127)

The following example illustrates the feasibility in applying this method to the

EMRAAT missile.

Example 2
We would like to apply the iterative Lyapunov design method to the EM-

RAAT missile. The missile was flown in a simulation through a trajectory











using another autopilot design. The model was linearized, and the fol-

lowing system was taken at 4.00 seconds into the flight.


x: = Ax + Bu


39 -.1
13 -.4
36 -1
.2 -6.
06 8;



B=


612 .0086
100 .1079
557 -2.078 -
097 .0173 -
L3.29 -.0261 -

0.0 -.1296
-.0149 0.0
-1150 -31.06
-4.494 -121.7
1.234 -.2802


(2.128)


.9997
0.0
.1763
.6403
.1597


0
.10-
-12:
-4.7-
-108


0.0
-.9997
.0442
.1611
-.5701

1.0
14
22
47
.6


(2.129)






(2.130)


x = [a,0,p,q,r]T


(2.131)


Step 1)

Let P = I. The minimum and maximum uncontrollable normal velocity

components were found by computing the limiting values in equations

(2.47) and (2.48). Since P = I, these terms simplify and are evaluated as

follows:


minlxT[PA+ATp]x= A'[mr[BT(A + AT)B.]
XS .7263
= -.7263


(2.132)

(2.133)


max xT [PA + ATp]x
xEs 2


= A [axc)I(A + A)B]

= -.3017


where


-.99;
.16
-64.:
-252
-.584


and


and


(2.134)

(2.135)










Theorem 2 guarantees the nonemptyness of T and K if f > -.3017 and

cf < -.7263. With this in mind, we let Zf = -.3 and cf = -6000. Since
the intersection of T and K is not guaranteed, cf was chosen to be very

negative to increase the probability of getting an answer.

Step 2)



A = A/,,[(A+ AT)] (2.136)
2
= -781.4 (2.137)

(2.138)


A = Ama,,[(A + AT)] (2.139)
2
= 778.9 (2.140)

(2.141)

So


co = min(cf,A) (2.142)

= -6000 (2.143)


Co = max(Zf,A) (2.144)

= 778.9 (2.145)

Steps 3) 7)
A program was written for MATLAB to carry out the iterations in steps

3) 7). The program would terminate if Kf was found or when

6 := I[i _i-1 < 6 = .0001 (2.146)











kf was found before 6 < .0001. The program ran 16 iterations. The final

result is


Kf =


.00046
2.057
-.00029


2.125
-.0012
-.7259


-.7974
2.339
.0263


.0805
-6.577
-.1346


.7535
.2591
-.7985


(2.147)


Now to check the result.


1
Af = A.n[(A+AT BKf KjBT)]
2f
= -818.0


Af = Amax[(A + AT BKf KfTBT)]

= -.3001

This meets the desired constraint.


af < Af < Af < Cf


(2.148)

(2.149)


(2.150)

(2.151)











CHAPTER 3
A TIME VARYING SECOND ORDER EXAMPLE

The following problem gives a case when pole placement succeeds in giving eigen-
values with negative real parts, but fails to stabilize the system. The Lyapunov
design method is then employed, and the resulting closed loop system is shown to be
asymptotically stable for all time.
We would like to find a feedback control law that stabilizes the system

x= A(t)x + B(t)u (3.1)

where

A (t) -1 + 1.Scos(t)[cos(t) + cos(t + Ir/18)] 1 1.5sin(t)[cos(t) + cos(t + r/18)] 1
A -1 1.5cos(t)[sin(t) + sin(t + 7r/18)] -1 + 1.5sin(t)[sin(t) + sin(t + 7r/18)]

(3.2)
and
B(t)= [ cos(t + r/18) (33)
sin(t + 7r/18) (
The eigenvalues of A(t) are -.4889 and 1.4661 for all time. The following control law
is proposed.

u = [ 1.5cos(t) -1.5sin(t) ]x (3.4)

The resulting closed loop system can be found in example 5.3,109 by Vidyasagar
[10] and also in Khalil [13]. The eigenvalues for the resulting closed loop system are
A = -.25 j.6614. Since the eigenvalues have negative real parts, one would expect
the closed loop system to be stable. However, Vidyasagar shows that the transition
matrix is
(t,, 0) = e5tcos) esin(t) (3.5)
1 _e.51sin^) e t cos(t)I












Trajectory of a pole placement time varying system


2501-


200

150

V 100


-50 -


-n00 -100 0 100 200 300 400 500
xl


Figure 3.1. The trajectory of the above closed loop system based
The initial conditions are x0 = [1 O]T. The system is unstable.


on pole placement.


If the initial conditions of the system are xo = [1 0]T, the resulting trajectory is

unstable as figure 3.1 shows.

We now turn to the Lyapunov design method. We choose P to be the identity

matrix. Before giving the design constraints, we need to check the the value of the

uncontrollable normal velocity components for all time. At t = 0


B = [.9848 .1736]T


(3.6)


Since P is the identity matrix, we are interested in the normal velocity component

which is on the part of the unit circle whose tangent is parallel to B. So we let


x = B = [.1736 .9848]T


The uncontrollable normal velocity component at t = 0 is


xTp = =xT[PA(O) + AT(0)P]x = -.9548
xT 2x


(3.7)


(3.8)










The uncontrollable velocity for this example is constant for all time. In selecting the

constraining values, we must have


c < -.9548 < Z (3.9)

The following assignments are made


c=-2 (3.10)

= -.5 (3.11)

Since the system is second order and has only one input it is possible to plot the set
of all feedback gains which satisfy the following.

C < lxT[P(A(t)-B(t)K(t))+(A(t)- B(t)K(t))TP]x < V x:xTx= 1, Vt (3.12)

This time varying feedback set is shown in figure 3.2. Since the time varying nature
of the system is periodic, and from inspection of figure 3.2, the following control law
is chosen.

u = 3[cos(t) sin(t)]x (3.13)

The feedback matrix in (3.13) is shown to be inside the moving feedback set in figure

3.2. The eigenvalues of the derivative of the resulting Lyapunov function is

A[A(t) B(t)K(t) + AT(t) KT(t)BT(t)] = -1.1193, -0.8579 Vt (3.14)

Therefore, the closed loop system is stable. Figure 3.3 shows a trajectory of the
Lyapunov based closed loop system where the initial condition is xO = [1 0]T.



























0


4 -2 2 4 -

____________ I) I p


4-
62 ---I -__________


6 -2 a 2 4
k11


-2


6 -4 -2 2 2 6
611
6 ____ d) t -3pir ___

4

2

0-


-6
-6 -O2 2 4
-8B 5 0 S ------ 4 ------


-26 7 2 4
k11 t.7p4


Figure 3.2. The time varying feedback set which satisfies the design constraints.








































-0.05


-0.1


-0.15


-0.2


-0.25


Trajectory of the Lyapunov based closed loop time varying system


: r /


-0.2 0 0.2 0.4 0.6 0.8 1
xl

Figure 3.3. A trajectory of the closed loop system using the Lyapunov design method.
xo = [1 O]J. The system is stable.













CHAPTER 4
DERIVATION OF A MODEL OF THE EMRAAT MISSILE

The next section derives a nonlinear model of the EMRAAT missile. A linearized

model is then generated in the following section and will be used for the design of the

autopilot. Aerodynamic and inertial cross coupling are assumed negligible in order

to reduce the order of the model. A linearized pitch model and a linearized roll-yaw

model results. All assumptions are clearly stated.

4.1 The Nonlinear Model

Before deriving the nonlinear model some variables and terms are defined. Figure

4.1 shows the missile body coordinate frame of the EMRAAT missile [2]. The three

axes, x, y, and z, are fixed to the missile as shown. The velocity of the missile is

represented by V. Angle of attack, a, is defined as the angle between x and the

projection of V onto the x-z plane. Sideslip, j, is the angle between x and the

projection of V onto the x-y plane. The velocity components, u, v, and w, are the

projections of V onto the x, y, and z axis respectively. The angle rates, p, q, and r,

are named roll rate, pitch rate, and yaw rate respectively and are defined as the rate

of rotation around the x, y, and z axis. Their directions obey the right hand rule.

These definitions are similar to those applied to aircraft as given by Etkin [17].

The following derivation of the nonlinear model is based on a similar model found
in Smith [2] and in Koenig [18]. Assumptions are made here to separate the system

into two lower order models: the pitch model and the roll-yaw model. We begin the

derivation by starting with the force equations. The forces acting on the missile are

thrust, gravity, and aerodynamic forces. The autopilot design in this paper applies

to the second phase of the flight when the engine is no longer burning so thrust is

















p



















r^ "- .- - ...q.


z




Figure 4.1. The missile body coordinate frame of the EMRAAT missile [2].










zero. Also, the weight of the missile is small in comparison to the aerodynamic force

so gravity is neglected. The aerodynamic force in the x direction is much smaller

than the aerodynamic force in the y and z directions, and therefore, will be ignored

for the rest of the paper. Newton's second law of motion implies the following:


Fy = m[i + ru pw] (4.1)

Fz = m[b + pv qu] (4.2)

Fy and Fz are the aerodynamic forces in the y and z directions. The quantities

inside the brackets are the total accelerations in the y and z directions. We know

that

V = [u2 + v2 + w2]1 (4.3)

Since v and w are much smaller than u then


V u (4.4)

Angle of attack and sideslip are given by


a = arctan (-) (4.5)

=3 arctan (-) (4.6)

If we assume that a and are small, then

w
a w- (4.7)
u
U

and
V
3-- (4.8)
U
Equations (4.1) and (4.2) can be rewritten as

Fy = mu + r p (4.9)











Fz = mu [+ p q]

which simplifies to

Fy = mV [- +r pal

Fz =mV [ + po3- q]

We assume that the forward velocity changes slowly so that


(4.13)


Then


w
U


V


and


So (4.11) and (4.12) become
Fy
mV
Fz
mV
The aerodynamic forces are given by


=O+r-pa (4.16)

-& +p p -q (4.17)


Fy = QS [Cyfi + Cyp + Cyjr + Cy6,bp + Cyl\6r]

Fz = -QS [CNQ + CNQ& + CNq + CNd6]q I


(4.18)

(4.19)


Q is the dynamic pressure and is defined as


Q= Ppv2


(4.20)


where p is the air density. S is the surface area of the wing. The aerodynamic

coefficients come from wind tunnel tests. They depend on mach number, and some

depend on angle of attack. The values of these coefficients have been put in tabular


(4.10)


(4.11)

(4.12)


(4.14)


(4.15)








form and are given in the first appendix along with other data pertaining to the
EMRAAT missile.
Substituting (4.18) and (4.19) into (4.16) and (4.17) and solving for & and /3 gives
QS
p= a r + -t [Cy + Cyp + Cyr + Cy,6/ + Cy,lr] (4.21)

and
S+ QSCN. q = QS [CN. + CNqq + CN,,6q] (4.22)

= [ + QSC [q p (CN.aa + CNq + CN,) (4.23)

Equations (4.21) and (4.23) are two nonlinear state equations. In order to separate
pitch dynamics from roll-yaw dynamics, /3 is assumed close to zero in (4.23). Thus,

[I [1 i QSCN]1 \ QSC \ + (I-' CAq) q +--2S N6q]

(4.24)
Equation (4.21) can be written as
S= ( y -" "-p(-1 -Cr) r-(--Cy6p) p-( P+ CY6,) 4,
\mV p/ V \ mV ) \mV p \mV
(4.25)
We now turn to the moment equations of the missile. Since thrust is zero and since
gravity does not contribute any moment to the missile, then the moments around
the x, y, and z axis are given by 1, m, and n respectively, the moments due to
aerodynamic pressure. They are given by

I = QSd [Ci3 + Cip + Cr + C,6p + Clfr\] (4.26)

m = QSd [Cma + Cm,& + Cmqq + C,6q 6,q] (4.27)
and


S= QSd [Co, + Cn,p + C1r + Cn,,p6 + Cn6r6r


(4.28)










where d is the missile diameter. Again, the aerodynamic coefficients come from
wind tunnel testing and are tabulated in the first appendix. Euler's three moment
equations can be written as follows.


m =J c + H(p,q,r) (4.29)
n r
where
Ixx -IXY -Ixz
J= -IXY IYY -IYz (4.30)
-Ixz -IYz Izz
and
-(Iyy Izz)qr + Iyz(r2 q2) Ixypq + Ixyrp
H = -(Izz- Ixx)rp + Ixz(p2 r2) Ixyqr + Iyzpq (4.31)
-(Ixx Iyy)pq + Ixy(q2 p2) Iyzrp + Ixzqr
We will assume that the inertial cross products, Ixy, Ixz, and Iyz are small. Also
due to symmetry of the airframe, Iyy and Izz are assumed to be equal. Inertial data
for the EMRAAT missile is given in the second appendix. Solving (4.29) for p, 4,
and r gives
S= --I (4.32)
'xx
= -- [m + (Izz Ixx)rp] (4.33)
IYY
and
S= -- [n + (Ixx Iyy) pq] (4.34)
'zz
Substituting (4.26), (4.27), and (4.28) into (4.32), (4.33), and (4.34) gives

QSd [ + Cp + Cr + C1 6+ C,] (4.35)

iQSd Izzxx
q QSd [Cmoa + Ca& + Cmq + Cm6qSq] + Izz Ixxrp (4.36)
IYY i 1* YY
and
____ '~ xx Iyy
S= [C/3 + Cp + Cnr + C 6,6p + C, 5] + 1- pq (4.37)
Izz Izz









If gyroscope effects are considered small, then (4.24), (4.25), (4.35), (4.36), and (4.37)
can be written as two separate systems: the pitch dynamics, and the roll-yaw dy-
namics. The pitch dynamics are as follows.

&=[ + QS\ QNa l[(-2-. Qc e + (I ( -'CN,) q+ (+Q CN6) 6q]
L rV J \mV } \ mV *} \mV *
(4.38)
QSd [C.a + Cm& + Cqq + C,,,q (4.39)

The roll-yaw dynamic equations are

QS (c + 2 Cy ))p (I + Q- (,) 2Cr+(Q2S C, ('S ) c r)
\mV */ \ mV *) \ mV mV p) \mV
(4.40)
sd[C,3 + C1pp + C r + C6pSP + C \r] (4.41)
Ix-x

= Q-d- [C,3 + Cnp + Cn1r + C,6p6P + Cn,65,] (4.42)

The states of the pitch model are a and q with 6q as its input. For the roll-yaw
model, the states are /3, p, and r, and the inputs are 6p and 64.

4.2 The Linear Model
The previous section gave a model of the pitch dynamics and the roll-yaw dynam-
ics in the following form.
[ = fq(a, q, q) (4.43)

[/ ]T fpT(),p,r,6p,6r) (4.44)

We would like to have two linearized models for use with the proposed design method.
The result will be two systems in the following form.

x = Ax + Bu (4.45)









where x contains the states and u contains the inputs. A and B are matrices which
are functions of several time varying flight parameters and are computed as follows.

A = df (4.46)
dx (X,W)nominal

dB f (4.47)
B=du
d(X,W ).ominal
where w contains additional flight parameters. Note that x and u are now pertur-
bations from the point around which the linearization is taken.
We linearize the pitch model first. From inspection of (4.38) we see that

=1 ( QSCNQ 1 (-QS CN
aqll = [1 +-'---- mI --- a
M mV M mV
aq12 +( QSCN (1- QS CN,)

bq= (I + QSCN) (- CN q)
M n-V M \mV q
To linearize (4.39), we must first substitute (4.38) in for &. Then we differentiate as
in (4.46) and (4.47).
QSd C + & QSCN (-QS \
aq22 = 7y [cm + Cm ( + ~mV~) ()]
h[U [( QSCNj1 (--QSCN ]

bq21 = QSd Cm6q + Cm (11 + QSCN) (QS0)]
Ivy Irn V M
The resulting linearized pitch model is
F 1 a=1, aq12 a I + bgil I 86 (4.48)
aq2l aq22 q bqg21
The same procedure is applied to the roll-yaw model. Assuming that a is constant
in (4.40), and from inspection of equations (4.40) to (4.42) the following results.
QQSCy QSCy, QSCyr
aprIl MV apr12 = a + ,-V apr13 = -1 + m"V







59


bpr1i QSCYP bpr12 QSCy6,
mV 'mV '
QSdCi QSdCi, QSdCh,
apZ1 ~ xI apr22 ~ X ~,apr23 -
QSdCj QSdCj ,
bpr21 XX Q pr22 Ixx
QSdC,, QSdCn, QSdCnr
apr31 = Z apr32 IZZ -I pr3 ~ IZZ '

_Q~rdC'6^ QSdCn6T
bpr3l QSdC,6 bpr32 QSdC

The linearized model is given by

apri apr12 apr13 + bpr11 bpr2 1 (p
p = apr21 apr22 apr23 p + bp21 bpr22 (449
1 apr31 apr32 apr33 i "r bpr31 bpr32












CHAPTER 5
THE DEPENDENCE OF GAINS ON FLIGHT PARAMETERS

This chapter describes the procedure that was used to determine which flight
parameters would be scheduled against the gains. Figure 5.1 gives a block diagram
of the system used for this procedure. The feedback gains are computed using the

iterative Lyapunov design method described in Chapter 2. The gains depend on
the linear model which in turn depends on seven flight parameters. Six of these
parameters are held constant while the seventh one changes. The resulting gains are
checked to see if they depend on this changing parameter. The process is repeated
for each flight parameter. It is found that the gains depend on angle of attack, mach
number, and dynamic pressure. This information will be used in the gain scheduling

process. The next section will discuss the atmospheric tables used to generate p and
V from M and Q. The flight parameter generator will then be presented. The third
section describes the initializer. The details of the iterative Lyapunov design method
are given in the fourth section. Finally, the results of the comparison between the
gains and flight parameters will be given in the fifth section.

5.1 Generating p and V from M and Q

By inspection, the linear models from Chapter 4 clearly depend on a, 0, p, q, r.
p, V, and the aerodynamic coefficients. The coefficients, however, can be eliminated
from the list because they depend on mach number and angle of attack. It is desirable

to replace p and V with M and Q since the later two can be easily measured on the
missile. We know the following.

2
Q = ^P2(5.1)

M = (5.2)
Vs0
60



















































Dependent
Flight
Parameters


Figure 5.1. A block diagram of the system used to determine the dependence of
gains on flight parameters.









where Vsos is the speed of sound. Both p and Vos are functions of altitude.

p = fp(h)

V8s0 = f(h)


where h is altitude in feet above sea level. Here,
atmospheric tables and are implemented by linear
and substituting the result into (5.1) gives


Q =


We generate a third table in the following way.


fp and fs are functions based on
interpolation. Solving (5.2) for V


(5.5)


f3(h) := f(h)f(h) = pV


= 2


(5.6)


(5.7)


The function, f3, is a one-to-one function so that its inverse can be found by reading
the table backwards. With this in mind we can solve (5.7) for h.


h f (2Q)
h= f31M


(5.8)


From (5.2) and (5.4)


V = MV,,, = Mf,(h)


(5.9)


Substituting (5.8) into (5.9) to eliminate h gives


S= Mf, (f- (2Q))

f (2Q))
-fP ( M2^


(5.10)


Also from (5.3)


(5.11)


(5.3)

(5.4)










Equations (5.10) and (5.11) are used to eliminate p and V in the linear model. As
a result, the linearized model can now be generated from the following seven flight

parameters.

[M Q ca 3 p q r] (5.12)

5.2 The Flight Parameter Generator

A series of flight conditions are made and used to generate many linear models.
Feedback gains are generated for each condition. The first of the series is called
the nominal flight condition. The values of the parameters for the nominal flight

condition are

M = 2, Q = 1250 psf, c = 8,/3 = 0,

p = 00/s,q = 100/s,r = 00/s.

Next, one of the parameters is allowed to vary while the other six are held constant.

This process is repeated six times so that each parameter can be tested. Table 5.1

shows the starting point, and the minimum and maximum values of each changing

parameter. Parameters with nonzero starting points begin at the starting point,

increment to the maximum value, return to the starting point, and then decrement

to the minimum value. Due to symmetry, the remaining variables have starting points

at zero and increment to their maximum values. M and Q sweep through a narrow

range because of restrictions of the atmospheric tables.

5.3 Initializing the Iterative Lyapunov Design Method

The iterative Lyapunov design method requires an initial guess, Kqo and Kpro and
two positive definite matrices, Pq and Pp,,. The initializer supplies these values and

will be discussed in this section.







64


Table 5.1. The initial point and range of changing flight parameters.
Initial point Minimum value Maximum value
M 2 1 2.6
Q 1250 psf 700 psf 5000 psf
a 8 _-80 200
/3 00 00 100
p 00/s 00/s 500/S
q 10/s -101/s 200/s
r 00/s 0/s 200/s

The initial feedback gains are found by using a pole placement algorithm. At the
nominal flight condition, the linear models are given by

A [ -1.1345 0.9996 [Bq -0.14631 (5.13)
S -261.4732 0.6209 ,Bq -123.1091 '

Apr -25. [ 24 1 .1350.9996 -017.5143 5. (5.13)
-0.459 0.140 -.100 0.018 0.117
Ap, = -2255.5 -2.41 .066 ,Bp = -1173.5 -1335.6 (5.14)
73.0 -0.181 -.648 L 2.01 -114.4
The desired eigenvalues for the closed loop pitch dynamics have been chosen to be
-40jlO0. For the closed loop roll-yaw model the desired pole locations are -20j5,
-80. The resulting feedback gains are

S\ 11.1633 -.6233 1 = -1.4166 -.0758 .3758 (5.15)
L 2.9337 .0086 -.3300 -



For both closed loop systems, P must be found so that xTpx is a Lyapunov
function. The following problem is stated.

Given a stable linear system : = Ax, find a positive definite function,
V = xTpx so that V1 = xT(PA + ATP)x is a negative definite function.

Let A be put into Jordan canonical form.

A = SJS-1 (5.16)










where S is an invertible matrix. The diagonals of J are the real parts of the eigen-

values of A, and imaginary parts of the eigenvalues lie in skew symmetric locations

off of the diagonals. For example, if the eigenvalues of A are a jb, c, then

'a -b 0
J= b a 0 (5.17)
0 0 c
Let

J = Jdiag + Jskew (5.18)

where Jdiag is the symmetric part of J and Jakew is the skew symmetric part of J. In

the above example

a 0 0 0 -b 0
Jdiag 0 a 0 Jskew = b 0 0 (5.19)
0 0 c 0 0
Making the following transformation on the system k = Ax, let


x = Sz (5.20)

Then

Si = ASz (5.21)

and

z = S-'ASz = Jz. (5.22)

Let Pz = -Jdiag. We wish to check the velocities of the system z = Jz on the ellipsoid

zTPzz = -zTJd"gz = 1 (5.23)

The normal to the ellipsoid at z is -Jdiagz. The projection of z = Jz onto the normal

is JdiagZ. Here the normal velocity component has the same magnitude but opposite

direction to the normal of the ellipsoid. If the system has no complex eigenvalues

then the velocities are orthogonal to the ellipsoid everywhere. For this reason, the










choice of P- = -Jdiag is the best choice for a positive definite function for the system

S= Jz.


V(z) = -zT JdiagZ


(5.24)


Making the following transformation into the x coordinate system gives


z = S-ix


(5.25)


which implies


V(x) = -xT[S-1]TJdiagS-X


Our choice of P is

P = -S-1]TJdiagS-.

For the nominal flight condition, the Jordan canonical form of Ag -
Apr- BprKpr is found and from (5.27)


S261772 60 1 ] 6792 10.7
Pq 609 14.9 Pp= 10.7 4.02
P 60 14[9 -316.4 -.509

The eigenvalues of (PqAq + ATpq PqBqKq KTBTpq) at


-316.4
-.509 (5.28)
15.7 j

the nominal point are


-1.0715 x 106, -40.002


(5.29)


Likewise, for the closed loop roll-yaw system the eigenvalues are


-1.3614 x 105, -320.00,-20.003


(5.30)


5.4 The Iterative LvaDunov Design Method


The iterative Lyapunov design method generates feedback gains so that xT PqX
and xTPp'x are Lyapunov functions for each closed loop system. The algorithm
requires the initial guesses Kgo and KprO, for the first point, and the positive definite
matrices Pq and Ppr. As a given flight condition changes, the feedback gains from


(5.26)


(5.27)

BqKq and










the previous point become the initial guess for the present point. The result of

this algorithm is a series of gains; one set for each flight condition generated. The

next portion of this section discusses some modifications made to the algorithm from

section 2.4. The material which follows describes how the design constraints are

selected.

The design method of section 2.4 finds a K so that the eigenvalues of (PA+ATp)

fall between cf and Cf where A is the closed loop system. The design algorithm

used in Figure 5.1 has been modified so that the resulting K satisfies the following

requirements.

cl < Ai ([P(A-BK) + (A-BK)TP]] < Z,
c2 < A2 [P(A- BK) + (A -BK)TP]] < 2 (5.31)
... ... ... ... ,...

c, < A,[I[P(A BK)+ (A BK)TP]] < n
where A1 is the smallest eigenvalue, A2 is the second smallest eigenvalue, and so on.

The values of the c's are supplied by the designer. This modification has been made

with the belief if more design constraints are made, then the performance will vary

less with changes in the flight conditions. The modified algorithm is as follows.

1) Choose P, e1f, cif, ... c,, and Z,,f. The selection of clf and cnf must

obey Theorem 2. However, Theorem 2 does not guarantee the nonempti-

ness of kC.

2) Compute cl0Cio, ... ,cn,,no so that the initial guess, K0 E Co0 and

JCf C /Co.

3) Let i = 0

4) Let i = i + 1

5) Compute Ki so that Ki E AC-1 and K, 9iC,-I.

6) Compute cli. ci, ... cni, Cni so that Ki E 9C, and KACf C Ci.







68

7) Repeat steps 4) 5) and 6) until one of two events occur.

1. cli= =Cl, Zli = Cll, ... cini = Cnf and Zni = Enf.
2. (Cli cli-1) *.. (ni- -,i-1) become very small.


Event 1. implies that the final answer has been found. Event 2. implies that /f
is empty and Ki satisfies the constraints corresponding to cli, ,i, ... cn, Z,. Steps

2), 5), and 6) are accomplished in the following way.


Steps 2) and 6)



Let
e = A1 [!(PA + ATp PBKi KTBTP)]

e = An [(PA + ATP- PBKi KTBTP) (5.32)
Then,
Cli = min[clf, el]
Eli = max[-cl, el]
... ...... (5.33)
fc = mzn[f _, e-]
-,i = max[nf, en]

Step 5)



Let K = Ki
For j = 1 to m
For k = 1 to n

P1 = -(PBQj,k + QBTP)
PO = P(A BKO,k) + (A BKo,k)TP 2cI

Koj,k = K kj,kQj,k










Solve det[kPi + P0] = 0 for in terms of k for c = cli, ... Ci.

Reject all values of k which do not meet the constraints from
steps 2) or 6). The result is 2n intervals whose lower bounds are

designated by k1, ... k2,n and whose upper bounds is named k, ..., k2n.

Find the intersection of these intervals by evaluating

k = max[k1, ...,1k2,]

k = mn [kl,...,k2n]
Find the midpoint of the interval by computing

kj,k = (k+k).
Let K = kj,kQj,k + Koj,k

Next k

Next j

KA+i = K



Remarks:

Let the feedback sets ),C1, k 2i &2, 2... ,, and C n be defined respectively as the

set of all K so the constraints (5.31) are met. Theorem 2 provides conditions for the

nonemptiness of K, and kAn. But conditions for the nonemptiness of the remaining

sets are still unknown. Also, )CI, ... ,_ are not convex in general. These are the

limitations of using the modified design algorithm.

We now turn to selecting constraining values for the eigenvalues of [P(A BK)+

(A BK)TP]. It is necessary to evaluate the uncontrollable normal velocity compo-

nents for each flight condition that will be generated in Figure 5.1. Tables 5.2 and 5.3

show for both models the minimum and maximum uncontrollable normal velocities

for each changing flight parameter.











Table 5.2. Uncontrollable normal velocity components for the pitch model
changing parameter minimum uncontrollable velocity maximum uncontrollable velocity
M -43.1456 -41.7253
Q -45.0144 -42.2289....
........ -43.3856 -42.2362....
__________ -42.6047 -42.6047
p -42.6047 -42.6045
q ..... -42.6047 -42.6047
r -42.6047 -42.6044


Table 5.3. Uncontrollable normal velocity components for the roll-yaw model
changing parameter minimumuncontrollable velocity maximum uncontrollable velocity
M -21.3937 -20.9751
Q -22.3016 -21.1574
_a_______ -21.7136 -21.1140
________ -21.3180 -21.3180
p -21.3181 -21.3180
q -21.3184 -21.3180
r -21.3180 "-21.3180..


For the pitch model, the uncontrollable normal velocity components range from

-45.0144 to -41.7253. Theorem 2 requires that


-41.7253 < q2
!ql < -45.0144


(5.34)


In addition, from (5.29), we want


cq1 <
_q2 <


-1.0715 x 106
-40.002


< Cql
< Cq2


(5.35)


From this the constraining values for the pitch model have been chosen to be


!21 = -1.2 X 106, qC- = -1 x 106,
Sq2 = -45, zq2 = -35


(5.36)


Likewise for the roll-yaw model, when looking at Table 5.3, Theorem 2 requires that


-20.9751
gprl


Cpr3
-22.3016


(5.37)










The eigenvalues in (5.30) suggest the following.

Cp l < -1.3614 x 105 < cprl
cpr2 < -320.00 < Cpr2 (5.38)
Cp,3 < -20.003 < Cp,3
The following constraining values have been chosen for the roll-yaw model.

cpi = -150000, Cprl = -110000
Cp,2 = -340, Zp, = -300 (5.39)
Cpr3 = -22, Cpr3 = -17
5.5 Formulation of a State Tracker

The autopilot of the EMRAAT missile will be a state tracker. That is, we want to
be able to change the location of the equilibrium point in order to control the values

of some of the states. The following shows how this will be accomplished.

Given the linear system

x= Ax + Bu (5.40)

y =Cx, (5.41)

we would like to find a control law


u = -Kx + KrefV (5.42)

so that y, the output, tracks v ,the reference input, asymptotically. We require y = v

when 5 = 0. When 5 = 0, then


0 = Ax BKx + BKfv. (5.43)

Since K is chosen so that the system is stable, then A BK is invertible and


x = -(A BK)-1BKrefV (5.44)

Also,


v = y = Cx = -C(A BK)-BKref V


(5.45)









Because (5.45) is true for any v, then

I = -C(A BK)-BKref, (5.46)

Controllability of the system implies that C(A-BK)-'B is invertible. Solving (5.46)
for Kref gives

Kref = -[C(A BK)-'B]-1 (5.47)

The EMRAAT missile has three inputs and therefore only three states can be
tracked. Controlling a in the pitch model and /3 and p in the roll yaw model is
desirable. For the pitch model y = a, implying that

Cq =[1 0] (5.48)

and, therefore,

Kef, = -[Cq(Aq BqKq)-'Bq]- (5.49)

For the roll-yaw model

Pi 0 o1 J (5.50)
Y = p 0 1 0
and, thus,

Krefr = -[Cpr(Apr BpKp1Bpr]'1 (5.51)

Krfq and Krefpr are computed for each flight condition and then compared along
with Kq and Kpr to the flight conditions.

5.6 Comparing Gains with Flight Parameters

A series of gains have been generated as a function of different flight conditions.
Each flight parameter has been swept through a range of points while the remaining
six have been held constant. In order to compare different gains on the same input
it has been decided to use the products of gains and their corresponding terms at
typical values. For example a typical value of a is 8. So we set ao equal to 8











Table 5.4. Extreme values and range of the pitch channel control terms as individual
flight parameters vary. Gains depend mostly on M, Q, and a._ __
_____M_. M] Q ... p q r
control min min min min min min min
term max max max max max max max
diff diff diff diff diff diff diff
kqlIao -2.15 -3.1 -2.1 -1.60 -1.61 -1.61 -1.60
-.72 -.16 -1.4 -1.59 -1.59 -1.59 -1.59
1.43 2.95 .69 .017 .017 .017 .017
kqi2qo -.1509 -.21 -.16 -.118 -.118 -.118 -.118
-.0529 -.031 -.10 -.117 -.117 -.117 -.117
.099 .179 .054 .0005 .0005 .0005 .0005
kref qllcO -2.53 -3.5 -2.7 -1.98 -1.98 -1.98 -1.98
-.916 -.53 -1.7 -1.97 -1.96 -1.96 -1.96
S 1.61 3.0 .94 .017 .017 .017 .017

and look at kqllao. We set qo and aco equal to 10/s and 8 respectively so that

we can look at kql2qo and kref qllao. The sum of these three terms are fed into to

the elevator. For the terms which are fed into the remaining inputs, the following

assignments are made.
/o = = 2
po = p = 100/s (5.52)
ro = 25/s
Figures 5.2a-g show kqi ao plotted against all seven flight parameters. These seven
figures show that kqnao changes with M, Q, and a but remains nearly constant when

/3, p, q, and r change. Similar figures exist for the remaining eleven gains and are

summarized in Tables 5.4 and 5.5. The minimum and maximum values of each term

is listed for each changing flight parameter along with the difference between the

minimum and maximum values. Angles are expressed in radians. From this table it
was determined that all gains will be scheduled against M, Q, and a.






















































































-1.59



-I,








.1.61 o v ) ) --------
0 20 40 6o a6 106 120 140 160 10 206



Figure 5.2. kqliao vs. a)M ; b)Q; c)a; d)3; e) p; f) q; g) r














Table 5.5. Extreme values and range of the roll-yaw channel control terms as indi-
vidual flight parameters vary. Gains depend mostly on M, Q, and a.
SM Q 1 p q -r
control min min min min minm imm mmin
term max max max max max max max
diff diff diff diff diff diff diff
kpli/3o -.050 -.11 -.22 -.050 -.050 .050 .050
.022 .045 -.013 -.038 -.036 -.036 -.036
.072 .157 .21 .011 .014 .014 .013
kpTrl2po -.17 -.14 -.29 -.13 -.13 -.13 -.13
-.011 -.061 -.13 -.13 -.13 -.13 -.13
.16 .20 .15 .001 .001 .001 .001
kpr13ro .074 .047 .15 .16 .16 .16 .16
.18 .28 .25 .16 .17 .17 .17
.10 .23 .099 .0006 .001 .001 .0009
kpr2ito .026 .0062 .065 .093 .092 .092 .092
.11 .167 .15 .10 .101 .101 .101
.080 .16 .084 .008 .009 .009 .009
kpr22po -.054 -.18 .0015 .014 .014 .014 .014
.015 .015 .152 .014 .015 .015 .015
.068 .20 .15 .0007 .0008 .0005 .0007
kpr23ro -.151 -.25 -.24 -.15 -.15 -.15 -.15
-.067 -.034 -.11 -.14 -.14 -.14 .14
.084 .22 .12 .001 .002 .001 .002
kref pr11ii/3co -.14 -.21 -.16 -.14 -.14 -.14 -.15
-.064 -.050 -.13 -.13 -.13 -.13 -.13
.08 .16 .028 .011 .014 .013 .013
kref pr12PcO -.071 -.046 -.39 -.040 -.040 -.045 -.040
.029 .15 .056 -.039 -.040 .013 -.038
.10 .19 .45 .001 .001 .057 .002
kref pr21l3co .055 .030 .11 .12 .12 .12 .12
.126 .19 .17 .13 .13 .13 .13
.071 .16 .058 .008 .009 .010 .009
kref pr22PcO -.010 -.33 -.17 -.071 -.071 -.12 -.071
-.070 -.070 .24 -.070 -.070 -.066 -.070
.027 .26 .41 .001 .001 .051 .002












CHAPTER 6
COMPUTING LOOK-UP TABLES

In the last chapter we showed that the gains depend mostly on a, M, and Q.
This chapter describes the process of generating a look-up table of gains verses M,

Q, and a. First a grid of points is formulated. Design constraints are then formulated

based on the knowledge of uncontrollable velocity components of the linear models.

Finally, the gains are computed.

6.1 Determining a Grid of Points

A two dimensional grid of points for M and Q has been made and used for each
entry of a in the table for both the pitch channel and the roll-yaw channel. Q sweeps
through a wide range of values starting with 100 psf and ending at 15,000 psf. The

values of M were chosen so that each entry of M and Q lie in the atmospheric tables

used to compute p and V. As a result, the grid points are not rectangular. All

entries are restricted to values between M = .6 and M = 3.5 and must correspond

to altitudes between sea level and 50,000 ft.

Table 6.1 shows the values of a used in the look-up table for both channels. M
and Q sweep through all values of the grid previously mentioned for each value of a
in Table 6.1. The spacing between the grid points was determined in a trial and error

process. During the iterative procedure for computing feedback gains, the initial
guess for each point came from the result of an adjacent grid point. The closer the

spacing between adjacent points, the fewer iterations were needed to find the next

feedback gain. Numerical problems as described in section 2.4 and shown in figure
2.10 were encountered. When this happened some of the constraints were relaxed

so interior points in the feedback set could be found. Later these constraints were








77


Table 6.1. Values of a in the pitch controller look-up table.
a
Pitch Roll-Yaw
-3.0 -1.0
1.0 1.0
4.0 2.5
8.0 4.0
12.0 8.0
16.0 12.0
20.0 16.0
_____ 20.0

made more restrictive and returned to their original assignments. This became a

tedious process for some parts of the grid. When the number of iterations exceeded

100 it was decided to add more grid points so that the desired feedback gains could

be found in fewer iterations.

6.2 Formulation of the Design Constraints

Before using the iterative Lyapunov design algorithm, the uncontrollable normal

velocity components for the entire M-Q-a grid must be determined. This information

is needed to formulate the design constraints. This process is described by the block

diagram in Figure 6.1. Table 6.2 presents the minimum and maximum uncontrollable

normal velocity components of the pitch model throughout the M Q grid for each

value of a. Here P = Pq, the matrix computed during the initializing procedure

of the last chapter. Table 6.3 gives the same result for the roll-yaw model where

P = Ppr,. The extreme values of uncontrollable normal velocities for the pitch model

are -49.5985 and -37.3480. Also the eigenvalues of

I[Pq(Aq BqKq) + (Aq BqKq)TPq] (6.1)







P P
q'^ pr


Uncontrollable
Normal
Velocities


Formulation
of Design
Constraints


Figure 6.1. Formulation of the Design Constraints.


Table 6.2. Uncontrollable normal velocity components for the pitch model.
a(degrees) minimum uncontrollable velocity maximum uncontrollable velocity
-3 -47.1539 -37.4260
1 -47.0806 -37.3873
4 -47.1611 -37.3480
8 -47.9797 -37.4247
.. 12 -48.8796 -37.3906
16 -49.5915 -37.6144
20 -49.5985 -37.4355

Table 6.3. Uncontrollable normal velocity components for the roll-yaw model.
a(degrees) minimum uncontrollable velocity [maximum uncontrollable velocity
-1 -23.3399 -19.6307
1 -23.3774 -19.6409
2.5 -23.4445 -19.6883
4 -23.4971 -19.7450
8 -24.5852 -17.2696
12 -25.3163 -19.9600
16 -25.6480 -19.8625
20 -26.4797 -20.30717


Flight
Parameter
Generator










at the nominal flight condition from the previous chapter were found to be -1.0715 x

106 and -40.002. Theorem 2 requires that

-37.3480 < cq2 (6.2)
cqI < -49.5985
But we also want
cq1 < -1.0715 x 106 < ql (6.3)
-q2 < -40.002 < (6q.
since we desire the eigenvalues of (6.1) to be close to those at the nominal flight

condition. From this, the constraining values were chosen to be

CqI = -1.2 x 106, q1 = -106,

Cq2 =-45, cq2 =-35

Likewise, for the roll-yaw model, the uncontrollable normal velocity components range

from -26.4797 to -17.2696. From Theorem 2 we must have

-17.2696 < p3 (6.4)c
cpr1 < -26.4797 )
With the eigenvalues of

S[Pp,(Apr BprKpr) + (Ap. BprKp,)TPp.]
2
at the nominal flight condition being -1.3614 x 105, -320.00, and -20.003, the

following is desirable.

cp,1 < -1.3614 x 105 < CplI
cpr2 < -320.00 < Cp,2 (6.5)
cpr3 < -20.003 < Cp,3
With this in mind, the following selections were made.


.prl = -200000, CprIl =-90000,

pr2 = -400, Tp,2 = -250,


cpr3 = -25, Cp,3 = -15











6.3 Generating the Look-Up Table

With the design constraints set, feedback gains and feedforward gains are gen-

erated for each grid point. Figure 6.2 gives a block diagram of the system used to

accomplish this. For the pitch model, the initial guess comes from one of the gains

that was generated in the previous chapter when determining the dependence be-

tween gains and flight parameters. The operating point from which this initial guess

originates is

M = 1, Q = 1250psf, a = 8 (6.6)

and is one of the extreme values listed in Table 5.1. The result of the first point is

used to start adjacent points which, in turn, start new adjacent points until gains

have been computed for the entire grid. The look-up table for the roll-yaw model is

made in the same way.

Numerical problems were encountered in parts of the look-up table for the roll-

yaw model. They were similar to the problems that were predicted in step 5) of the

iterative design method given in Chapter 2. To overcome these difficulties, some of the

constraining values were relaxed for a number of iterations and were later returned to

their original assignments in the algorithm. Eventually, the desired feedback matrix

was found.

Figure 6.3 shows kq ii verses mach number and dynamic pressure when a = 8. As

this figure would indicate, the gains generated from the iterative Lyapunov method

are smooth with respect to the dependent flight parameters. This fact gives hope

that the gain scheduling scheme will be easy.

















































Gain
Schedules


Figure 6.2. Formulation of the look-up table.




















































10000


5000


0 0.5


Figure 6.3. kql, verses M and Q when a = 80













CHAPTER 7
GAIN SCHEDULING

The result of the previous chapter is thirteen tables of gains in terms of mach

number, dynamic pressure, and angle of attack. This chapter discusses the process

of curve fitting used to implement the look-up table. The results of a test of this

scheme will follow.

7.1 Curve Fitting

It was decided to use a combination of polynomial fitting and interpolation to

implement the autopilot. Third order polynomials were fit to the tables as a function

of mach number. However, low order polynomials could not achieve close fits as a

function of angle of attack or dynamic pressure, so linear interpolation was used for

these two variables. Figure 7.la shows kq11 as a function of M for various constant

values of Q at a = 8. An example of polynomial fitting of one of these curves is

shown in Figure 7.1b where Q = 150psf. A polynomial has been made for every

a-Q pair and the polynomial coefficients are interpolated as a function of these two

variables. The tables of the three pitch controller gains each have 7 entries for a

and 22 entries for Q. Since each polynomial has 4 coefficients the total number of

coefficients for each gain is 7 x 22 x 4 = 616. Similarly the tables for the roll-yaw

controller have 8 entries for a and the same number of entries for Q. Each of the ten

gains are then scheduled using 8 x 22 x 4 = 704 polynomial coefficients.

7.2 Testing the Fit

Figure 7.2 gives the system for testing the polynomial and interpolation routines.
Gains were generated from these routines at locations centered between the original

grid points. These new locations were found by taking the average of the coordinates

83












a 0
a
10
0_ .......

.20 .- -47
30,
i- .. 48.


-60
-70-50
.5 1 1.5 2 as 3 3.5 .6 0.65 0.7 0.75 0.8 0.A5 0.9 0.95 1
M M

Figure 7.1. A plot of the kqii verses M with (a) various constant values of Q and
Q = 8, and (b) a least squares third order polynomial fit for the plot where Q =
150psf and ac = 8.


of adjacent grid points. Linear models were also made at each test point where /, p,

q, and r were set to zero. The eigenvalues of


I[Pq(Aq- BqKq) + (Aq BqKq)TPq] (7.1)


and

j[Pp,(Ap,, Bp,Kp,) + (Ap. BpKp)TPP] (7.2)

were computed to see if they remained within the desired limits. Table 7.1 shows

the maximum eigenvalue of (7.2) for all of the test points at a = 1.75. These values

are plotted against their indices in Figure 7.3. Most of the eigenvalues of Table

7.1 lie within the desired limits of -25 and -15. The eigenvalue in seventeenth row,

second column, however, is -3.96, the worst value found out of all of the test points.

Although the deviation is high, this value is still negative indicating stability for the

closed loop system. For the pitch controller, the actual limiting values of A1 and A2

at all the test points are

-1.37 x 106 < Aq1 < -9.85 x 105 (73)
-47.7 < Aq2 < -38.1 ( )

















































Figure 7.2. A test of the curve fitting routines used to implement the autopilot.












Table 7.1. The maximum eigenvalues of (6.1) for all test points at a = 1.75.
Q/M indices 1 2 3 4 5 6 7[ 8
1 -20.79 -20.82 -20.84 -20.87 -20.88 -20.89 -20.90 -20.90
2 -20.68 -20.69 -20.70 -20.72 -20.74 -20.76 -20.76 -20.76
3 -20.84 -20.85 -20.84 -20.83 -20.81 -20.79 -20.78 -20.79
4 -20.59 -20.61 -20.49 -20.30 -20.18 -20.33 -20.34 -20.21
5 -20.30 -20.22 -19.93 -19.71 -19.96 -19.88 -20.01 -20.62
6 -20.41 -19.97 -19.77 -20.16 -19.88 -20.20 -20.79 -20.97
7 -19.10 -17.70 -17.86 -18.92 -20.03 -20.42 -19.75 -20.81
8 -18.62 -19.79 -20.67 -20.37 -16.16 -19.14 -20.21 -19.48
9 -19.90 -20.97 -19.33 -14.48 -19.58 -21.08 -20.93 -19.64
10 -20.23 -20.88 -11.67 -16.89 -19.96 -20.40 -19.72 -19.41
11 -20.57 -20.38 -12.40 -18.26 -20.60 -20.85 -20.28 -20.30
12 -20.77 -17.82 -17.66 -20.97 -21.85 -21.65 -21.14 -21.32
13 -19.83 -18.51 -21.91 -20.81 -20.87 -22.17 -22.13 -22.15
14 -20.45 -22.23 -20.57 -17.42 -20.84 -22.21 -22.42 -22.39
15 -22.42 -22.13 -18.83 -18.24 -20.65 -21.23 -21.49 -21.90
16 -21.82 -19.45 -15.17 -20.54 -21.09 -20.92 -20.94 -21.73
17 -21.85 -3.96 -20.63 -22.26 -22.35 -22.13 -22.03 -22.48
18 -18.88 -19.74 -21.48 -22.29 -22.58 -22.65 -22.68 -22.87
19 -20.85 -21.06 -21.71 -22.27 -22.69 -22.98 -23.19 -23.37
20 -21.65 -21.93 -22.29 -22.63 -22.93 -23.19 -23.39 -23.52
21 -20.62 -22.15 -22.52 -22.62 -22.61 -22.53 -22.46 -22.42


-10

M~ -20,
E
t-30,

-40-

-50,,
30 -


10 04
I ,,^ F 0 0 . .


Indices of M


Figure 7.3. A plot of the eigenvalues from Table 6.1.


lll in i c -








87

Likewise, for the roll-yaw controller, the limiting values at all the test points are

-2.03 x 105 < Aprl K -9.28 x 104
-2392.7 < Apr2 < -182.6 (7.4)
-26.27 < Apr3 < -3.96

Some of these values differ significantly from the desired constraints; however, since

these values are still negative, these deviations are acceptable and indicate that the

closed loop system will be stable. It should also be noted that most eigenvalues

remain well within their desired constraints as shown in Table 7.1.













CHAPTER 8
NONLINEAR SIMULATIONS

A nonlinear simulation has been used to test the proposed autopilot for the EM-

RAAT missile. First a section follows giving an overview of the nonlinear simulation.

A test module is then made to generate state commands in order to evaluate the

autopilot's tracking ability. Finally, a series of flight scenarios are run to determine

the ability the missile has to intercept the target.

8.1 The Nonlinear Simulation

Figure 8.1 shows a block diagram of the simulation used to test the EMRAAT

missile. The program is written in FORTRAN. Initial conditions of the target and

missile are specified by the user. The simulation is then run and a trajectory of both

the target and missile results. All target and missile variables can be observed. The

target is programmed to fly in a straight line until the range between the target and

missile falls below 5,000 ft. The target then makes a 9 g turn to the right. The

simulation terminates when the closing velocity becomes positive.

The seeker measures the line of sight angles and the range rate of the target.

The simulation uses exact measurements of these values and does not assume any

noise. These values are sent to the guidance law which, in this case, implements

proportional navigation. A derivation of this guidance law can be found in Bryson

and Ho [19]. The outputs of the guidance law are two desired accelerations, a, and

a,. The BTT logic makes the conversion from the acceleration commands to the

three state commands ac, 0, and pc. Since the missile can achieve a much higher

acceleration with angle of attack than with sideslip, the BTT logic uses Pc to rotate

the desired accelerations into the pitch plane. If this roll maneuver is successful then


















Seeker
(RF)


Gain
Schedule




M Q .


K

K~f


n Guidance Law
R (Pro-Nay)
OQ
vc


ayc, azc

BTT Logic


aI O cl


PC


oi p q r


Sp
6q
Sr


Nonlinear
Missile
Dynamics




a Iaz


Exact Computation of Missile and Target Variables


Figure 8.1. A Block Diagram of the Nonlinear Missile Simulation.


Target
Position


U = -Kx + K,,fv


i[ l, i










ay, will become small and a, will become positive. ac and /30, are computed in an

attempt to match a^ and ayo respectively.

The autopilot implements the control law


u = -Kx + Krefv (8.1)

where x is a vector containing the actual states and v contains the state commands

from the bank-to-turn (BTT) logic. The states come from exact measurements in the

simulation. If this autopilot were to be implemented in an actual missile, the states

would be measured using an inertial platform. The gains K and Krf come from the

gain schedule implemented with a combination of polynomials and interpolation. In

this simulation there is no delay in the gain schedule and K and h,,f are produced

instantaneously. The output of the autopilot is the control surface angles 6p, 6q, and

86. Linear and angular accelerations are computed by the missile dynamics module
of the program. The simulation uses the output of the missile dynamics to compute

all of the flight variables including the position and velocity of the missile.

8.2 A Test of State Tracking

The model for the EMRAAT missile has five states and three inputs. The autopi-

lot is designed to track three state commands: ac, 3, and pc. Before running missile

target scenarios it was decided to test the autopilot's tracking ability. The BTT logic

was disconnected, and the following commands were applied to the reference inputs

of the autopilot.
100 for 0 s < t < .5 s
0 for .5 s < t < 2.75 s (8.2)
c= 10 for 2.75 s < t <3.75s (.
0 for 3.75 s < t
0' for 0 s /= 5 for 2 s 0 for 2.5s < t











Table 8.1. The rise times of each commanded state.
Altitude(ft) Mach t,(s) t, (s) 4.(s)
20,000 2.0 .100 .243 .0284
50,000 3.0 .108 .316 .0181

0/s for 0 s < t < 1 s
100/s for 1 s < t < 1.5 s
0/s for 1.5 s < t< 2.75 s .
Pc 100o/s for 2.75 s < t< 3.25 s (8.4)
-100/s for 3.25 s 0/s for 3.75 s < t
The experiment was performed once at an initial altitude of 20,000 ft with an initial

mach number of 2.0 and a second time at an initial altitude of 50,000 ft with an initial

mach number of 3.0. Figure 8.2 shows the results of the first test. In addition to the

commanded states, figure 8.2 also shows the remaining two states q and r. The rise

time as defined as the time needed to achieve 90% of the desired value was found

for each commanded state and is shown in table 8.1. It should be noted that mach

number and dynamic pressure do not change significantly during this test. This is a

test of accuracy in state tracking and is not a valid test of gain scheduling in terms

of M and Q. However, a changes very rapidly and does not appear to hinder the

performance.

Cross coupling is evident from this experiment. The step in the roll rate of figure

8.2c at t = 2 s is due to the sideslip command in figure 8.2b. A lesser degree of cross

coupling occurs during the roll command for 2.75 s < t < 3.75 s when a which is at

10 is rotated into sideslip. The same effects are present in the second experiment at

50,000 ft.

Figure 8.2d and 8.2e show the two states that have no commands. The spikes in

the pitch rate occur when ac changes value, because a pitching maneuver is required

to change the angle of attack. The two spikes in the yaw rate occur for the same





































b) C--) Co,,m,..c 0ta C-)


0 = a a a r a

a) p C -) ~,. Oo,,a~ 0 (-)
'Sc.


'ool


S.o


-0)


Figure 8.2. A test of state tracking at an initial altitude of 20,000 ft with an initial

mach number of 2.0. The states shown are (a) a, (b) 03, (c) p, (d) q, and (e) r.


S.---


. T


. -a ora.










reason when sideslip changes. The two steps in the yaw rate, however, are due to

cross coupling with the roll rate.

The rise times in table 8.1 are small and indicate that the missile will perform well

in flight scenarios. While cross coupling effects are noticeable, they are not believed

to be great enough to significantly hinder the performance of the missile; thus it was

decided to run the missile through a series of flight scenarios to see how well the

missile can intercept the target.

8.3 Simulation of Flight Scenarios.

A series of flight scenarios has been run to test the autopilot. Figure 8.3 shows

the trajectory of the missile and target for a flight scenario at an altitude of 20,000

ft. The miss distance is .64 ft. A hit is considered to be any miss distance under 10

ft. Figure 8.4 shows the commanded and the actual y and z accelerations. The devi-

ations from the commanded normal accelerations are mostly due to a simplification

used in implementing the BTT logic. Instead of using two aerodynamic coefficients,

proportionality constants were assumed. The errors are not large enough to prevent

the interception of the target. The state tracking of the reference inputs appears to

be working well as seen in figure 8.5 and indicates that the autopilot is performing

well.

The trajectories of a 40,000 foot altitude scenario is shown in figure 8.6. The miss

distance is 0.05 ft. A similar scenario at 10,000 ft is shown in figure 8.7. The miss

distance is 2.8 ft.

Figure 8.8 gives a case were a miss occurred. The scenario took place at 50,000

ft and the target was missed by 452 ft. Figure 8.9a and b shows that the missile was

unable to achieve the desired z acceleration and angle of attack after 1.3 seconds into

the flight. This is because the elevator reached its -40 limit as figure 8.9c indicates.























a) Top View of Missile (-) and Target (-) Trajectory


6000

5000

4000

3000


0 1000 2000 3000 400 5000 6008 7000 800O 9
Position in the x direction in ft.
xl04 b) Side View of Missile (-) and Target (-) Trajectory


-2.05 k


0 1000 2000 3000 4000 5000 6000 7000 8000
Position in thex direction in ft.


Figure 8.3. (a) Top view and (b) side view of missile and target
scenario which occurred at 20,000 ft. The miss distance is 0.64
numbers of the target and missile are 2.5 and .92 respectively.


trajectories of a
ft. Initial mach






















a) Commanded z Acceleration (-) and Acwal z Acceleration (-)


Seconds
b) Conmmanded y Acceleration (-) and Actal y Acceleration (-)


Seconds


Figure 8.4. Commanded and actual (a) z accelerations and (b) y accelerations for
the scenario in figure 8.3.




Full Text
68
7) Repeat steps 4) 5) and 6) until one of two events occur.
1. Cjj Q-ifi cu C\}i i Qni Qnf ^nd Cni Cnf.
2. (clt c1_1) ... (c cm_ 1) become very small.
Event 1. implies that the final answer has been found. Event 2. implies that ICf
is empty and K{ satisfies the constraints corresponding to cu, cu, ... cn, cn. Steps
2), 5), and 6) are accomplished in the following way.
Steps 2) and 6)
Let
Then,
ei
Ai
\{PA + ATP PBKi KjBrP)
\{PA + AtP PBKi KjBTP)\
(5.32)
cu = min[clf, ex]
cu = max[ci/,ei]
Qni ?7iZ7i[cny, en]
Cjii 77]
(5.33)
Step 5)
Let K = Ki
For j = 1 to m
For k = 1 to n
Pi = ~(PBQi,k + QT,.kBTP)
Po = P(A BK0j,k) + (A BKaj,t)TP 2cl
K0j,k kj,kQj,k


using another autopilot design. The model was linearized, and the fol
lowing system was taken at 4.00 seconds into the flight.
43
X
= Ax + Bu
(2.128)
where
' -.9939
-.1612
.0086
.9997
0.0 '
.1613
-.4100
.1079
0.0
-.9997
A =
-64.36
-1557
-2.078
-.1763
.0442
(2.129)
-252.2
-6.097
.0173
-.6403
.1611
-.5806
83.29
-.0261
-.1597
-.5701
0.0
-.1296
0.0 '
-.0149
0.0
.1044
B =
-1150
-31.06
-1222
-4.494
-121.7
-4.747
1.234
-.2802
-108.6
and
x = [a,/3,p,?,r]r
(2.130)
(2.131)
Step 1)
Let P = I. The minimum and maximum uncontrollable normal velocity
components were found by computing the limiting values in equations
(2.47) and (2.48). Since P /, these terms simplify and are evaluated as
follows:
min -xT [PA + A1 Phi
X6S 2 J
and
max -xT[PA + AT P\x
xes 2
A rmn[\Bl(A + A^B,] (2.132)
-.7263 (2.133)
\maX[\Bl(A + AT)BL] (2.134)
(2.135)
-.3017


79
at the nominal flight condition from the previous chapter were found to be 1.0715 x
106 and 40.002. Theorem 2 requires that
-37.3480 < cq 2
cql < -49.5985
(6.2)
(6.3)
But we also want
cql < 1.0715 x 106 < Cqi
cq 2 < 40.002 < cq2
since we desire the eigenvalues of (6.1) to be close to those at the nominal flight
condition. From this, the constraining values were chosen to be
Cql = -1.2 X 106, Cql = -106,
cq2 = -45, Cq2 = -35
Likewise, for the roll-yaw model, the uncontrollable normal velocity components range
from 26.4797 to 17.2696. From Theorem 2 we must have
17.2696 < cpr3
cpn < -26.4797
(6.4)
With the eigenvalues of
PpT ( Apr
- B
pr Kpr ) h (-^.pr
Bpr
KpT)T PpT]
condition being 1.3614 x
105
, -320.00,
and 20.003, the
prl
<
-1.3614 x 105
<
Cpr 1
pr 2
<
-320.00
<
Cpr 2
(6.5)
Lpr3
<
-20.003
<
Cpr 3
With this in mind, the following selections were made.
cpri = 200000, cpri = 90000,
Cpr2 = -400, Cpr2 = -250,
pr3 CpT 3 = 15


2
Figure 1.1. The EMRAAT missile [1].


81
Gain
Schedules
Figure 6.2. Formulation of the look-up table.


28
Then
~axT[PA + AtP PBK1 K?BTP]x < ac V xTx = 1 (2.65)
and
^(l-a)xT[PA+ATP-PBK2-K^BrP]x < (1 -a)c VxTx = 1 (2.66)
where 0 < a < 1. Adding (2.65) to (2.66) gives
xT[PA + ATP]x \axT[PBKx + K?BtP]x
|(1 a)xT[PjBAr2 + K^BTP]yi which becomes
(2.67)
^T[PA+ATP-PB(aK1+{l-a)K2)-(aK1+(l-a)K2)TBTP} (2.68)
So aKi + (1 ol)K2 G 1C for 0 < a < 1. /C is convex.
Now that we have conditions on the nonemptiness and convexity of 1C and 1C it
would be desirable to find the boundary of these sets. We know that if one of the
eigenvalues of the square matrix Q is c then
det[Q cl] = 0. (2.69)
This is helpful in understanding the following exhaustive search method for comput
ing the boundary of 1C. A similar procedure exists for finding dlC.
1) All eigenvalues of
i[PA + AtP PBK KtBtP] cl (2.70)
need to be less than or equal to zero. Fix all but one of the entries of K.
Let the jth entry be the free entry and be represented by k. Let Kqj


37
the line that passes through K{ and is parallel to the kn axis. Point c is found by
computing \Ki + \K{2. Since 1C1 is convex, then c 6 AC,. This step is repeated again
by searching along the kx2 axis. Using similar arguments, point / is also in AC. Kt+\
is set equal to point /. For higher order systems, additional boundary points are
found and interior points computed by searching along directions which are parallel
to the axis of the coordinate system.
Under normal conditions, this procedure works well. However numerical problems
do occur. These problems will now be discussed along with a cure. Figure 2.10b shows
a second order example of when the above method fails. The boundaries are shaped
in such a way that searching along line lx and l2 yields no new boundary points. A
solution to this problem is to relax one of the constraining values by setting, for ex
ample, Ci equal to c + 6. d)Ci will then move so that point b can be found as shown in
figure 2.10c. Later, the relaxed constraining value can take on its original assignment.
2. We need to show that c, > c,+i when c, > Cf. In this case
Ci Ai > Cf (2.114)
Since Ki+i ICl and Ki+1 £ <9AA, then the following condition holds with strict
inequality.
^xt[P(A BKi+1) + (A BKi+1)TP]x < c V x : xTx = 1 (2.115)
Taking (2.110) for i + 1 then
^xt[P(A BKi+1) + (A BA'!+1)tF]x < A,-+i V x : xTx = 1 (2.116)
and there exists an x which satisfies the above condition for equality. Therefore
Ci > At-+i and Ci > max\c.f,\i+]] = c,-+1. Using similar arguments, it can be shown


K,¡ was found before 6 < .0001. The program ran 16 iterations. The final
result is
45
.00046
2.125
-.7974
.0805
.7535 '
II
2.057
-.0012
2.339
-6.577
.2591
(2.147)
-.00029
-.7259
.0263
-.1346
-.7985 _
Now to check the result.
II
^1
^min [ ,-y ( ^4
+ Ar-
BKf -
(2.148)
=
-818.0
(2.149)
Xf -
+ Ar-
BKj -
KjBT)]
(2.150)

-.3001
(2.151)
This meets the desired constraint.
Cf < Xf < \f < c¡


25
np
c > maxx
xe5
P Ax = max
xes
txT[iM + AtP]x
(2.46)
Comments:
Conditions (2.45) and (2.46) are respectively equivalent to
c < Amin[^(y/H)-lBl[AP-x + P~l At]B ^VH)~1] (2.47)
and
c > Amax[\('/HTlBl[AP-1 + P-lAT\B^VH)-1] (2.48)
where
H = (P-1BA.)TP~1B (2.49)
The matrix H is square, full rank, and has dimension n m. The above is true for
the following reasons. Equation (2.46) can be rewritten as
c > max ^xT[PA + AT P]x (2.50)
= max )-nTBl[AP_1 + P~lAT]Bxfi (2.51)
H1 H i=l
Since H is positive definite then y/H exists, is square, and has an inverse. By making
the substitution, // = (v/#)~1z, (2.51) becomes
c > max T(VH)-1BI[AP~1 + P~lA)B^VH)~1'z (2.52)
zTz=i2
which is equivalent to
C > Amax^iVHy'BliAP-1 + P-lA]B^VH)-1} (2.53)
Equation (2.45) is equivalent to (2.47) for similar reasons. We now give a proof of
Theorem 2.


Table A.6. Tabular data for Cy0
a/M
.8
1.2
2.0
2.7
3.5
-8
-.2226
-.2052
-.2175
-.1717
-.1758
-4
-.2870
-.2668
-.2305
-.2195
-.1911
0
-.3096
-.3099
-.2546
-.2478
-.2096
2
-.3210
-.3235
-.2660
-.2579
-.2299
4
-.3200
-.3363
-.2717
-.2629
-.2478
8
-.3113
-.3614
-.2851
-.2623
-.2812
12
-.2918
-.3774
-.3238
-.2925
-.3178
16
-.2445
-.3447
-.3849
-.3698
-.3783
20
-.2515
-.2730
-.4800
-.4440
-.4631
Table A.7. Tabular data for Cn6r
a/M
.8
1.2
2.0
2.7
3.5
-8
-1.0842
-.9703
-.4520
-.4236
-.2831
-4
-1.0743
-.9505
-.4600
-.3709
-.2739
0
-1.0446
-.9683
-.4900
-.3818
-.3015
2
-1.0356
-.9802
-.4900
-.4000
-.3162
4
-1.0257
-1.0050
-.5080
-.4091
-.3272
8
-1.0000
-1.0535
-.5200
-.4436
-.3640
12
-.9505
-1.0891
-.5300
-.5200
-.4412
16
-.8851
-1.0931
-.5760
-.6400
-.5441
20
-.7921
-1.0475
-.6840
-.7709
-.6838


84
M M
Figure 7.1. A plot of the kqu verses M with (a) various constant values of Q and
a = 8, and (b) a least squares third order polynomial fit for the plot where Q =
150psf and a = 8.
of adjacent grid points. Linear models were also made at each test point where /?, p,
q, and r were set to zero. The eigenvalues of
\[P,(A, ~ B,K,) + (A, B,K,fP,} (7.1)
and
^ (Apr Sp'pJ\.pr'j ~f" Bpr Kpr ) Ppr ] (7-2)
were computed to see if they remained within the desired limits. Table 7.1 shows
the maximum eigenvalue of (7.2) for all of the test points at a = 1.75. These values
are plotted against their indices in Figure 7.3. Most of the eigenvalues of Table
7.1 lie within the desired limits of 25 and 15. The eigenvalue in seventeenth row,
second column, however, is 3.96, the worst value found out of all of the test points.
Although the deviation is high, this value is still negative indicating stability for the
closed loop system. For the pitch controller, the actual limiting values of and A2
at all the test points are
-1.37 x 106 < A,i < -9.85 x 105
-47.7 < A?2 < -38.1
(7.3)


67
the previous point become the initial guess for the present point. The result of
this algorithm is a series of gains; one set for each flight condition generated. The
next portion of this section discusses some modifications made to the algorithm from
section 2.4. The material which follows describes how the design constraints are
selected.
The design method of section 2.4 finds a K so that the eigenvalues of |(PA+ATP)
fall between c¡ and c/ where A is the closed loop system. The design algorithm
used in Figure 5.1 has been modified so that the resulting K satisfies the following
requirements.
Qi < Ai
< Ao
2
\[P(A BK) + (A BK)tP]
\[P{A BK) + (A BK)tP}
< cx
< c2
(5.31)
cn < An [\[P{A-BK) + (A-BK)tP}\ < cn
where Ai is the smallest eigenvalue, A2 is the second smallest eigenvalue, and so on.
The values of the cs are supplied by the designer. This modification has been made
with the belief if more design constraints are made, then the performance will vary
less with changes in the flight conditions. The modified algorithm is as follows.
1) Choose P, cxj, ci/, ... cnf, and cnf. The selection of and cn¡ must
obey Theorem 2. However, Theorem 2 does not guarantee the nonempti
ness of K.
2) Compute c10,Cio, ... ,cn0,cn0 so that the initial guess, Ko £ and
K,} C /C0.
3) Let f = 0
4) Let i i + 1
5) Compute Kt so that K{ 6 i and ' ^ 5AT,_i.
6) Compute cu. ci, ... cn, cn so that Kt 6 dICi and C /Ct.


APPENDIX A
AERODYNAMIC DATA FOR THE EM R A AT AIRFRAME
This appendix gives aerodynamic data for the extended medium range air-to-air
technology (EMRAAT) airframe. The aerodynamic coefficients are given in tabular
form. All coefficients that correspond to angle rates have no dimensions and must be
multiplied by All coefficients that correspond to angle positions are given in per
degrees and must be multiplied by ^.
Missile Reference Diameter
Missile Reference Area
d = .625ft
S = .3067ft2
Table A.l. Tabular data for C/v
a/M
.8
1.2
2.0
2.7
3.5
-8
-4.816
-5.472
-5.104
-4.502
-3.952
-4
-2.408
-2.736
-2.552
-2.251
-1.976
0
0.000
0.000
0.000
0.000
0.000
2
1.204
1.368
1.276
1.126
.988
4
2.408
2.736
2.552
2.251
1.976
8
5.264
6.072
5.644
5.011
4.456
12
8.613
10.068
9.456
8.458
7.528
16
12.467
14.674
13.898
12.490
11.092
20
16.800
19.730
18.610
16.732
14.860
106


56
where d is the missile diameter. Again, the aerodynamic coefficients come from
wind tunnel testing and are tabulated in the first appendix. Eulers three moment
equations can be written as follows.
l '
m
- J
P '
q
n
r
where
J =
+ H(p,q,r)
Ixx IxY Ixz
Ixy Iyy Iyz
Ixz Iyz Izz
(4.29)
(4.30)
and
-(Iyy Izz)qr + Wz(r2 q2) Ixvpq + Ixyrp
H = ~(Izz Ixx)rp + Ixz(p2 ~ r2) Ixyqr + Iyzpq (4-31)
-(Ixx lYY)pq + Ixy(q2 p2) Ivzrp + Ixzqr _
We will assume that the inertial cross products, Ixy, Ixz, and Iyz are small. Also
due to symmetry of the airframe, Iyy and Izz are assumed to be equal. Inertial data
for the EMRAAT missile is given in the second appendix. Solving (4.29) for p, q,
and f gives
P =
Ixx
l
q =
Iyy
and
1
r =
Izz
[m + (Izz ~ Ixx) rp]
[n + (Ixx Iyy) pq]
(4.32)
(4.33)
(4.34)
Substituting (4.26), (4.27), and (4.28) into (4.32), (4.33), and (4.34) gives
QSd
q =
p =
QSd
h
Ixx
\plpP + ClpP + Clrr + Cl6pdp + Cl6rST
YY
~t~ Cmqq 4" SJm6c¡8q
+
Izz Ixx
Iyy
-rp
(4.35)
(4.36)
and
QSd r
Izz
Cng/3 + CnpP + Cnrr + Cns 8p + Cn¡j.S
, Ixx Iyy
+ j pq
izz
(4.37;


91
Table 8.1. The rise times of each commanded state.
Altitude(ft)
Mach
%(s)
ir/3(s)
r-p(s)
20,000
2.0
TOO
.243
.0284
50,000
3.0
.108
.316
.0181
0/s
for 0 s < t < 1 s
100%
for 1 s < t < 1.5 s
0%
for 1.5 s < t < 2.75 s
100%
for 2.75 s < t < 3.25 s
-100%
for 3.25 s < t < 3.75 s
0%
for 3.75 s < t
(8.4)
The experiment was performed once at an initial altitude of 20,000 ft with an initial
mach number of 2.0 and a second time at an initial altitude of 50,000 ft with an initial
mach number of 3.0. Figure 8.2 shows the results of the first test. In addition to the
commanded states, figure 8.2 also shows the remaining two states q and r. The rise
time as defined as the time needed to achieve 90% of the desired value was found
t
for each commanded state and is shown in table 8.1. It should be noted that mach
number and dynamic pressure do not change significantly during this test. This is a
test of accuracy in state tracking and is not a valid test of gain scheduling in terms
of M and Q. However, a changes very rapidly and does not appear to hinder the
performance.
Cross coupling is evident from this experiment. The step in the roll rate of figure
8.2c at t = 2 s is due to the sideslip command in figure 8.2b. A lesser degree of cross
coupling occurs during the roll command for 2.75 s < t < 3.75 s when a which is at
10 is rotated into sideslip. The same effects are present in the second experiment at
50,000 ft.
Figure 8.2d and 8.2e show the two states that have no commands. The spikes in
the pitch rate occur when ac changes value, because a pitching maneuver is required
to change the angle of attack. The two spikes in the yaw rate occur for the same


16
Figure 2.4. An example of a second order system with negative radial velocity com
ponents
Figure 2.5. One velocity vector on the ellipse xTPx = 1


63
Equations (5.10) and (5.11) are used to eliminate p and V in the linear model. As
a result, the linearized model can now be generated from the following seven flight
parameters.
M Q a ¡3 p q r
5.2 The Flight Parameter Generator
(5.12)
A series of flight conditions are made and used to generate many linear models.
Feedback gains are generated for each condition. The first of the series is called
the nominal flight condition. The values of the parameters for the nominal flight
condition are
M = 2, Q = 1250 psf, a = 8, (3 = 0,
p = 0/s, q = 10/s, r = 0/s.
Next, one of the parameters is allowed to vary while the other six are held constant.
This process is repeated six times so that each parameter can be tested. Table 5.1
shows the starting point, and the minimum and maximum values of each changing
parameter. Parameters with nonzero starting points begin at the starting point,
increment to the maximum value, return to the starting point, and then decrement
to the minimum value. Due to symmetry, the remaining variables have starting points
at zero and increment to their maximum values. M and Q sweep through a narrow
range because of restrictions of the atmospheric tables.
5.3 Initializing the Iterative Lyapunov Design Method
The iterative Lyapunov design method requires an initial guess, Kqo and Kpro and
two positive definite matrices, Pq and Ppr. The initializer supplies these values and
will be discussed in this section.


13
inside one of the positive regions. Since the velocity of the system always has an
outer radial component, the magnitude of the state vector is always increasing so
that the system (2.1) is unstable.
A direct relationship exists between the positive and negative region boundaries
and the eigenvectors of the system. In this case, the eigenvectors are rotating clock
wise at the same angular velocity as the region boundaries. It can be shown that if a
system has constant eigenvectors with time varying eigenvalues which have negative
real parts for all time then the system is stable. The stability question, however, is
not as easy to answer for the case with moving eigenvectors.
If the system has no positive regions, then it will be stable regardless of how
quickly it changes. However the converse is not true in general. Figure 2.3 shows a
state vector and its corresponding velocity for some dynamic system. The angle d
can be computed in the following way.
cos($) =
T
XxX
XX
For x to have an inner radial component then
This is true when
where
t 3tt
- < 9 <
2 ~ 2
xTx < 0
x = Ax
(2.5)
(2.6)
(2.7)
(2.8)
So if xT,4x is negative everywhere for all time, then the system is stable. Since the
system is linear, it is sufficient to check the sign of xTAx on the unit circle. If all
velocities on the unit spheroid point inside that boundary, then the same is true for


A CONTROLLER DESIGN METHOD WHICH APPLIES TO TIME VARYING
LINEAR SYSTEMS
By
KURT WALTER KOENIG
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1994


9
and A is bounded. If 0 is an asymptotically stable equilibrium point of z = A(t)z for
all time, then 0 is a locally stable equilibrium point for the system
X = /(*,x)
These ideas can be applied in the following way. Given a nonlinear system
x = /(x,w,u)
define
A(x,w) =
df
dx
B{x,w) =
(x.wfi
df
du
(1.19)
(1.20)
(x,w)
where x is a vector containing the states and w is a vector containing additional
system parameters. At regions near the operating point, the system becomes close
to
Ax = A(x, w)Ax + B(x, w)Au (1.21)
where Ax and Au are small perturbations between the states and inputs and the
operating point. We would like to find a feedback control law,
u = K(x, w)x
(1.22)
so that
-P[A(x, w) B(x, w)K(x, w)] + [A(x, w) B(x, w)K(x, w)]rP (1.23)
is negative definite for all x and w where P is positive definite. If such a control law
is found, then the perturbations from the nominal trajectory are locally stable for
the system
x = /(x,w,-T(x,w)x) (1.24)
Shahruz and Behtash [12] give one control law which places some of the eigenvalues
of (1.23) for the case where P = I. However unnecessary limitations on where the


65
where S is an invertible matrix. The diagonals of J are the real parts of the eigen
values of /l, and imaginary parts of the eigenvalues lie in skew symmetric locations
off of the diagonals. For example, if the eigenvalues of A are a jb, c, then
J =
a b 0
b a 0
0 0c
Let
J Jdiag ~t" Jskew
(5.17)
(5.18)
where Jdiag is the symmetric part of J and Jskew is the skew symmetric part of J. In
the above example
a
0
0 '
' 0
-b
0 '
0
a
0
Jskew
b
0
0
0
0
c
0
0
0
Jdiag
Making the following transformation on the system x = /lx, let
x = Sz
Then
Si = ASz
and
(5.19)
(5.20)
(5.21)
i S 1ASz = J z. (5.22)
Let Pz = Jdiag We wish to check the velocities of the system z = Jz on the ellipsoid
'Td~- ~Tr- --1 (5.23)
ZTPzZ = -zTJdiagz = 1
The normal to the ellipsoid at z is JdiagZ- The projection of z = Jz onto the normal
is JdiagZ. Here the normal velocity component has the same magnitude but opposite
direction to the normal of the ellipsoid. If the system has no complex eigenvalues
then the velocities are orthogonal to the ellipsoid everywhere. For this reason, the


95
a) Commanded z Acceleration (-) and Actual z Acceleration ()
Figure 8.4. Commanded and actual (a) z accelerations and (b) y accelerations for
the scenario in figure 8.3.


The design algorithm finds feedback gains which place the eigenvalues of the
derivative of a given Lyapunov function. Limitations on placement of these eigenval
ues are stated. This concept is applied to a second order linear time varying system
where ordinary pole placement techniques fail. The design method is then applied
to a linearized model of the EMRAAT missile which is a function of time varying
flight parameters. Feedback gains are generated as a function of these flight param
eters. It is discovered that the gains depend on dynamic pressure, mach number,
and angle-of-attack. A combination of interpolation and polynomial fitting is used
to create a look-up table for the gains. The resulting controller is programmed into
a nonlinear simulation which runs missile and target scenarios. Small miss distances
are achieved.
v


57
If gyroscope effects are considered small, then (4.24), (4.25), (4.35), (4.36), and (4.37)
can be written as two separate systems: the pitch dynamics, and the roll-yaw dy
namics. The pitch dynamics are as follows.
a =
1 +
QSCna
-ll r
mV
QSr \ (y Sr \
77CNa ) a + 1 tyCV 9 +
mV J \ mV
QS
mV
CNSq ) tig
9 =
QSd
iyy
T Ct -(- Cm<¡q T Crfiiq^q
(4.38)
(4.39)
The roll-yaw dynamic equations are
' (W *(+%>) *(-*+3*)^0 *
(4.40)
(4.41)
P =
r =
QSd
Ixx
QSd
[chp + clpp + Clrr + Chp8p + C,trSr
izz
\Cnpfi + CnpP + CnTr + Cnsp6p + Cner6r
(4.42)
The states of the pitch model are a and q with Sq as its input. For the roll-yaw
model, the states are ¡3, p, and r, and the inputs are 6P and Sr.
4.2 The Linear Model
The previous section gave a model of the pitch dynamics and the roll-yaw dynam
ics in the following form.
[q q }T = fg(a,q,Sq) (4.43)
[/5 p r }T = fpr(P,p,r,8p,8r) (4.44)
We would like to have two linearized models for use with the proposed design method.
The result will be two systems in the following form.
x = Ax + Bn
(4.43)


Table A. 16. Tabular data for CUp
a/M
.8
1.2
2.0
2.7
3.5
-8
121.7839
121.3270
44.3658
28.8446
16.7480
-4
98.7909
94.1548
41.3766
27.5697
16.7480
0
76.4321
66.1137
37.2075
26.7729
16.4228
2
64.9356
51.5798
34.2183
25.9761
15.9450
4
53.1219
37.9147
30.9931
25.0996
15.7317
8
30.2081
9.3997
24.0708
22.3904
13.4959
12
7.6115
-18.1675
16.9912
21.0359
13.0081
16
-15.7780
-46.9984
10.9341
20.3487
12.8455
20
-38.5332
-77.0142
6.3717
20.1594
12.7642
Table A. 17. Tabular data for Civ,
M
cnq
.8
-149.068
1.2
-134.226
2.0
-87.602
2.7
-77.176
3.5
-75.149
Table A.18. Tabular data for Cmq
M
Cmq
.8
-1981.268
1.2
-1900.576
2.0
-1218.300
2.7
-1040.346
3.5
-1007.925


44
Theorem 2 guarantees the nonemptyness of 1C and 1C if c¡ > .3017 and
cj < .7263. With this in mind, we let c¡ = .3 and = 6000. Since
the intersection of 1C and 1C is not guaranteed, Cj was chosen to be very
negative to increase the probability of getting an answer.
Step 2)
So
A ^min [^(A + Ar)]
(2.136)
= -781.4
(2.137)
(2.138)
^ = + AT)]
(2.139)
= 778.9
(2.140)
(2.141)
c0 = mm(cy, A)
(2.142)
= -6000
(2.143)
Co = max(cf, A)
(2.144)
= 778.9
(2.145)
Steps 3) 7)
A program was written for MATLAB to carry out the iterations in steps
3) 7). The program would terminate if 1C j was found or when
8 |Ci ct_i| < e = .0001
(2.146)


40
1) Choose P, Cf, and c/. This selection must obey Theorem 2. It should
be noted that Theorem 2 guarantees the nonemptiness of tCf and ICf, but
not their intersections. If the system is time varying, then, in order to
use this algorithm, P must be found so that
\Bl{t)[A{t)P~l + p-1AT(t)\BL{t) < 0 Vi (2.119)
Otherwise this algorithm cannot guarantee stability.
2) Compute c0 and c0 so that K0 = 0 6 dK-o and JCf C fCo- K0 will be
the initial guess in the search for Kf G
3) Let i = 0
4) Let i = i T 1
5) Find Ki so that K{ G and K{
6) Compute ct- and c so that Kl G dKx and K¡ C or so that G ICf.
7) Repeat steps 4) 5) and 6) until one of two events occur.
1. c = Cj- and c = cj.
2. (c ci_1) and (c2 c_i) become very small.
Remarks:
If event 1 occurs, then )Cf is nonempty and Kt G )Cj. If event 2 occurs, then IC/ is
empty and Ki yields a closed loop system that meets the constraints corresponding
to c and c,.
We now explain how to perform steps 2), 5), and 6).
Step 2)


7
Sigma max[E(t)] vs. Time
Figure 1.4. A plot of crmax[E{t)\ versus time.
E(t) is the result of subtracting this matrix from the closed loop system from the
rest of the trajectory. Solving for P in (1.12) gives
P =
17.29 -.0454
-.0454 .012
;i.i5)
Computing the upper bound in (1.13) gives
1
= .0578
(1.16)
Figure 1.4 shows the plot of crmax[E(t)\. Obviously since amax[E{t)\ is greater than
.0587 for most of the trajectory then (1.13) is not satisfied. This procedure was
repeated by taking the constant closed loop matrix from every point in the trajectory,
and the result was the same. The time varying linear system from this trajectory is
known to be stable because a Lyapunov function exists which can show this. The
upper bound in (1.13) can not be easily used if at all to support the claim of stability.
Desoer [9] gives a stability limit on the rate of variation of a given closed loop
time varying system. Wilson, Cloutier, and Yedavalli [1] give a stability limit on


97
a) Top View of Missile (-) and Target () Trajectory
Figure 8.6. (a) Top view and (b) side view of missile and target trajectories of
a scenario which occurred at 40,000ft. The miss distance is .05 ft. Initial mach
numbers of the target and missile are 3.0 and .92 respectively.


55
form and are given in the first appendix along with other data pertaining to the
EMRAAT missile.
Substituting (4.18) and (4.19) into (4.16) and (4.17) and solving for and /? gives
3 pot r + [Cy3/3 + Cypp + Cyrr + CySp6p + Cy6rST
and
QSCNa QS ^
4 77 ot = q-pp ¡7 CNaa + CNqq + CNsq8q
a =
1 +
mV
QSCN
-il r
mV
mV
QS
q- p(3 y. (CNaa + CNqq + CNiqSq
(4.21)
(4.22)
(4.23)
Equations (4.21) and (4.23) are two nonlinear state equations. In order to separate
pitch dynamics from roll-yaw dynamics, /5 is assumed close to zero in (4.23). Thus,
a =
1 +
QSCNo
mV
-i
Qs \ f QS
:Cn0 ) a + ( 1 7777Cjv, ) q +
mV
mV
QS
mV
a
NSq q
(4.24)
Equation (4.21) can be written as
*-(&) *(+*(-* + 3*)-O*) *
(4.25)
We now turn to the moment equations of the missile. Since thrust is zero and since
gravity does not contribute any moment to the missile, then the moments around
the x, y, and 2 axis are given by /, m, and n respectively, the moments due to
aerodynamic pressure. They are given by
l QSd [Cip/3 4- Cipp + Cirr + Ci6pSp 4- Ci6t8t
m QSd CrriaQ 4 CmtC T Cmqq 4~ Cm6q8q
(4.26)
(4.27)
and
n QSd Cnpfl + CnpP + Cnrr + CnSp8p 4" CnSr8-
(4.28)


CHAPTER 6
COMPUTING LOOK-UP TABLES
In the last chapter we showed that the gains depend mostly on a, M, and Q.
This chapter describes the process of generating a look-up table of gains verses M,
Q, and a. First a grid of points is formulated. Design constraints are then formulated
based on the knowledge of uncontrollable velocity components of the linear models.
Finally, the gains are computed.
6.1 Determining a Grid of Points
A two dimensional grid of points for M and Q has been made and used for each
entry of a in the table for both the pitch channel and the roll-yaw channel. Q sweeps
through a wide range of values starting with 100 psf and ending at 15,000 psf. The
values of M were chosen so that each entry of M and Q lie in the atmospheric tables
used to compute p and V. As a result, the grid points are not rectangular. All
entries are restricted to values between M = .6 and M = 3.5 and must correspond
to altitudes between sea level and 50,000 ft.
Table 6.1 shows the values of a used in the look-up table for both channels. M
and Q sweep through all values of the grid previously mentioned for each value of a
in Table 6.1. The spacing between the grid points was determined in a trial and error
process. During the iterative procedure for computing feedback gains, the initial
guess for each point came from the result of an adjacent grid point. The closer the
spacing between adjacent points, the fewer iterations were needed to find the next
feedback gain. Numerical problems as described in section 2.4 and shown in figure
2.10 were encountered. When this happened some of the constraints were relaxed
so interior points in the feedback set could be found. Later these constraints were
76


53
zero. Also, the weight of the missile is small in comparison to the aerodynamic force
so gravity is neglected. The aerodynamic force in the x direction is much smaller
than the aerodynamic force in the y and 2 directions, and therefore, will be ignored
for the rest of the paper. Newtons second law of motion implies the following:
Fy = m[v + ru pw] (4.1)
Fz = m[w + pv qu\ (4.2)
Fy and Fz are the aerodynamic forces in the y and 2 directions. The quantities
inside the brackets are the total accelerations in the y and 2 directions. We know
that
V = [u2 + 2 + w2}* (4.3)
Since v and w are much smaller than u then
V ~ u
Angle of attack and sideslip are given by
a arctan
ft = arctan |
If we assume that a and ft are small, then
w
a ~
u
and
u
Equations (4.1) and (4.2) can be rewritten as
Fy = mu
V w
b r-p
.u u.
(4.4)
(4.5)
(4.6)
(4.7)
(4.S)
(4.9)


96
a) Commanded Alpha (-) and Actual Alpha (--)
Figure 8.5. Commanded and actual states, (a) a, (b) /3, and (c) p. for the scenario
in figure 8.3.


105
varying system. Under this restriction, the algorithm can be used to achieve local
stability for a nonlinear system by formulating a linearized model around a time
varying operating point. Using this model, feedback gains are then computed as a
function of parameters which characterize the operating point. There is no restriction
on the rate of change of the system parameters.
In stabilizing an time varying linear system, it is not known how to choose a
constant positive definite P, or even if a P exists so B'[(t)[A(t)P~1 + P~l AT(t)]Bj_(t)
is positive definite. Further study in this area would be warranted.


Table A. 19. Tabular data for C/v
M
CN;
.8
163.215
1.2
198.109
2.0
168.132
2.7
172.199
3.5
164.634
Table A.20. Tabular data for Cm
M
cm.
.8
442.575
1.2
532.153
2.0
94.406
2.7
15.694
3.5
-27.525
Table A.21. Tabular data for Cyr
M
Cyr
.8
219.964
1.2
186.303
2.0
92.659
2.7
53.447
3.5
40.197


86
Table 7.1. The maximum eigenvalues of (6.1) for all test points at a 1.75.
Q/M indices
1
2
3
4
5
6
7
8
1
-20.79
-20.82
-20.84
-20.87
-20.88
-20.89
-20.90
-20.90
2
-20.68
-20.69
-20.70
-20.72
-20.74
-20.76
-20.76
-20.76
3
-20.84
-20.85
-20.84
-20.83
-20.81
-20.79
-20.78
-20.79
4
-20.59
-20.61
-20.49
-20.30
-20.18
-20.33
-20.34
-20.21
5
-20.30
-20.22
-19.93
-19.71
-19.96
-19.88
-20.01
-20.62
6
-20.41
-19.97
-19.77
-20.16
-19.88
-20.20
-20.79
-20.97
7
-19.10
-17.70
-17.86
-18.92
-20.03
-20.42
-19.75
-20.81
8
-18.62
-19.79
-20.67
-20.37
-16.16
-19.14
-20.21
-19.48
9
-19.90
-20.97
-19.33
-14.48
-19.58
-21.08
-20.93
-19.64
10
-20.23
-20.88
-11.67
-16.89
-19.96
-20.40
-19.72
-19.41
11
-20.57
-20.38
-12.40
-18.26
-20.60
-20.85
-20.28
-20.30
12
-20.77
-17.82
-17.66
-20.97
-21.85
-21.65
-21.14
-21.32
13
-19.83
-18.51
-21.91
-20.81
-20.87
-22.17
-22.13
-22.15
14
-20.45
-22.23
-20.57
-17.42
-20.84
-22.21
-22.42
-22.39
15
-22.42
-22.13
-18.83
-18.24
-20.65
-21.23
-21.49
-21.90
16
-21.82
-19.45
-15.17
-20.54
-21.09
-20.92
-20.94
-21.73
17
-21.85
-3.96
-20.63
-22.26
-22.35
-22.13
-22.03
-22.48
18
-18.88
-19.74
-21.48
-22.29
-22.58
-22.65
-22.68
-22.87
19
-20.85
-21.06
-21.71
-22.27
-22.69
-22.98
-23.19
-23.37
20
-21.65
-21.93
-22.29
-22.63
-22.93
-23.19
-23.39
-23.52
21
-20.62
-22.15
-22.52
-22.62
-22.61
-22.53
-22.46
-22.42
Figure 7.3. A plot of the eigenvalues from Table 6.1.


TABLE OF CONTENTS
ABSTRACT iii
CHAPTERS
1 INTRODUCTION 1
1.1 Eci.rlier Works 3
1.2 Purpose 10
2 THEORY BEHIND THE DESIGN METHOD 11
2.1 A Geometric Interpretation of Lyapunovs Linear Stability Theorem 11
2.2 Adding Control to the System 18
2.3 A Linear Feedback Set to Control xTPx 23
2.4 One Linear Feedback Matrix to Control xTPx 34
3 A TIME VARYING SECOND ORDER EXAMPLE 46
4 DERIVATION OF A MODEL OF THE EMRAAT MISSILE 51
4.1 The Nonlinear Model 51
4.2 The Linear Model 57
5 THE DEPENDENCE OF GAINS ON FLIGHT PARAMETERS 60
5.1 Generating p and V from M and Q 60
5.2 The Flight Parameter Generator 63
5.3 Initializing the Iterative Lyapunov Design Method 63
5.4 The Iterative Lyapunov Design Method 66
5.5 Formulation of a State Tracker 71
5.6 Comparing Gains with Flight Parameters 72
6 COMPUTING LOOK-UP TABLES 76
6.1 Determining a Grid of Points 76
6.2 Formulation of :he Design Constraints 77
6.3 Generating the Look-Up Table 80
ii


49
Figure 3.2. The time varying feedback set which satisfies the design constraints.


20
Figure 2.7. A case in which the normal velocity component cannot be controlled.
Theorem 1
Consider the system, x = A(t)x + B(t)u, where dim(x) = n, dim(u) = m
and A(t) and B(t) have compatible dimensions. Let P be a constant
positive definite matrix. The normal velocity component, xTPx, can be
arbitrarily set to any value with the right choice of u at a given time t,
at any point x, on the spheroid, xTx = 1, unless x is contained in the set
S(t) := P~1span(B(t)) fl {x : xTx = 1} (2.22)
where B(t) is a basis of column vectors orthogonal to span(B(t)). If
x S(t) then
xTPx = xTPA(t)x (2.23)
and this velocity component cannot be controlled at time t.
Theorem 1 gives the parts of the unit spheroid for which the velocity component
xTPx is uncontrollable and can therefore be used to determine if a linear time varying


19
Figure 2.6. A velocity vector (a) without control and (b) with control.
To answer question 1), consider a second order single input system frozen at some
instant in time.
x=Ax + F?u (2.21)
A is a 2 x 2 matrix, and B is a two dimensional column vector, u is a scalar input
and can take on any real value. If u = 0 then the velocity field of the system is
x = Ax. Figure 2.6a shows the velocity, x = Ax, of some state vector, x, in the
system. The control vector, B, is also shown. When m / 0, x has the extra term,
Bu. The direction of Bu is constant, but its magnitude is directly proportional to u.
Figure 2.6b shows many possible values for x by sweeping u through a wide range of
values in small increments. As this diagram shows, the arrow head of each velocity
vector can be placed on the line drawn parallel to B and intersecting the arrow head
of Ax. This demonstrates that velocities can be controlled along the space spanned
by the columns of B. The following theorem answers question 2).


Table A. 10. Tabular data for Ci0
a/M
.8
1.2
2.0
2.7
3.5
-8
.22653
.35501
.19609
.16015
.10417
-4
.20408
.26612
.17228
.12767
.09510
0
.13163
.12245
.08447
.04391
.02323
2
.07857
.03469
.01228
-.00701
-.01779
4
-.02106
-.06000
-.05953
-.06421
-.05336
8
-.17469
-.27837
-.18083
-.17934
-.10744
12
-.24204
-.33265
-.22214
-.22066
-.12232
16
-.24408
-.30857
-.22214
-.22030
-.11416
20
-.20857
-.27714
-.19200
-.19077
-.10417
Table A.ll. Tabular data for C¡Sp
a/M
.8
1.2
2.0
2.7
3.5
-8
-.13341
-.16085
-.07361
-.06506
-.05441
-4
-.14801
-.17360
-.09268
-.08207
-.07059
0
-.15265
-.17539
-.10111
-.09057
-.07353
2
-.15044
-.17204
-.10288
-.08725
-.06949
4
-.14314
-.16957
-.10067
-.07874
-.05956
8
-.13119
-.15459
-.09224
-.06617
-.04265
12
-.12345
-.13714
-.08115
-.06174
-.03860
16
-.11062
-.12371
-.07251
-.05989
-.04669
20
-.10111
-.11655
-.06829
-.05767
-.04853


kql 1alphaO
hq11*alphaO kqirajphaO
4
4^


93
reason when sideslip changes. The two steps in the yaw rate, however, are due to
cross coupling with the roll rate.
The rise times in table 8.1 are small and indicate that the missile will perform well
in flight scenarios. While cross coupling effects are noticeable, they are not believed
to be great enough to significantly hinder the performance of the missile; thus it was
decided to run the missile through a series of flight scenarios to see how well the
missile can intercept the target.
8.3 Simulation of Flight Scenarios.
A series of flight scenarios has been run to test the autopilot. Figure 8.3 shows
the trajectory of the missile and target for a flight scenario at an altitude of 20,000
ft. The miss distance is .64 ft. A hit is considered to be any miss distance under 10
ft. Figure 8.4 shows the commanded and the actual y and 2 accelerations. The devi
ations from the commanded normal accelerations are mostly due to a simplification
used in implementing the BTT logic. Instead of using two aerodynamic coefficients,
proportionality constants were assumed. The errors are not large enough to prevent
the interception of the target. The state tracking of the reference inputs appears to
be working well as seen in figure 8.5 and indicates that the autopilot is performing
well.
The trajectories of a 40,000 foot altitude scenario is shown in figure 8.6. The miss
distance is 0.05 ft. A similar scenario at 10,000 ft is shown in figure 8.7. The miss
distance is 2.8 ft.
Figure 8.8 gives a case were a miss occurred. The scenario took place at 50,000
ft and the target was missed by 452 ft. Figure 8.9a and b shows that the missile was
unable to achieve the desired 2 acceleration and angle of attack after 1.3 seconds into
the flight. This is because the elevator reached its 40 limit as figure 8.9c indicates.


4
Alpha vs. Time
Figure 1.2. Angle of attack from one trajectory of the EMRAAT missile.
bound on supt>0 ||A(i)|| is given which, if enforced, guarantees asymptotic stability
of (1.1). As will now be shown, this limit is too restrictive for the EMRAAT missile.
Figure 1.2 shows the angle of attack of the trajectory of a missile flown using an
already existing autopilot. Desoer proposes the Lyapunov function
V'(x,i) = xT(£l/ + P(f))x (1.2)
where ej > 0 and P(t) is chosen so that
AT(t)P{t) + P{t)A{t) = -3/
Desoer gives a bound on V,
V < xTx[3 + 2eicim +
where the following definitions are made:
A/ 3m4
4^0
a\f := sup ||A(i)|| < oo
t> o
(1.3)
(1.4)
(1.3)


23
Case 2: x ^ S(t)
Suppose we want to set xTPx = c where c is some arbitrarily chosen
value. Then we want
c = xtPA()x + xtPB()u (2.36)
Since x S(t) then xTP5(t)u ^ 0 and there exists at least one u such
that 2.36 is true.
One could solve (2.36) for u to use as a control law, but this would not be practical
to implement because A and B can not easily be computed. Linear state feedback is
more desirable. The next section will develop a procedure for computing the set of
feedback gains that will implement a feedback control law keeping the normal velocity
component, xTPx, within some specified limit.
2.3 A Linear Feedback Set to Control xTPx
Recall from Section 2.1 that the stability of a time varying linear system
x = A(i)x (2.37)
could be analyzed by evaluating the bounds of x1 PA(t)x on the spheroid
xTx = 1 (2.38)
Now we want to find the set of all linear state feedback gains for the control law,
u = -K(t)x (2.39)
for the system
x = A(t)x + B(t)u (2.40)
so that condition (2.19) holds for the closed loop system where c and c are now
specified. This section will give conditions for the nonemptiness of such a set followed


41
The lower and upper bound of xTPx for the open loop system i
tively,
Ao ^min[~^{P A + F P)\
Ao = Amax[(P A + AT P)}
Then,
c0 = min[cf, Aq]
c0 = max[cf, A0]
Step 5)
The algorithm is now given.
K = Ki
For j = 1 to m
For k = 1 to n
Po j,k P kjQj,k
Pi = -(PBQj,k + QlkBTP)
Po = P(A BKoj'k) + (A BKo,,k)TP 2cl
Solve detfkPi + P0] = 0 in terms of k for c = c,, and c,-.
E.eject all values which do not meet the constraints from
steps 2) and 6). The result is two intervals whose
lower bound is ki_ and k2 and whose upper bound is
k\ and 2
Find the intersection of these intervals by evaluating
respec-
(2.120)
(2.121)
(2.122)
(2.123)


CHAPTER 4
DERIVATION OF A MODEL OF THE EMRAAT MISSILE
The next section derives a nonlinear model of the EMRAAT missile. A linearized
model is then generated in the following section and will be used for the design of the
autopilot. Aerodynamic and inertial cross coupling are assumed negligible in order
to reduce the order of the model. A linearized pitch model and a linearized roll-yaw
model results. All assumptions are clearly stated.
4.1 The Nonlinear Model
Before deriving the nonlinear model some variables and terms are defined. Figure
4.1 shows the missile body coordinate frame of the EMRAAT missile [2]. The three
axes, x, y, and z, are fixed to the missile as shown. The velocity of the missile is
represented by V. Angle of attack, a, is defined as the angle between x and the
projection of V onto the x-z plane. Sideslip, is the angle between x and the
projection of V onto the x-y plane. The velocity components, u, u, and to, are the
projections of V onto the x, y, and z axis respectively. The angle rates, p, q, and r,
are named roll rate, pitch rate, and yaw rate respectively and are defined as the rate
of rotation around the x, y, and z axis. Their directions obey the right hand rule.
These definitions are similar to those applied to aircraft as given by Etkin [17].
The following derivation of the nonlinear model is based on a similar model found
in Smith [2 and in Koenig [18]. Assumptions are made here to separate the system
into two lower order models: the pitch model and the roll-yaw model. We begin the
derivation by starting with the force equations. The forces acting on the missile are
thrust, gravity, and aerodynamic forces. The autopilot design in this paper applies
to the second phase of the flight when the engine is no longer burning so thrust is
51


59
pr 21
b
pr 11
QSCySp
mV
QSdCip
, CLpr22
XX
bprl2
Ixx
QSCy6r
mV
, Clpr23

5pr21
QSdCn
QSdCl6p QSdC[(r
Ixx
"? bpr22
Ixx
'np QSdCnp QSdCnr
dpr31 j i &pr32 j ? &pr33
Izz
)pT 31
l QSdcnsp L
^nr31 j ) Vpr32
ZZ
Izz F-J Izz
QSdCn6r
Izz
The linearized model is given by
0'
dprll dprl 2 ^prl3
/J'
bprll bpri2
p
=r
&pr21 dpr22 dpT 23
P
+
6pr21 bpr22
r
Ctpr31 dpr32 dpr 33
r
bpr31 bpr32
(4.49)


Table A. 14. Tabular data for C¡.
ol/M
.8
1.2
2.0
2.7
3.5
-8
61.3158
4.0263
-15.2343
-9.6414
-10.8909
-4
74.2982
40.0000
-1.7188
-5.3386
-4.6446
0
86.3158
79.4748
9.0625
0.0000
2.2422
2
92.4561
99.2560
13.3594
.1593
5.2853
4
97.6316
117.4617
16.3281
.9562
7.6877
8
91.5789
151.6849
16.0156
3.1076
3.6036
12
86.8421
179.9562
15.6250
4.3825
0.0000
16
88.0702
198.4245
21.4063
6.0558
-1.2813
20
93.9474
294.0919
25.9375
7.8884
3.0430
Table A.15. Tabular data for Cyp
a/M
.8
1.2
2.0
2.7
3.5
-8
-9.2259
-7.7656
-3.9717
-3.1351
-2.0118
-4
-9.2417
-7.6484
-3.9717
-3.1583
-2.0197
0
-9.2338
-7.7656
-3.9152
-3.1390
-2.0039
2
-9.1945
-7.9375
-3.8799
-3.0579
-2.0039
4
-9.1866
-8.0937
-3.8233
-2.9730
-2.0039
8
-9.1945
-8.3281
-3.7385
-2.8108
-1.8619
12
-9.2259
-8.3203
-3.6396
-2.7027
-1.7357
16
-9.1081
-8.2969
-3.5830
-2.6718
-1.6331
o
O
-9.0688
-7.9609
-3.6254
-2.6873
-1.6095


5
(To is positive and
Re A,[A(t)] < 2ar0 Vz, Vi > 0
m is a constant and depends only on cr0 and aj^ so that
Also,
exp(T/l(i))|| < mexp( cr0r), Vt > 0 Vi > 0
hM := sup [|()||.
z>o
If ex is allowed to be very small, then from (1.4), stability would result if
4cr2
au <
77T
(1.6)
(1.7)
(1.8)
(1.9)
For the linear pitch model of the given trajectory, cr0 = 15.75. At .04 seconds into
the trajectory the closed loop matrix is
A =
-3.2533 .8997
-1678.2 -83.50
(1.10)
At this instant in time, m = 22.88 and au 2231. ^ = .0036. From this we see
that the linear system is changing much faster than the limit shown in (1.9). The
system is not slowly varying. Note that m was only found for i = .04 sec. and not
for all time. If a greater m were found for the rest of the trajectory, then the limit in
1.9 would be even more restrictive and the result would be the same. This test does
not show that the system is unstable. In fact, a Lyapunov function is known which
shows that this linear system is stable but Desoers theory can not support this fact.
Wilson, Cloutier, and Yedavalli [1] give a stability condition for a constant linear
system with time varying uncertainties. Given a system,
x [A + E(t)]x
(1.11)
if a positive definite P can be found which solves the Lyapunov equation.
PA + AtP + 21 = 0
(1.12)


52


99
The miss occurred due to the reduced elevator effectiveness from the low air density
at high altitudes.
Another miss is shown in figure 8.10. Figure 8.11 shows that the desired z ac
celeration was not achieved because the commanded angle of attack was at its 18
limit. The commanded normal acceleration was too great to be realized. Again, the
autopilot performed as well as it could, but the miss occurred because of physical
limitations of the missile. The results of these simulations indicate that the autopilot
does its job and, in reasonable scenarios, allows the missile to intercept the target.


REFERENCES
[1] Robert F. Wilson, James R. Cloutier, and R. K. Yedavalli, Control Design for
Robust Eigenstructure Assignment in Linear Uncertain Systems, Proceedings
of the 30th IEEE Conference on Decision and Control, Vol. 3, pp. 2982-6, 1991.
[2] Roger L. Smith, An Autopilot Design Methodology for Bank-to-Turn Missiles,
Air Force Armament Laboratory, Eglin AFB, August 1989.
[3] Jeff S. Shamma and Michael Athans, Analysis of Gain Scheduled Control for
Nonlinear Plants, IEEE Transactions on Automatic Control Vol. 35, No.8, pp.
898-907, 1990.
[4] Jeff S. Shamma and Michael Athans, Gain Scheduling: Potential Hazards and
Possible Remedies, IEEE Control Systems Magazine, Vol. 12, No. 3, pp. 101-7,
1992.
[5] Jeff S. Shamma and Michael Athans, Stability and Robustness of Slowly Time-
Varying Linear Systems, Proceedings of the 26th Conference on Decision and
Control, Los Angeles, December 1987.
[6] Jeff S. Shamma and James R. Cloutier, Gain-Scheduled Missile Autopilot De
sign Using Linear Parameter Varying Transformations, Journal of Guidance,
Control, and Dynamics, Vol. 16, No. 2, pp. 256-63, 1993.
[7] Kevin A. Wise, Barry C. Mears, and Kameshwar Poolla, Missile Autopilot De
sign Using Hoo Optimal Control with //-Synthesis, Proceedings of the American
Control Conference, Vol. 3, San Diego, 1990.
[8] Mario A. Rotea and Pramod P. Khargonekar, Simultaneous H2/H00 Optimal
Control with State Feedback, Proceedings of the American Control Conference,
Vol. 3, San Diego, 1990.
[9] Charles A. Desoer, Slowly Varying System x = A(f)x, IEEE Transactions on
Automatic Control, Vol. AC-14, pp. 780-781, December 1969.
[10] M. Vidyasagar, Nonlinear Systems Analysis, Prentice-Hall, Englewood Cliffs,
NJ, 1978.
[11] Wolfgang Hahn, Stability of Motion, Springer-Verlag, New York, 1967.
[12] S. M. Shahruz and S. Behtash, Design of Controllers for Linear Parameter-
Varying Systems by the Gain Scheduling Technique, Journal of Mathematical
Analysis and Applications, Vol. 168, pp. 195-217, July 1992.
118


6
Figure 1.3. Angle of attack from one trajectory of the EMRAAT missile,
then the system will be asymptotically stable if
max[E{t)\ < l- Vi (1.13)
&max\E )
where crmax is the maximum singular value. This idea is initially interesting because
it requires the error, E(i), to be bounded, but does not constrain the time variation.
The above result can not be easily applied to the EMRAAT missile. This will be
shown by applying the result to a trajectory of the EMRAAT missile whose angle of
attack is shown in figure 1.3. In order to check the stability of the EMRAAT missile,
A must be chosen since it may be any matrix which is stable. It is believed that the
best choice of A should come from a closed loop matrix in the trajectory. With this
in mind, A was taken at t = 2 sec into the trajectory.
-4.477 .9526
-1.684 -79.47
A-BK =
(1.14)


103
a) Commanded z Acceleration (-) and Actual z Acceleration (--)
c) Elevator Deflection Angle
Figure 8.11. Commanded and actual accelerations (a) in the 2 direction, (b) com
manded and actual angles of attack, and (c) elevator deflection angle for the scenario
shown in figure 8.10. Saturation of the a command resulted in the miss.


CHAPTER 7
GAIN SCHEDULING
The result of the previous chapter is thirteen tables of gains in terms of mach
number, dynamic pressure, and angle of attack. This chapter discusses the process
of curve fitting used to implement the look-up table. The results of a test of this
scheme will follow.
7.1 Curve Fitting
It was decided to use a combination of polynomial fitting and interpolation to
implement the autopilot. Third order polynomials were fit to the tables as a function
of mach number. However, low order polynomials could not achieve close fits as a
function of angle of attack or dynamic pressure, so linear interpolation was used for
these two variables. Figure 7.1a shows kqU as a function of M for various constant
values of Q at a = 8. An example of polynomial fitting of one of these curves is
shown in Figure 7.1b where Q = 150psf. A polynomial has been made for every
ol-Q pair and the polynomial coefficients are interpolated as a function of these two
variables. The tables of the three pitch controller gains each have 7 entries for a
and 22 entries for Q. Since each polynomial has 4 coefficients the total number of
coefficients for each gain is 7 x 22 x 4 = 616. Similarly the tables for the roll-yaw
controller have 8 entries for a and the same number of entries for Q. Each of the ten
gains are then scheduled using 8 x 22 x 4 = 704 polynomial coefficients.
7.2 Testing the Fit
Figure 7.2 gives the system for testing the polynomial and interpolation routines.
Gains were generated from these routines at locations centered between the original
grid points. These new locations were found by taking the average of the coordinates
83


34
2.4 One Linear Feedback Matrix to Control xT.Px
The search procedure given in section 2.3 becomes impractical for high order
systems with multiple inputs because the number of grid points for the fixed entries
of K becomes very large. This section discusses an iterative Lyapunov design method
which saves on computations and finds one element K /C, if it exists, where K, =
1C fl 1C. By applying this procedure at every instant in time to a time varying linear
system, one can find a control law stabilizing the system if a constant positive definite
P exists which satisfying the following condition.
^BTL(t)[A(t)P~l + P~1AT(t)\BL(t) < 0 Vi (2.107)
By Theorem 2, this condition guarantees the existence of a c < 0 such that K{t)
is nonempty for every instant in time. This procedure applies to all systems which
can be stabilized with respect to a Lyapunov function given by a constant positive
definite P. The following is a discussion of the iterative Lyapunov method followed
by the algorithm itself. An example is then given applying this procedure to one
operating point of a fifth order linearized model of the EMRAAT missile.
Figure 2.9 illustrates the iterative procedure. fCf is the feedback set which satisfies
the designers predetermined constraints for some specified P. The constraints are
£f < xt[PA + AtP PBK KTBTP\x Let ICi be defined as the set of all K which satisfies the following constraints.
SU < xt[PA + AtP PBK KtBtP]x Given Ki, we would like to find c, and c, so that if Ki ^ 1Cf then Kz 6 d)Ci. We also
require that ICf C K{. If Kz £ JCf, then we want JC{ = K¡. The following definitions
for Cj and c, meet these requirements. Let
A, := max -x.T[P(A BICC + (A BKCTPht
11*11=1 2
(2.110)


26
Proof:
We will show the nonemptiness of 1C. 1C can be shown to be nonempty in
a similar manner.
Case 1:
nr
maxx PAx < c
xeS
then
xTPAx < c V x S.
(2.54)
(2.55)
Let K = kBT P where k is a scalar value. Then
xTPx = ^xt[PA + AtP PBK KTBTP]x (2.56)
= \xT[PA + AtP]x ^k[xTPB}[BTPx\ (2.57)
Since xtPB = [BTPx]T then
xtPBBtPx> OVx^S, VxTx = 1 (2.58)
If (2.55) is true and if k is made large enough, then the second term in
(2.57) will dominate V x S and xTPx < c V x : xTx = 1. Since
kBTP e X then X is nonempty.
Case 2:
maxxTJDAx > c (2.59)
-x.es '
Then
nr
x P Ax > c for some x G S.
(2.60)
By Theorem 1, xPAx can not be controlled V x £ <5; so 1C is empty.


BIOGRAPHICAL SKETCH
Kurt Walter Koenig was born on July 30, 1967 in Rolla, Missouri. He graduated
with high honors from the University of Florida with a bachelors degree in electrical
engineering in December of 1989. He received a Florida Graduate Scholarship in
1990 to pursue a masters degree in electrical engineering at the University of Florida
which he received in December of 1991. He was admitted into the Palace Knight
program in 1992 and received support from the United States Air Force to work on
a Ph.D. in electrical engineering at the University of Florida.
120


92
Figure 8.2. A test of state tracking at an initial altitude of 20,000 ft with an initial
mach number of 2.0. The states shown are (a) a, (b) /3, (c) p, (d) q. and (e) r.


48
The uncontrollable velocity for this example is constant for all time. In selecting the
constraining values, we must have
c < -.9548 < c (3.9)
The following assignments are made
c -2 (3.10)
c = .5 (3.11)
Since the system is second order and has only one input it is possible to plot the set
of all feedback gains which satisfy the following.
c < ^xT[P(A(t) B(t)K(t)) + (A(t) B(t)K(t))TP]x < c Vx : xTx = 1, Vi (3.12)
This time varying feedback set is shown in figure 3.2. Since the time varying nature
of the system is periodic, and from inspection of figure 3.2, the following control law
is chosen.
u = 3[cos(i) sin(i)]x (3.13)
The feedback matrix in (3.13) is shown to be inside the moving feedback set in figure
3.2. The eigenvalues of the derivative of the resulting Lyapunov function is
A[A(<) B{t)K{t) + AT{t) KT{t)BT{t)} = -1.1193, -0.8579 Vi (3.14)
Therefore, the closed loop system is stable. Figure 3.3 shows a trajectory of the
Lyapunov based closed loop system where the initial condition is x0 = [1 0]T.


17
point on the ellipse, and its velocity, x, is also shown. The normal to the ellipse at x
is Px. To ensure stability, it is sufficient to require that the projection of x onto the
normal of the ellipse, Px, is negative everywhere. The resulting term, xTPx, is the
normal velocity component on the ellipse. The main thrust of this paper is to be able
to control xTPx and to make this normal velocity component negative everywhere
for all time in order to achieve stability. As before, the linearity of the system allows
one to only check xTPx on the unit circle. This generalizes the previous stability
condition to
Stability Condition 2: For the system x = A(t)x, if xTPx = xTPA(i)x <
0 V x : xTx =1, V t > 0 for some constant positive definite matrix P,
then the origin is a global asymptotically stable equilibrium point.
The above condition applies to linear systems of any order.
Lyapunov [14] stated that if any positive definite function of the system states
F(x) is always decreasing, i.e. V(x) < 0, then the system is stable (asymptotically
stable for linear systems) [10]. For the linear case, let
F(x)=xTPx (2.10)
V(x) is positive definite if and only if P is a positive definite matrix. Taking the
derivative gives
V(x) = xTPx + xTPx
(2.11)
= xTArPx + xTPAx
(2.12)
= xt[AtP + PA]x
(2.13)
Lyapunovs stability condition becomes
xt[AtP + PA]x <0 V x, V t > 0
(2.14)


15
Figure 2.3. x and x
all spheroids centered on the origin. This leads to the following stability condition
which is given without proof.
Stability Condition 1: If xTx = xTAx < 0 V x : xTx = 1, V i > 0 for
the system, x = A(i)x, then the origin is a global asymptotically stable
equilibrium point.
Figure 2.4 illustrates an example of a second order system which meets Stability
Condition 1. Note, that while all figures given so far represent second order linear
systems, the above stability condition applies to linear systems of any order.
The above stability condition can be made less conservative by expanding the
class of shapes to ellipsoids defined in the following way.
xTPx = 1 (2.9)
where P is a symmetric, positive definite matrix. Figure 2.5 shows an ellipse of the
form (2.9) defined for a second order system. The state vector, x, is drawn to some


35
Figure 2.9. A geometric view of the iterative Lyapunov design method.


CHAPTER 9
CONCLUSION
A controller design algorithm has been made which can stabilize time varying
linear systems. The motivation behind this work was the need to achieve local sta
bility of a nonlinear system by stabilizing a linearized model which is a function of
time varying system parameters. This work began by studying second order linear
time varying systems that were stable when frozen at any given time but were un
stable when allowed to vary with time. Geometric techniques were used to rederive
Lyapunovs linear stability theorem. These techniques were carried further to give a
condition for the existence of state feedback which places an upper and lower bound
on the eigenvalues of the derivative of a given Lyapunov function. This led to a design
algorithm which makes the derivative of the Lyapunov function negative definite. A
time varying second order system was then stabilized when ordinary pole placement
techniques failed. The algorithm was then applied to a linearized model of the EM-
RAAT missile and used to formulate an autopilot. Feedback gains were generated
along a grid of flight parameters. The gains were scheduled against angle of attack,
mach number, and dynamic pressure. The resulting autopilot was tested using a non
linear simulation. The missile was able to intercept the target in reasonable scenarios
and flight conditions.
The significance of the proposed design algorithm is that no assumption is made
on time invariance of the system. However as stated in chapter 2, this algorithm
can only be used to stabilize a system if there exists a constant positive definite P
such that ()[A()P_1 P~lAT(t)]BjL(t) is negative definite for all time. If this
is true then a K(t) can be found so that xTPx is a Lyapunov function for the time
104


30
Then (2.74) becomes
det[ kP1 + P0] = 0 (2.78)
Solve (2.78) for k.
3) Reject all complex roots. If all the roots are complex then skip the
next step.
4) Test the intervals between the real roots by checking to see if
Amax[PA + AtP Tel PBK KtBtP] < 0 (2.79)
The K's that bound the interval which satisfies (2.79) lie on the boundary
of 1C. Convexity of 1C implies that no more than two K's bound this
interval.
5) Repeat the process for all possible values for the fixed entries in K.
The result is dlC.
djC can be computed in a similar way by reversing the inequalities in the above
procedure and by replacing \max with Amn. To find 1C the intersection of 1C and 1C
can be found in step 4).
The fixed entries in step 5) must be assigned to a finite number of grid points if
the above procedure is to be executed on a real system. The spacing of these grid
points must be smaller than the size of the feedback set. If the system has a high
order or a multiple number of inputs, the number of grid points will become too large,
and it will not be practical to implement this method. Understanding this procedure,
however, leads to the formulation of an iterative method that can be used on high
order, multiple input systems and will be presented in the following section. First,
an example is given applying the exhaustive search method to a second order, single
input system.


64
Table 5.1.
'he initial point and range of changing flight parameters
Initial point
Minimum value
Maximum value
M
2
1
2.6
Q
1250 psf
700 psf
5000 psf
a
8
-8
20
P
0
0
10
P
0 /s
0 /s
500/s
Q
10 /s
1
H-
o
o
0)
200/s
r
0 /s
0 /s
200/s
The initial feedback gains are found by using a pole placement algorithm. At the
nominal flight condition, the linear models are given by
4 =
-1.1345 0.9996
-261.4732 0.6209
,Bq =
-0.1463
-123.1091
A
r-Lrpr
L
L
-0.459
0.140
-.100
0.018
0.117
-2255.5
-2.41
.066
, Bpr
-1173.5
-1335.6
73.0
-0.181
-.648
2.01
-114.4
(5.13)
(5.14)
The desired eigenvalues for the closed loop pitch dynamics have been chosen to be
40jl0. For the closed loop roll-yaw model the desired pole locations are 20j5,
80. The resulting feedback gains are
Kq =
-11.1633 -.6233
, KpT
-1.4166 -.0758 .3758
2.9337 .0086 -.3300
(5.15)
For both closed loop systems, P must be found so that xTPx is a Lyapunov
function. The following problem is stated.
Given a stable linear system x = Ax, find a positive definite function,
V = xTPx so that V = xt(PA + ATP)x is a negative definite function.
Let A be put into Jordan canonical form.
A = SJS
(5.16)


75
Table 5.5. Extreme values and range of the roll-yaw channel control terms as indi
vidual flight parameters vary. Gains depend mostly on M, Q, and a.
M
Q
a.
P
P
<7
r
control
min
min
min
min
min
min
min
term
max
max
max
max
max
max
max
diff
diff
diff
diff
diff
diff
diff
kprllfio
-.050
-.11
-.22
-.050
-.050
.050
.050
.022
.045
-.013
-.038
-.036
-.036
-.036
.072
.157
.21
.011
.014
.014
.013
fcprl2PQ
-.17
-.14
-.29
-.13
-.13
-.13
-.13
-.011
-.061
-.13
-.13
-.13
-.13
-.13
.16
.20
.15
.001
.001
.001
.001
^prl3^0
.074
.047
.15
.16
.16
.16
.16
.18
.28
.25
.16
.17
.17
.17
.10
.23
.099
.0006
.001
.001
.0009
kpr 2l/^0
.026
.0062
.065
.093
.092
.092
.092
.11
.167
.15
.10
.101
.101
.101
.080
.16
.084
.008
.009
.009
.009
kpr22Po
-.054
-.18
.0015
.014
.014
.014
.014
.015
.015
.152
.014
.015
.015
.015
.068
.20
.15
.0007
.0008
.0005
.0007
kpr 23^0
-.151
-.25
-.24
-.15
-.15
-.15
-.15
-.067
-.034
-.11
-.14
-.14
-.14
.14
.084
.22
.12
.001
.002
.001
.002
kref prll/^cO
-.14
-.21
-.16
-.14
-.14
-.14
-.15
-.064
-.050
-.13
-.13
-.13
-.13
-.13
.08
.16
.028
.011
.014
.013
.013
kref prl2PcO
-.071
1
O
-.39
-.040
-.040
-.045
-.040
.029
.15
.056
-.039
-.040
.013
1
O
CO
00
.10
.19
.45
.001
.001
.057
.002
kref pr2lfic0
.055
.030
.11
.12
.12
.12
.12
.126
.19
.17
.13
.13
.13
.13
.071
.16
.058
.008
.009
.010
.009
kref pr22PcO
-.010
-.33
-.17
-.071
-.071
-.12
-.071
-.070
-.070
.24
-.070
-.070
-.066
-.070
.027
.26
.41
.001
.001
.051
.002


Table A.8. Tabular data for Ci6r
a/M
.8
1.2
2.0
2.7
3.5
-8
-.11703
-.11700
-.08247
-.06413
-.04209
-4
-.12446
-.12390
-.08446
-.05906
-.04209
0
-.13189
-.12870
-.08705
-.06123
-.04420
2
-.13131
-.13020
-.08904
-.06304
-.04638
4
-.13249
-.13540
-.09223
-.06775
-.04964
8
-.13738
-.14780
-.09900
-.07536
-.05761
12
-.14305
-.16186
-.11036
-.08804
-.06812
16
-.14932
-.16660
-.11813
-.10181
-.08188
o
O
(M
-.14266
-.15690
-.13008
-.11775
-.09565
Table A.9. Tabular data for CySr
a/M
.8
1.2
2.0
2.7
3.5
-8
.1635
.1386
.06877
.04874
.04270
-4
.1639
.1331
.06996
.05054
.03745
0
.1649
.1357
.07036
.05269
.04007
2
.1608
.1408
.07194
.05305
.04232
4
.1535
.1474
.07292
.05556
.04719
8
.1452
.1494
.07273
.06201
.05618
12
.1467
.1553
.07668
.07168
.06780
16
.1414
.1498
.08360
.08208
.08052
o
O
.1101
.1231
.09368
.09570
.08951


42
k = max[ku k2\
k min[ki, 2],
Find the midpoint of the interval by computing
k j,k = \(k+k)-
Let I\ kj,kQj,k d- Kqj,k
Next k
Next j
Ki+i = K
Step 6)
Let
A = Xmin[\(PA + ATP-PBKi-KTBTP)\ (2.124)
A = Amax[i(PA + ATP-PBKi-KjBTP)] (2.125)
&
So,
Cj = min[cfi A] (2.126)
Ci = max[c/,A] (2.127)
The following example illustrates the feasibility in applying this method to the
EMRAAT missile.
Example 2
We would like to apply the iterative Lyapunov design method to the EM
RAAT missile. The missile was flown in a simulation through a trajectory


APPENDIX B
INERTIAL DATA FOR THE EMRAAT AIRFRAME
This appendix gives the weight and moments and products of inertia of the EM
RAAT missile with no fuel.
Missile Weight
W = 227 lb
Moments of Inertia
Ixx = 1-08 slug-ft2
Iyy = 70.13 slug-ft2
Izz = 70.66 slug-ft2
Products of Inertia
Ixy = 0.274 slug-ft2
Ixz = 0.704 slug-ft2
Iyz 0.017 slug-ft2
117


7 GAIN SCHEDULING 83
7.1 Curve Fitting 83
7.2 Testing the Fit 83
8 NONLINEAR SIMULATIONS 88
8.1 The Nonlinear Simulation 88
8.2 A Test of State Tracking 90
8.3 Simulation of Flight Scenarios 93
9 CONCLUSION 104
APPENDICES
A AERODYNAMIC DATA FOR THE EMRAAT AIRFRAME 106
B INERTIAL DATA FOR THE EMRAAT AIRFRAME 117
REFERENCES 118
BIOGRAPHICAL SKETCH 120


38
a b
kX2 k\2
Figure 2.10. Step 5 of the iterative Lyapunov design method for (a) a second order
example, (b) how it sometimes fails, and (c) how this problem is corrected.


Table A.2. Tabular data for Cm
a/M
.8
1.2
2.0
2.7
3.5
-8
6.976
12.336
11.376
8.445
7.472
-4
3.488
6.168
5.688
4.222
3.736
0
0.000
0.000
0.000
0.000
0.000
2
-1.746
-3.084
-2.844
-2.111
-1.868
4
-3.488
-6.168
-5.688
-4.222
-3.736
8
-7.762
-11.238
-10.662
-8.174
-6.908
12
-11.802
-15.530
-15.074
-11.439
-9.972
16
-14.880
-19.330
-19.282
-14.603
-13.624
O
CM
-16.030
-25.700
-29.560
-24.218
-25.930
Table A.3. Tabular data for C^6q
a/M
.8
1.2
2.0
2.7
3.5
-8
.153
.195
.095
.077
.048
-4
.181
.215
.106
.073
.055
0
.197
.219
.107
.069
.056
2
.199
.212
.104
.068
.055
4
.198
.207
.101
.067
.051
8
.199
.195
.091
.056
.046
12
.199
.175
.076
.047
.044
16
.188
.161
.064
.044
.047
20
.184
.150
.056
.047
.055


119
[13] Hassan K. Khalil, Nonlinear Systems, p. 175, Macmillan Publiching Company,
New York, 1992.
[14] A. M. Lyapunov, Stability of Motion, Academic Press, New York, 1966.
[15] Shannon Fields, unpublished Ph. D. work, private communication, University of
Florida, February 1994.
[16] H. L. Royden, Real Analysis, The Macmillan Company, New York, 1964.
[17] Bernard Etkin, Dynamics of Flight Stability and Control, Second Edition, John
Wiley and Sons, New York, 1982.
[18] Kurt W. Koenig, The Design of a Robust Autopilot for an Air-to-Air Missile,
M.S. Thesis, University of Florida, December 1991.
[19] Arthur E. Bryson and Yu-Chi Ho, Applied Optimal Control, p. 155, Hemisphere
Publishing Corporation, Washington, 1975.


54
Fz mu
'w v
+P <7
Lu u
which simplifies to
Fy = mV
Fz = mV
'v
f- r pa.
.u
'w .
+ p/3 q
L u
We assume that the forward velocity changes slowly so that
~ 0.
Then
and
So (4.11) and (4.12) become
w
a ~
u
P=~
u
mV
Fz_
mV
= (3 + r pet
L + p/3 q
The aerodynamic forces are given by
Fy = QS [cYl,0 + CYp +CYrr + Cr6r + CyJ,
Fz = QS Cjvaa + Cn0c* + Cjv9 Q is the dynamic pressure and is defined as
Q = \pv2
(4.10)
(4.11)
(4.12)
(4.13)
(4.14)
(4.15)
(4.16)
(4.17)
(4.18)
(4.19)
(4.20)
where p is the air density. S is the surface area of the wing. The aerodynamic
coefficients come from wind tunnel tests. They depend on mach number, and some
depend on angle of attack. The values of these coefficients have been put in tabular


20 >
Figure 6.3. kqU verses M and Q when a = 8


47
Trajectory of a pole placement time varying system
Figure 3.1. The trajectory of the above closed loop system based on pole placement.
The initial conditions are x0 = [1 0]T. The system is unstable.
If the initial conditions of the system are Xo = [1 0]r, the resulting trajectory is
unstable as figure 3.1 shows.
We now turn to the Lyapunov design method. We choose P to be the identity
matrix. Before giving the design constraints, we need to check the the value of the
uncontrollable normal velocity components for all time. At t = 0
B = [.9848 .1736]t (3.6)
Since P is the identity matrix, we are interested in the normal velocity component
which is on the part of the unit circle whose tangent is parallel to B. So we let
x = Bl = [.1736 .98481er (3.7)
The uncontrollable normal velocity component at t = 0 is
xTPx = ^xt[PA(0) + .4r(0)P]x = -.9548
(3.8)


102
a) Top View of Missile (-) and Target () Trajectory
Figure 8.10. (a) Top view and (b) side view of missile and target trajectories of a
scenario which occurred at 35,000ft. The target was missed 767 ft. Initial mach
numbers of the target and missile are 1.9 and .92 respectively.


73
Table 5.4. Extreme values and range of the pitch channel control terms as individual
flight parameters vary. Gains depend mostly on M, Q, and a.
M
Q
a
0
P
<7
r
control
min
min
min
min
min
min
min
term
max
max
max
max
max
max
max
diff
diff
diff
diff
diff
diff
diff
kqnao
-2.15
-3.1
-2.1
-1.60
-1.61
-1.61
-1.60
-.72
-.16
-1.4
-1.59
-1.59
-1.59
-1.59
1.43
2.95
.69
.017
.017
.017
.017
kquQo
-.1509
-.21
-.16
-.118
-.118
-.118
-.118
-.0529
-.031
-.10
-.117
-.117
-.117
-.117
.099
.179
.054
.0005
.0005
.0005
.0005
kref qll^cO
-2.53
-3.5
-2.7
-1.98
-1.98
-1.98
-1.98
-.916
-.53
-1.7
-1.97
-1.96
-1.96
-1.96
1.61
3.0
.94
.017
.017
.017
.017
and look at kqncio. We set q0 and equal to 10/s and 8 respectively so that
we can look at kqi2q0 and kref 9nac0. The sum of these three terms are fed into to
the elevator. For the terms which are fed into the remaining inputs, the following
assignments are made.
/% = &o = 2
Po = Pco = l00/s (5.52)
r0 = 25/s
Figures 5.2a-g show kqUa0 plotted against all seven flight parameters. These seven
figures show that kqnflo changes with M, Q, and a but remains nearly constant when
0, Pi q, and r change. Similar figures exist for the remaining eleven gains and are
summarized in Tables 5.4 and 5.5. The minimum and maximum values of each term
is listed for each changing flight parameter along with the difference between the
minimum and maximum values. Angles are expressed in radians. From this table it
was determined that all gains will be scheduled against M, Q, and a.


22
always a control, u, which can place x inside the ellipsoid. This is true because the
velocity field can always be controlled in the B direction. This is not true, however,
for points on the ellipsoid whose normal is parallel to span{B\_) as shown in figure
2.7. These points on the ellipsoid can be found by computing
P_1 span(B _l) (2.29)
where B_ is a basis of column vectors orthogonal to span(B). We now give the proof
of Theorem 1.
Proof:
Case 1: x S(t)
Substituting 2.20 into xTPx, we have
xTPx = xTPA()x + xTPfi(i)u (2.30)
Since xe5 then
x = p-'liB^t)) (2.31)
where l(Pj_(£)) is any linear combination of Bj_(t). So
xTfifi() u = l(P1(i))rp-1PP(i)u (2.32)
= \(B(t))TB(t)u (2.33)
= 0 (2.34)
and therefore from 2.30 we have,
xTPi = xTPA(i)x (2.35)


72
Because (5.45) is true for any v, then
I = C(A BK)~lBKref
(5.46)
Controllability of the system implies that C(A BK) 1B is invertible. Solving (5.46)
for Kref gives
Kref = ~[C(A BK)-lB]~l (5.47)
The EM R A AT missile has three inputs and therefore only three states can be
tracked. Controlling a in the pitch model and ¡3 and p in the roll yaw model is
desirable. For the pitch model y = a, implying that
Cg = [ 1 0 ] (5.48)
and, therefore,
Krefq = ~[Cq(Aq BqKq)-1 Bq]~l (5.49)
For the roll-yaw model
and, thus,
1 0 0
0 1 0
B-cfpr [Cr)rf Apr Bprhpr) B
1-1
pr J
(5.50)
(5.51)
Krefq and Krefpr are computed for each flight condition and then compared along
with Kq and Kpr to the flight conditions.
5.6 Comparing Gains with Flight Parameters
A series of gains have been generated as a function of different flight conditions.
Each flight parameter has been swept through a range of points while the remaining
six have been held constant. In order to compare different gains on the same input
it has been decided to use the products of gains and their corresponding terms at
typical values. For example a typical value of a is 8. So we set ao equal to 8


Table A.22. Tabular data for CUr
M
cnr
.8
-1857.534
1.2
-1854.110
2.0
-1182.192
2.7
-937.671
3.5
-849.315
Table A.23. Tabular data for C\p
M
Cip
.8
-91.499
1.2
-114.634
2.0
-67.057
2.7
-47.486
3.5
-35.671


31
Example 1
Given the system
where
and
let
x = Ax + B u
(2.80)
A =
B =
-10 12
0 -1
cos(lb)
sin(75)
P = I
We want to find all feedback gains which satisfy the following constraint.
c < x1 P < c Vx:x1x = l
(2.81)
Before specifying c and c, we must check the normal velocity components
for x E xi = Bl =
sin 75
cos 75
Then
xf Axi = 12.40
(2.82)
(2.83)
From Theorem 2 we must have
c < -12.40 < c
(2.84)
From this we choose
c = -14
c = 9
(2.85)
(2.86)


36
and
A := mm ^xT[P(,4 BKt) + (A BI Then let
c, = max[cj,\i] (2.112)
and
c, = min[cf, A,] (2.113)
Kq is the initial guess in the search for Kf Kf. c0 and Cq are computed using
(2.112) and (2.113) so that Kq is a member of dJCo and Kf C Kq. Then a new
feedback matrix, K\ is found which lies inside of Kq, but not on the boundary. New
constraining values are found in the same way as before so that the boundary of the
next feedback set contains K\. K% is then found so that it lies inside of the present
feedback set, but not on its boundary. This process is continued until Kl = Kf Kf.
The success of finding Kf depends on the following conditions.
1. Given that Ki dKi we must be able to find fG+i such that K{+1 G K{.
2. We must show that c,+i < c when c > c/ and ci+l > c when c{ < Cf.
3. We must show that Kf C Ki+1 C Ki.
4. Kf must be nonempty.
We now address these four points.
1. Given Ki dKi, we need to find a second feedback matrix K2 0K{ where
K2 Ki. Then, due to convexity, \Ki + \Ka is a member of Kt. Figure 2.10a shows
a second order example of a procedure for finding Ki+i. The algorithm will be given
shortly. The horizontal and vertical axis are assigned to kn and 12 respectively. The
region Ki is enclosed by 'dK; and dKi- Ki is known. Ki2 is found by searching along


89
Figure 8.1. A Block Diagram of the Nonlinear Missile Simulation.


39
that c, < ci+1 when ct < Cj.
3. We now show that ICf C JCi+1 C /C. Since c = max[c/, A], then c,- > c/. We have
already seen that ct > A+i. So c > max[c/, A+1] = C{+\. Similarly c < c+1. XA+1 is
the set of all K so that
ci+1 < ^xT[P(A BK) + (A M)rP]x < Ci+i V x : xTx = 1 (2.117)
Since Ci > Ci+i and c < c+1, then for every element of the following holds.
c,- < ixT[P(^ PA") + (A PA")tP]x < c V x : xTx = 1 (2.118)
Therefore, JCi+1 C /C. Since c+i > Cj and ci+1 < Cf. then using a similar argument
ICf C ICi+1.
4. In using this design procedure, P is chosen so that the maximum uncontrol
lable normal velocity component is negative. Then from Theorem 2, c/ can be made
negative in an attempt to achieve stability. c¡ must be greater than the maximum
uncontrollable normal velocity component, and c¡ must be less than the minimum
uncontrollable normal velocity component. From Theorem 2 this will guarantee the
nonemptiness of ICf and ICf. The nonemptiness of the intersection of these two sets,
however, is unknown. If ICf is nonempty, then, as i becomes large, A", £ ICf. If ICf
is empty then cl and c will converge to values which do not match the desired con
straints and K{ will yield a closed loop system that meets the constraints given by c
and ct. The designer will either have to accept this result or try again with a different
P or different constraining values or both. Since stability is desired, one approach
would be to keep P and c¡ and lower Cf until ICf becomes large enough to intersect 1C/.
The outline of the iterative Lyapunov design method is as follows.


32
In this example n = 2 and m = 1. It was decided to set i = j 1 so
that
Qi,j ~ Qi,i i
0 '
(2.87)
A0t\j A 01,1 = 0
a2
(2.88)
and
A kQip + Aqi,i
k K2
(2.89)
From equations (2.76) and (2.77) come
Ac ~[PQi,i + Qi,iBT]
(2.90)
Pie = -[BQhi + Qi'1BT]
(2.91)
Poc = A + AT 2c/ BKoi'i KqXIBt (2.92)
Po-c = A + AT 2c/ BKoi'i KqI XBt (2.93)
The roots of the following polynomials are computed in terms of k while
incrementing K2 through a wide range of values.
det[hPic + Poe] = 0 (2.94)
det[kPic T Toe] = 0 (2.95)
Rejecting complex roots and checking the regions separated by the real
roots give
kc(K2), kc(K2) (2.96)
kc(K2), k-c{K2) (2.97)
The intersection of these regions are found.
k(A2) maxfk^kc]
k(A'2) = mznfkc, kc]
(2.98)
(2.99)


CHAPTER 2
THEORY BEHIND THE DESIGN METHOD
The material in this section gives the theory leading to the proposed controller
design method. The first section presents a geometric interpretation of Lyapunovs
linear stability theorem. The effects of control on the velocity field of a system is
studied in the next section. The nonemptiness and convexity of the set of all feedback
gains which stabilize a system with respect to a given Lyapunov function are then
discussed. Finally, an iterative procedure that finds one element of this set is given.
2.1 A Geometric Interpretation of Lyapunovs Linear Stability Theorem
To understand why time invariance is a necessary assumption for eigenstructure
design methods the following second order linear time varying system was studied.
The example given now is from Vidyasagars example 5.3,109 [10] and can also be
found in Khalil [13]. Given the following system
x = A(t)x
(2.1)
where
m
l + acos2(t) 1 a sin(i) cos(i)
1 a sin() cos(t) l + asin2(i)
Vidyasagar [10] notes that the transition matrix is given by
*(*,0)
e( 1) cos(f) e 4 sin(i)
_eD-i) sin() e-cos()
and the characteristic equation is
(2.2)
(2.3)
A2 + (2 a)A + (2 a) 0 (2-4)
The roots of (2.4) have negative real parts for 1 < a < 2. The exponents in the first
column of 4>(i, 0) indicate, however, that the system is unstable for these values of a.
11


66
choice of jP, = Jdiag is the best choice for a positive definite function for the system
z = J z.
V(z) = ~zTJdiagz (5.24)
Making the following transformation into the x coordinate system gives
z = S x
(5.25)
which implies
V(x) = -xT[S~1]T JdiagS'1* (5.26)
Our choice of P is
P = -[S'1]7, Jd.za.gS~1. (5.27)
For the nominal flight condition, the Jordan canonical form of Aq BqKq and
Apr BpTKpr is found and from (5.27)
P =
26772 609
609 14.9
P
1 pr
6792 10.7
10.7 4.02
-316.4 -.509
-316.4 '
-.509
15.7
(5.28)
q KqBjPq) at the nominal point are
-40.002
(5.29)
-1.0715 x 106,
Likewise, for the closed loop roll-yaw system the eigenvalues are
-1.3614 x 105, -320.00, -20.003 (5.30)
5.4 The Iterative Lyapunov Design Method
The iterative Lyapunov design method generates feedback gains so that xTP,x
and xTPprx are Lyapunov functions for each closed loop system. The algorithm
requires the initial guesses Kqo and Kpr0, for the first point, and the positive definite
matrices Pq and PpT. As a given flight condition changes, the feedback gains from


58
where x contains the states and u contains the inputs. A and B are matrices which
are functions of several time varying flight parameters and are computed as follows.
A = l
(x,w )
df
(4.46)
.ominal
(4.4?)
(^fW ) nominal
where w contains additional flight parameters. Note that x and u are now pertur
bations from the point around which the linearization is taken.
We linearize the pitch model first. From inspection of (4.38) we see that
aq\l 1 1 +
QSCNa
mV
-l
-QS
C
ail2=M+^r(1-
mV
QS
Na
bqll ( 1 +
mV
Qsc^y1 f-QS
7}CNq
mV
mV
mV
Cn6,
To linearize (4.39), we must first substitute (4.38) in for . Then we differentiate as
in (4.46) and (4.47).
aq21
aq22
5921 =
QSd
Iyy
QSd
Iyy
QSd
Iyy
Cma + Cm ( 1 +
QSCN
-i
Cm<¡ + Cm* i 1 +
mV
QSCN -1
mV
-QS
C
1 -
mV
QS
Na
mVNq
Cm6q + Cm f 1 +
QSCN
mV
-i
-QSr
lvCN
The resulting linearized pitch model is
Ct
qll
Uql2
a
+
bqll
kq21
9 .
aq21
aq22
. q.
(4.48)
The same procedure is applied to the roll-yaw model. Assuming that a is constant
in (4.40), and from inspection of equations (4.40) to (4.42) the following results.
ttprll
QSCy
mV
}i aprl2 a +
QSCYp
mV
i aprl3 1 +
QSCYr
mV '


100
a) Top View of Missile (-) and Target (-) Trajectory
Figure 8.8. (a) Top view and (b) side view of missile and target trajectories of a
scenario which occurred at 50,000 ft. The missile missed the target by 452 ft. Initial
mach numbers of the target and missile are 3.0 and .92 respectively.


71
The eigenvalues in (5.30) suggest the following.
Qprl
<
-1.3614 x 105
<
Cprl
pr2
<
-320.00
<
Cpr2
pr 3
<
-20.003
<
Cpr3
(5.38)
The following constraining values have been chosen for the roll-yaw model.
cprl = -150000, cpTl = -110000
cpr2 = -340, cpt2 = -300 (5.39)
Q-ptz 22, Cpr3 17
5.5 Formulation of a State Tracker
The autopilot of the EMRAAT missile will be a state tracker. That is, we want to
be able to change the location of the equilibrium point in order to control the values
of some of the states. The following shows how this will be accomplished.
Given the linear system
x Ax + Bu
(5.40)
y = Cx,
(5.41)
we would like to find a control law
u = Kx + ATe/V
(5.42)
so that y, the output, tracks v ,the reference input, asymptotically. We require y = v
when x = 0. When x = 0, then
0 = Ax BKx + BKTefV.
(5.43)
Since K is chosen so that the system is stable, then A BK is invertible and
x = -(A- BK)-lBKreSv
(5.44)
Also,
V = y = Cx = C(A BK)~lBKreiv
(5.45)


CHAPTER 1
INTRODUCTION
The original goal of this dissertation was to formulate an autopilot design method
which stabilizes the flight of a bank-to-turn air-to-air missile. The result was the de
velopment of a controller design method which applies to time varying linear systems.
When used on a nonlinear system, local stability is achieved. The remainder of this
chapter gives a brief description of the missile used in this study, a summary of related
works preceding this study, and the purpose and outline of this dissertation.
An air-to-air missile is launched from a military aircraft with the intent of in
tercepting an enemy aircraft or target. Bank-to-turn (BTT) missiles have wings
giving them greater maneuverability than the conventional skid-to-turn (STT) mis
siles which have fins. Thus, the target must work much harder to evade a BTT
missile. The dynamics of BTT airframes are highly unstable, making for a difficult
controls problem.
The missile under study in this paper is the extended medium range air-to-air
technology (EMRAAT) missile as shown in Figure 1.1 [1], The EMRAAT missile
is a theoretical missile formulated by the Air Force with the intent of studying the
feasibility of the BTT concept. The knowledge gained from the study of this missile
will be used by the Air Force if it ever decides to make a real BTT missile.
The EMRAAT missile is equipped with a seeker. If the seeker is infrared based,
then it measures the line of sight angles to the target. In the case that the seeker is
radar based, then, in addition to the line of sight angles, the range and range rate
of the target are measured. This information is passed to the guidance law which
determines the desired normal accelerations needed to intercept the target. The
1


101
a) Commanded z Acceleration (-) and Actual z Acceleration ()
Figure 8.9. Commanded and actual (a) accelerations in the 2 direction, (b) com
manded and actual angles of attack, and (c) the elevator deflection angle, for the
scenario shown in figure 8.8. Saturation of the elevator angle was the cause of the
miss.


24
by a discussion of its convexity. Finally this section will present an exhaustive search
method for finding the boundary of this set.
For the remainder of this chapter, A and B will be frozen at one instant in
time. A discussion follows on how to find a controller or set of controllers which
satisfies (2.19) at one given instant. If a given system is time varying then the results
which follow must be applied for all time with P being constant and positive definite.
These results apply to all systems which can be stabilized with respect to a Lyapunov
function using a constant P. The success of these methods depends on the existence
of a constant positive definite P so that
+ p-xAT(t)]BL{t) < 0 Vi (2.41)
Since we require P to be zero, these results are conservative.
Let 1C and 1C be the set of all K which satisfy the left and right inequalities
respectively of the following expression.
c < ~xt[PA + AtP PBK KTBTP]x < c V x : xTx = 1 (2.42)
j
where P is positive definite. The objective is to find
}C:=lCr\X (2.43)
The following theorem gives conditions for the nonemptiness of 1C and 1C
Theorem 2 Given the nth order m-input system,
x = Ax + Bn (2.44)
1C and K are nonempty if and only if there exists a positive definite P so
that the following conditions are true:
c < minxTPAx = min-^xT[.PA + ATP]x
xes xgs 2
(2.45)


29
contain all the fixed entries of K and be defined as follows.
^11 k ij k\ n
Then
where
LOi,j
kn 0 k{.
kmi kmj kmn
K = k Qitj + if.
Qi,j
m
o i,j
1 j n
0 0 0
0
0 1 0
0
0 0 0
(2.71)
(2.72)
(2.73)
2) Find all values of the free entry which make
det[PA + 1 P 2cl PBK I This problem reduces to finding the roots of a polynomial. After sub
stituting equation (2.72) into the argument of the above determinant, it
becomes
k(-PBQitj QJjBtP) + (PA + AtP PBKoj K^BTP) (2.75)
Let
p¡(i,j) := -(PBQ+ Ql,BTP) (2.76)
P0(Ko¡¡) = PA + AtP PBKoi.i Kju BrP (2.77)


70
Table 5.2. Uncontrollable normal velocity components for the pitch model
changing parameter
minimum uncontrollable velocity
maximum uncontrollable velocity
M
-43.1456
-41.7253
Q
-45.0144
-42.2289
a
-43.3856
-42.2362
P
-42.6047
-42.6047
P
-42.6047
-42.6045
-42.6047
-42.6047
r
-42.6047
-42.6044
Table 5.3. Uncontrollable normal velocity components for the roll-yaw model
changing parameter
minimum uncontrollable velocity
maximum uncontrollable velocity
M
-21.3937
-20.9751
Q
-22.3016
-21.1574
a
-21.7136
-21.1140
0
-21.3180
-21.3180
P
-21.3181
-21.3180
q
-21.3184
-21.3180
r
-21.3180
-21.3180
For the pitch model, the uncontrollable normal velocity components range from
45.0144 to 41.7253. Theorem 2 requires that
-41.7253 <
c < -45.0144
(5.34)
In addition, from (5.29), we want
Cql < -1.0715 X 106 < Cqi
Cq2
C9 < -40.002 < Cq2
(5.35)
From this the constraining values for the pitch model have been chosen to be
c?1 = -1.2 x 106, qcq = -1 x 106,
(5.36)
Cq2 45, Cq2 35
Likewise for the roll-yaw model, when looking at Table 5.3, Theorem 2 requires that
-20.9751 <
Cpr3
S-prl
-22.3016
(5.37)


Table A. 12. Tabular data for C,
a/M
.8
1.2
2.0
2.7
3.5
-8
.22168
.10270
-.05374
-.02637
-.02044
-4
.25664
.16052
-.02687
.01319
.00365
0
.31726
.22611
.00330
.05275
.03285
2
.33186
.24984
.02379
.06300
.04380
4
.33496
.26882
.03877
.07692
.05547
8
.32743
.25070
.07048
.10549
.07372
12
.34115
.19935
.08899
.13077-
.09270
16
.36991
.20453
.09956
.12161
.12190
o
O
.35973
.16915
.13084
.15824
.15547
Table A.13. Tabular data for CySp
a/M
.8
1.2
2.0
2.7
3.5
-8
-.033468
-.022174
.001062
.002825
-.004925
-4
-.040358
-.030174
-.002743
-.002082
-.000373
0
-.049933
-.035826
-.006726
-.008327
-.009403
2
-.051007
-.036957
-.007965
-.009814
-.010896
4
-.050828
-.037217
-.009204
-.010409
-.011045
8
-.049664
-.032957
-.010929
-.011004
-.010000
12
-.050380
-.025130
-.011770
-.011004
-.007313
16
-.055302
-.023391
-.010265
-.009665
-.009701
20
-.037673
-.014957
-.014159
-.015762
-.015821


b
T T r *
T"^ T j
7 T t ,
* A j
* A i i
T
* i i
T T t
^ i
r t t .
r t [
7
r T. t L
^ / y
.151 UX-a: L 1 1
15 -10 -5 0 S 10 15
XI
Figure 2.2. The trajectory and velocity field of (2.1)


98
a) Top View of Missile (-) and Target () Trajectory
Figure 8.7. (a) Top view and (b) side view of missile and target trajectories of
a scenario which occurred at 10,000ft. The miss distance is 2.8 ft. Initial mach
numbers of the target and missile are 2.0 and .92 respectively.


94
a) Top View of Missile (-) and Target () Trajectory
Figure 8.3. (a) Top view and (b) side view of missile and target trajectories of a
scenario which occurred at 20,000 ft. The miss distance is 0.64 ft. Initial mach
numbers of the target and missile are 2.5 and .92 respectively.


85
Figure 7.2. A test of the curve fitting routines used to implement the autopilot.


61
Dependent
Flight
Parameters
Figure 5.1. A block diagram of the systemy used to determine the dependence of
gains on flight parameters.


18
Since
xT Px = xT PAx = ^x.T[AT P + PA\x (2-15)
Stability Condition 2 is equivalent to Lyapunovs linear stability condition.
It would be useful to compute the lower and upper bound of the rate of decay of
a given positive definite function,
V(x) = xT Px
(2.16)
of the states of a time varying linear system by evaluating,
c := mm
t
c := max
t
Xmin^ \P A(t) + AT(t)P]
XmaX\[PA(t) + AT(t)P]
So,
c < ixT[PA() + AT(t)P]x < c V x : xTx =1, Vi
(2.17)
(2.18)
(2.19)
Then c and c are the lower and upper bounds respectively of the normal velocity
components, xTPx, on the unit spheroid. If c < 0, then the system is stable.
2.2 Adding Control to the System
The result in section 2.1 is an analytical tool only. A design procedure is needed
to control stability for the system
x = A(t)x + B(t)u (2.20)
where u is the control vector and x is the state vector. This section focuses on two
questions:
1) What effect does u have on the velocity field of (2.20)?
2) Can the normal velocity component xTPx on the spheroid xTx = 1
be controlled?


Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
A CONTROLLER DESIGN METHOD WHICH APPLIES TO TIME VARYING
LINEAR SYSTEMS
By
Kurt Walter Koenig
August 1994
Chairman: Dr. Thomas E. Bullock
Major Department: Electrical Engineering
A feedback controller design method has been formulated which applies to linear
time varying systems. The motivation behind this technique is the design of an
autopilot for an air-to-air missile. The missile under study is the extended medium
' range air-to-air technology (EMRAAT) missile, a theoretical bank-to-turn missile
which is under study by the United States Air Force. Knowledge gained by this
missile will be applied to future bank-to-turn missiles.
Conventional autopilot design techniques use pole placement, linear quadratic
regulators, linear quadratic Gaussian/loop transfer recovery techniques, or eigen-
structure assignment methods. These techniques are applied to a linear model which
depends on time varying flight parameters. In applying these methods, the false as
sumption is made that the linear model is slowly varying. Since the system is changing
rapidly, no theoretical basis exists to support the success of the resulting controller.
The proposed design method finds state feedback gains which cause the closed loop
linearized system to be stable with respect to a specified Lyapunov function. Even
though flight parameters change rapidly, local stability around the operating point is
achieved.
IV


80
6.3 Generating the Look-Up Table
With the design constraints set, feedback gains and feedforward gains are gen
erated for each grid point. Figure 6.2 gives a block diagram of the system used to
accomplish this. For the pitch model, the initial guess comes from one of the gains
that was generated in the previous chapter when determining the dependence be
tween gains and flight parameters. The operating point from which this initial guess
originates is
M = 1, Q = 1250psf, a = 8 (6.6)
and is one of the extreme values listed in Table 5.1. The result of the first point is
used to start adjacent points which, in turn, start new adjacent points until gains
have been computed for the entire grid. The look-up table for the roll-yaw model is
made in the same way.
Numerical problems were encountered in parts of the look-up table for the roll-
yaw model. They were similar to the problems that were predicted in step 5) of the
iterative design method given in Chapter 2. To overcome these difficulties, some of the
constraining values were relaxed for a number of iterations and were later returned to
their original assignments in the algorithm. Eventually, the desired feedback matrix
was found.
Figure 6.3 shows kqu verses mach number and dynamic pressure when a = 8. As
this figure would indicate, the gains generated from the iterative Lyapunov method
are smooth with respect to the dependent flight parameters. This fact gives hope
that the gain scheduling scheme will be easy.


CHAPTER 8
NONLINEAR SIMULATIONS
A nonlinear simulation has been used to test the proposed autopilot for the EM-
RAAT missile. First a section follows giving an overview of the nonlinear simulation.
A test module is then made to generate state commands in order to evaluate the
autopilots tracking ability. Finally, a series of flight scenarios are run to determine
the ability the missile has to intercept the target.
8.1 The Nonlinear Simulation
Figure 8.1 shows a block diagram of the simulation used to test the EMRAAT
missile. The program is written in FORTRAN. Initial conditions of the target and
missile are specified by the user. The simulation is then run and a trajectory of both
the target and missile results. All target and missile variables can be observed. The
target is programmed to fly in a straight line until the range between the target and
missile falls below 5,000 ft. The target then makes a 9 g turn to the right. The
simulation terminates when the closing velocity becomes positive.
The seeker measures the line of sight angles and the range rate of the target.
The simulation uses exact measurements of these values and does not assume any
noise. These values are sent to the guidance law which, in this case, implements
proportional navigation. A derivation of this guidance law can be found in Bryson
and Ho [19]. The outputs of the guidance law are two desired accelerations, aVc and
aZc. The BTT logic makes the conversion from the acceleration commands to the
three state commands ctc, /3C, and pc. Since the missile can achieve a much higher
acceleration with angle of attack than with sideslip, the BTT logic uses pc to rotate
the desired accelerations into the pitch plane. If this roll maneuver is successful then
88


Table A.4. Tabular data for C,
'<7
a/M
.8
1.2
2.0
2.7
3.5
-8
-1.035
-1.278
-.5951
-.5269
-.3970
-4
-1.200
-1.44
-.6862
-.5490
-.4377
0
-1.320
-1.4719
-.7089
-.5297
-.4151
2
-1.330
-1.4875
-.7008
-.5269
-.3925
4
-1.330
-1.456
-.6862
-.5297
-.3849
8
-1.330
-1.325
-.6276
-.4772
-.3396
12
-1.330
-1.216
-.5626
-.4200
-.3245
16
-1.286
-1.114
-.5100
-.3641
-.3547
o
O
-1.232
-1.0594
-.4715
-.3724
-.4226
Table A.5. Tabular data for Cm/3
a/M
.8
1.2
2.0
2.7
3.5
-8
1.0702
.8632
.2175
-.0923
-.3292
-4
1.2877
1.2182
.5090
.1754
-.0234
0
1.3544
1.3825
.6372
.3138
.1846
2
1.3544
1.4414
.6500
.3446
.2338
4
1.3509
1.4772
.6372
.3723
.2677
8
1.2561
1.5172
.491
.3200
.2862
12
.9502
1.4133
.2385
.2708
.3169
16
.6989
1.1228
.3808
.4338
.5046
o
O
.5123
.5179
.6179
.6554
.7446


62
where Vsoa is the speed of sound. Both p and Vsos are functions of altitude.
P = fAh)
(5.3)
Vsos = fs(h) (5.4)
where h is altitude in feet above sea level. Here, fp and fs are functions based on
atmospheric tables and are implemented by linear interpolation. Solving (5.2) for V
and substituting the result into (5.1) gives
Q = (5.5)
We generate a third table in the following way.
Mh) := fP(h)f*(h) = pVs2os (5.6)
So
Q = (5.7)
The function, /3, is a one-to-one function so that its inverse can be found by reading
the table backwards. With this in mind we can solve (5.7) for h.
From (5.2) and (5.4)
V = MVS0S = Mfs(h)
Substituting (5.8) into (5.9) to eliminate h gives
Also from (5.3)
(5.8)
(5.9)
(5.10)
(5.11)


I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
Scott L. Miller
Associate Professor of
Electrical Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
David C. Wilson
Professor of Mathematics
This dissertation was submitted to the Graduate Faculty of the College of
Engineering and to the Graduate School and was accepted as partial fulfillment
of the requirements for the degree of Doctor of Philosophy.
August 1994
^ ^>
Winfred M. Phillips
Dean, College of Engineering
Karen A. Holbrook
Dean, Graduate School


10
eigenvalues can be placed exist. Also, while many feedback control laws make (1.23)
negative definite, Shahruz and Behtash give only a small subset of these control laws.
1.2 Purpose
This paper gives an algorithm which can stabilize a linear time varying system by
placing the eigenvalues of (1.23) between desired bounds and has fewer limitations
than the control law in Shahruz and Behtash [12]. Limitations on the placement of
these bounds are stated. This algorithm applies to all systems for which a constant
positive definite P exists so that (1-23) can be made negative definite for all time.
First, theory is presented leading to the formulation of this algorithm. Then, to
demonstrate the usefulness of this algorithm, it is applied to a linear time varying
system where normal pole placement techniques fail. This algorithm is then applied
to the EMRAAT missile. A nonlinear model is made, and from this, a time vary
ing linearized model is generated which is a function of several flight parameters.
Gains are computed, and their dependence on flight parameters determined. A gain
scheduling scheme is then implemented yielding local stability around the operating
point. The resulting design is tested in a nonlinear simulation. Finally, the results of
this test are given, and the usefulness of this design technique is evaluated.


21
system is stabilizable with an appropriate choice of constant positive definite P. If
there exists a constant positive definite P so that
max max xTfid(i)x < 0
xes(t)
(2.24)
the system is stabilizable. The above condition is equivalent to requiring the following
matrix to be negative definite for all time.
1
(2.25)
This fact was established by Fields [15] and is true for the following reasons. Condi
tion (2.24) is true if and only if
<0V/i^0, Vi
(2.26)
Since the denominator of the above fraction is positive, with some simplification the
above statement is equivalent to
Bl{t)[A{t)P~l + p-1Ar(i)]JB1(i)/x <0V^0, Vi (2.27)
which is equivalent to
\Bl{t)[A{t)P~r + p-1AT(t)]BL(t) < 0 Vi (2.28)
If a constant positive definite P exists satisfying condition (2.28), then the given time
varying system can be stabilized. This result is more general than a similar one given
by Shahruz [12], and the two are equivalent when P = I.
Before giving a proof of Theorem 1, we give an intuitive explanation of Theorem
1 for the second order case with a single input. Figure 2.7 shows the B vector for
such a system along with some ellipsoid xTPx =1. If a state vector, x, is drawn
to some point on this ellipse whose normal is not perpendicular to B. then there is


33
Figure 2.8. A plot of the boundary of 1C which guarantees satisfaction of the design
constraints of Example 1.
Finally,
dlC = {[k{K2), K2] V K2] U {[k(K2), I<2] V I<2} (2.100)
A plot of dK, is shown in Figure 2.8. Let K = [7 11]. We can check to see
if K C 1C by evaluating
Ai = \mtn(A + AT BK KtBt) (2.101)
= -12.91 (2.102)
A2 = XmaX(A + At-BK- KtBt) (2.103)
= -10.52 (2.104)
As the following shows,
c < Ai (2.105)
A2 < c (2.106)


78
pr
Figure 6.1. Formulation of the Design Constraints.
Table 6.2. Uncontrollable normal velocity components for the pitch model.
a(degrees)
minimum uncontrollable velocity
maximum uncontrollable velocity
-3
-47.1539
-37.4260
1
-47.0806
-37.3873
4
-47.1611
-37.3480
8
-47.9797
-37.4247
12
-48.8796
-37.3906
16
-49.5915
-37.6144
20
-49.5985
-37.4355
Table 6.3. Uncontrollable normal velocity components for the roll-yaw model.
a(degrees)
minimum uncontrollable velocity
maximum uncontrollable velocity
-1
-23.3399
-19.6307
1
-23.3774
-19.6409
2.5
-23.4445
-19.6883
4
-23.4971
-19.7450
8
-24.5852
-17.2696
12
-25.3163
-19.9600
16
-25.6480
-19.8625
20
-26.4797
-20.3071


3
autopilot attempts to achieve these accelerations by applying the proper control to
the control surfaces.
1.1 Earlier Works
Early autopilot designs separated the missile dynamics into two or three lower
order models. These models were linearized, and classical techniques were used to
stabilize them. Other designs applied more advanced techniques to higher order lin
earized models. Some of these techniques are pole placement, linear quadratic regula
tors, linear quadratic Gaussian/loop transfer recovery techniques and eigenstructure
assignment methods [2] Feedback controllers would be designed as a function of the
linearized models. Gains would then be scheduled against the flight parameters on
which the models depend. Gain scheduling is a popular technique used in control
designs and has been studied in depth by Shamma, Athans, and Cloutier [3,4,5,6].
The problem with these design methods is that they assume that the system is
time invariant or slowly varying. The linearized models depend on rapidly changing
flight parameters. Although these methods often work, no theoretical basis exists
to support the success of these designs. More recently, H<*, and fi synthesis design
techniques have been used to formulate one dynamic controller which would stabilize
the system for all modeled uncertainties [7,8]. A disadvantage of Hoo controllers is
that they have very high orders. Also, the existence of a robust H^ controller which
satisfies the performance requirements is not guaranteed. The only way to determine
the stability of a given design is to test it using a nonlinear simulation.
Desoer [9] states if a given time varying system
x = A(t)x (1.1)
is stable for each t when t is frozen, then it is not necessarily stable when t is allowed
to change. However, (1.1) is stable if A(t) changes slowly. A conservative upper



PAGE 1

$ &21752//(5 '(6,*1 0(7+2' :+,&+ $33/,(6 72 7,0( 9$5<,1* /,1($5 6<67(06 %\ .857 :$/7(5 .2(1,* $ ',66(57$7,21 35(6(17(' 72 7+( *5$'8$7( 6&+22/ 2) 7+( 81,9(56,7< 2) )/25,'$ ,1 3$57,$/ )8/),//0(17 2) 7+( 5(48,5(0(176 )25 7+( '(*5(( 2) '2&725 2) 3+,/2623+< 81,9(56,7< 2) )/25,'$

PAGE 2

7$%/( 2) &217(176 $%675$&7 LLL &+$37(56 ,1752'8&7,21 (FLUOLHU :RUNV 3XUSRVH 7+(25< %(+,1' 7+( '(6,*1 0(7+2' $ *HRPHWULF ,QWHUSUHWDWLRQ RI /\DSXQRYfV /LQHDU 6WDELOLW\ 7KHRUHP $GGLQJ &RQWURO WR WKH 6\VWHP $ /LQHDU )HHGEDFN 6HW WR &RQWURO [73[ 2QH /LQHDU )HHGEDFN 0DWUL[ WR &RQWURO [73[ $ 7,0( 9$5<,1* 6(&21' 25'(5 (;$03/( '(5,9$7,21 2) $ 02'(/ 2) 7+( (05$$7 0,66,/( 7KH 1RQOLQHDU 0RGHO 7KH /LQHDU 0RGHO 7+( '(3(1'(1&( 2) *$,16 21 )/,*+7 3$5$0(7(56 *HQHUDWLQJ S DQG 9 IURP 0 DQG 4 7KH )OLJKW 3DUDPHWHU *HQHUDWRU ,QLWLDOL]LQJ WKH ,WHUDWLYH /\DSXQRY 'HVLJQ 0HWKRG 7KH ,WHUDWLYH /\DSXQRY 'HVLJQ 0HWKRG )RUPXODWLRQ RI D 6WDWH 7UDFNHU &RPSDULQJ *DLQV ZLWK )OLJKW 3DUDPHWHUV &20387,1* /22.83 7$%/(6 'HWHUPLQLQJ D *ULG RI 3RLQWV )RUPXODWLRQ RI KH 'HVLJQ &RQVWUDLQWV *HQHUDWLQJ WKH /RRN8S 7DEOH LL

PAGE 3

*$,1 6&+('8/,1* &XUYH )LWWLQJ 7HVWLQJ WKH )LW 121/,1($5 6,08/$7,216 7KH 1RQOLQHDU 6LPXODWLRQ $ 7HVW RI 6WDWH 7UDFNLQJ 6LPXODWLRQ RI )OLJKW 6FHQDULRV &21&/86,21 $33(1',&(6 $ $(52'<1$0,& '$7$ )25 7+( (05$$7 $,5)5$0( % ,1(57,$/ '$7$ )25 7+( (05$$7 $,5)5$0( 5()(5(1&(6 %,2*5$3+,&$/ 6.(7&+

PAGE 4

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n UDQJH DLUWRDLU WHFKQRORJ\ (05$$7f PLVVLOH D WKHRUHWLFDO EDQNWRWXUQ PLVVLOH ZKLFK LV XQGHU VWXG\ E\ WKH 8QLWHG 6WDWHV $LU )RUFH .QRZOHGJH JDLQHG E\ WKLV PLVVLOH ZLOO EH DSSOLHG WR IXWXUH EDQNWRWXUQ PLVVLOHV &RQYHQWLRQDO DXWRSLORW GHVLJQ WHFKQLTXHV XVH SROH SODFHPHQW OLQHDU TXDGUDWLF UHJXODWRUV OLQHDU TXDGUDWLF *DXVVLDQORRS WUDQVIHU UHFRYHU\ WHFKQLTXHV RU HLJHQ VWUXFWXUH DVVLJQPHQW PHWKRGV 7KHVH WHFKQLTXHV DUH DSSOLHG WR D OLQHDU PRGHO ZKLFK GHSHQGV RQ WLPH YDU\LQJ IOLJKW SDUDPHWHUV ,Q DSSO\LQJ WKHVH PHWKRGV WKH IDOVH DVn VXPSWLRQ LV PDGH WKDW WKH OLQHDU PRGHO LV VORZO\ YDU\LQJ 6LQFH WKH V\VWHP LV FKDQJLQJ UDSLGO\ QR WKHRUHWLFDO EDVLV H[LVWV WR VXSSRUW WKH VXFFHVV RI WKH UHVXOWLQJ FRQWUROOHU 7KH SURSRVHG GHVLJQ PHWKRG ILQGV VWDWH IHHGEDFN JDLQV ZKLFK FDXVH WKH FORVHG ORRS OLQHDUL]HG V\VWHP WR EH VWDEOH ZLWK UHVSHFW WR D VSHFLILHG /\DSXQRY IXQFWLRQ (YHQ WKRXJK IOLJKW SDUDPHWHUV FKDQJH UDSLGO\ ORFDO VWDELOLW\ DURXQG WKH RSHUDWLQJ SRLQW LV DFKLHYHG ,9

PAGE 5

7KH GHVLJQ DOJRULWKP ILQGV IHHGEDFN JDLQV ZKLFK SODFH WKH HLJHQYDOXHV RI WKH GHULYDWLYH RI D JLYHQ /\DSXQRY IXQFWLRQ /LPLWDWLRQV RQ SODFHPHQW RI WKHVH HLJHQYDOn XHV DUH VWDWHG 7KLV FRQFHSW LV DSSOLHG WR D VHFRQG RUGHU OLQHDU WLPH YDU\LQJ V\VWHP ZKHUH RUGLQDU\ SROH SODFHPHQW WHFKQLTXHV IDLO 7KH GHVLJQ PHWKRG LV WKHQ DSSOLHG WR D OLQHDUL]HG PRGHO RI WKH (05$$7 PLVVLOH ZKLFK LV D IXQFWLRQ RI WLPH YDU\LQJ IOLJKW SDUDPHWHUV )HHGEDFN JDLQV DUH JHQHUDWHG DV D IXQFWLRQ RI WKHVH IOLJKW SDUDPn HWHUV ,W LV GLVFRYHUHG WKDW WKH JDLQV GHSHQG RQ G\QDPLF SUHVVXUH PDFK QXPEHU DQG DQJOHRIDWWDFN $ FRPELQDWLRQ RI LQWHUSRODWLRQ DQG SRO\QRPLDO ILWWLQJ LV XVHG WR FUHDWH D ORRNXS WDEOH IRU WKH JDLQV 7KH UHVXOWLQJ FRQWUROOHU LV SURJUDPPHG LQWR D QRQOLQHDU VLPXODWLRQ ZKLFK UXQV PLVVLOH DQG WDUJHW VFHQDULRV 6PDOO PLVV GLVWDQFHV DUH DFKLHYHG Y

PAGE 6

&+$37(5 ,1752'8&7,21 7KH RULJLQDO JRDO RI WKLV GLVVHUWDWLRQ ZDV WR IRUPXODWH DQ DXWRSLORW GHVLJQ PHWKRG ZKLFK VWDELOL]HV WKH IOLJKW RI D EDQNWRWXUQ DLUWRDLU PLVVLOH 7KH UHVXOW ZDV WKH GHn YHORSPHQW RI D FRQWUROOHU GHVLJQ PHWKRG ZKLFK DSSOLHV WR WLPH YDU\LQJ OLQHDU V\VWHPV :KHQ XVHG RQ D QRQOLQHDU V\VWHP ORFDO VWDELOLW\ LV DFKLHYHG 7KH UHPDLQGHU RI WKLV FKDSWHU JLYHV D EULHI GHVFULSWLRQ RI WKH PLVVLOH XVHG LQ WKLV VWXG\ D VXPPDU\ RI UHODWHG ZRUNV SUHFHGLQJ WKLV VWXG\ DQG WKH SXUSRVH DQG RXWOLQH RI WKLV GLVVHUWDWLRQ $Q DLUWRDLU PLVVLOH LV ODXQFKHG IURP D PLOLWDU\ DLUFUDIW ZLWK WKH LQWHQW RI LQn WHUFHSWLQJ DQ HQHP\ DLUFUDIW RU WDUJHW %DQNWRWXUQ %77f PLVVLOHV KDYH ZLQJV JLYLQJ WKHP JUHDWHU PDQHXYHUDELOLW\ WKDQ WKH FRQYHQWLRQDO VNLGWRWXUQ 677f PLVn VLOHV ZKLFK KDYH ILQV 7KXV WKH WDUJHW PXVW ZRUN PXFK KDUGHU WR HYDGH D %77 PLVVLOH 7KH G\QDPLFV RI %77 DLUIUDPHV DUH KLJKO\ XQVWDEOH PDNLQJ IRU D GLIILFXOW FRQWUROV SUREOHP 7KH PLVVLOH XQGHU VWXG\ LQ WKLV SDSHU LV WKH H[WHQGHG PHGLXP UDQJH DLUWRDLU WHFKQRORJ\ (05$$7f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

PAGE 7

)LJXUH 7KH (05$$7 PLVVLOH >@

PAGE 8

DXWRSLORW DWWHPSWV WR DFKLHYH WKHVH DFFHOHUDWLRQV E\ DSSO\LQJ WKH SURSHU FRQWURO WR WKH FRQWURO VXUIDFHV (DUOLHU :RUNV (DUO\ DXWRSLORW GHVLJQV VHSDUDWHG WKH PLVVLOH G\QDPLFV LQWR WZR RU WKUHH ORZHU RUGHU PRGHOV 7KHVH PRGHOV ZHUH OLQHDUL]HG DQG FODVVLFDO WHFKQLTXHV ZHUH XVHG WR VWDELOL]H WKHP 2WKHU GHVLJQV DSSOLHG PRUH DGYDQFHG WHFKQLTXHV WR KLJKHU RUGHU OLQn HDUL]HG PRGHOV 6RPH RI WKHVH WHFKQLTXHV DUH SROH SODFHPHQW OLQHDU TXDGUDWLF UHJXODn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r DQG IL V\QWKHVLV GHVLJQ WHFKQLTXHV KDYH EHHQ XVHG WR IRUPXODWH RQH G\QDPLF FRQWUROOHU ZKLFK ZRXOG VWDELOL]H WKH V\VWHP IRU DOO PRGHOHG XQFHUWDLQWLHV >@ $ GLVDGYDQWDJH RI +RR FRQWUROOHUV LV WKDW WKH\ KDYH YHU\ KLJK RUGHUV $OVR WKH H[LVWHQFH RI D UREXVW +A FRQWUROOHU ZKLFK VDWLVILHV WKH SHUIRUPDQFH UHTXLUHPHQWV LV QRW JXDUDQWHHG 7KH RQO\ ZD\ WR GHWHUPLQH WKH VWDELOLW\ RI D JLYHQ GHVLJQ LV WR WHVW LW XVLQJ D QRQOLQHDU VLPXODWLRQ 'HVRHU >@ VWDWHV LI D JLYHQ WLPH YDU\LQJ V\VWHP [ $Wf[ f LV VWDEOH IRU HDFK W ZKHQ W LV IUR]HQ WKHQ LW LV QRW QHFHVVDULO\ VWDEOH ZKHQ W LV DOORZHG WR FKDQJH +RZHYHU f LV VWDEOH LI $Wf FKDQJHV VORZO\ $ FRQVHUYDWLYH XSSHU

PAGE 9

$OSKD YV 7LPH )LJXUH $QJOH RI DWWDFN IURP RQH WUDMHFWRU\ RI WKH (05$$7 PLVVLOH ERXQG RQ VXSW! __$Lf__ LV JLYHQ ZKLFK LI HQIRUFHG JXDUDQWHHV DV\PSWRWLF VWDELOLW\ RI f $V ZLOO QRZ EH VKRZQ WKLV OLPLW LV WRR UHVWULFWLYH IRU WKH (05$$7 PLVVLOH )LJXUH VKRZV WKH DQJOH RI DWWDFN RI WKH WUDMHFWRU\ RI D PLVVLOH IORZQ XVLQJ DQ DOUHDG\ H[LVWLQJ DXWRSLORW 'HVRHU SURSRVHV WKH /\DSXQRY IXQFWLRQ 9n[Lf [7eO 3Iff[ f ZKHUH HM DQG 3Wf LV FKRVHQ VR WKDW $7Wf3^Wf 3^Wf$^Wf 'HVRHU JLYHV D ERXQG RQ 9 9 [7[>f§ HLFLP ZKHUH WKH IROORZLQJ GHILQLWLRQV DUH PDGH $ P A D?I VXS __$Lf__ RR W! R f f f

PAGE 10

7R LV SRVLWLYH DQG 5H $>$Wf@ f§DU 9] 9L P LV D FRQVWDQW DQG GHSHQGV RQO\ RQ FU DQG DMA VR WKDW $OVR H[S7OLff__ PH[S f§FUUf 9W 9L K0 VXS >_ƒf__ ]!R ,I H[ LV DOORZHG WR EH YHU\ VPDOO WKHQ IURP f VWDELOLW\ ZRXOG UHVXOW LI FU DX 7 f f f f )RU WKH OLQHDU SLWFK PRGHO RI WKH JLYHQ WUDMHFWRU\ FU $W VHFRQGV LQWR WKH WUDMHFWRU\ WKH FORVHG ORRS PDWUL[ LV $ f $W WKLV LQVWDQW LQ WLPH P DQG DX f§ A )URP WKLV ZH VHH WKDW WKH OLQHDU V\VWHP LV FKDQJLQJ PXFK IDVWHU WKDQ WKH OLPLW VKRZQ LQ f 7KH V\VWHP LV QRW VORZO\ YDU\LQJ 1RWH WKDW P ZDV RQO\ IRXQG IRU L VHF DQG QRW IRU DOO WLPH ,I D JUHDWHU P ZHUH IRXQG IRU WKH UHVW RI WKH WUDMHFWRU\ WKHQ WKH OLPLW LQ ZRXOG EH HYHQ PRUH UHVWULFWLYH DQG WKH UHVXOW ZRXOG EH WKH VDPH 7KLV WHVW GRHV QRW VKRZ WKDW WKH V\VWHP LV XQVWDEOH ,Q IDFW D /\DSXQRY IXQFWLRQ LV NQRZQ ZKLFK VKRZV WKDW WKLV OLQHDU V\VWHP LV VWDEOH EXW 'HVRHUfV WKHRU\ FDQ QRW VXSSRUW WKLV IDFW :LOVRQ &ORXWLHU DQG @ JLYH D VWDELOLW\ FRQGLWLRQ IRU D FRQVWDQW OLQHDU V\VWHP ZLWK WLPH YDU\LQJ XQFHUWDLQWLHV *LYHQ D V\VWHP [ f§ >$ (Wf@[ f LI D SRVLWLYH GHILQLWH 3 FDQ EH IRXQG ZKLFK VROYHV WKH /\DSXQRY HTXDWLRQ 3$ $W3 f

PAGE 11

)LJXUH $QJOH RI DWWDFN IURP RQH WUDMHFWRU\ RI WKH (05$$7 PLVVLOH WKHQ WKH V\VWHP ZLOO EH DV\PSWRWLFDOO\ VWDEOH LI rPD[>(^Wf? Of§ 9L f tPD[?( f ZKHUH FUPD[ LV WKH PD[LPXP VLQJXODU YDOXH 7KLV LGHD LV LQLWLDOO\ LQWHUHVWLQJ EHFDXVH LW UHTXLUHV WKH HUURU (Lf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f

PAGE 12

6LJPD PD[>(Wf@ YV 7LPH )LJXUH $ SORW RI FUPD[>(^Wf? YHUVXV WLPH (Wf LV WKH UHVXOW RI VXEWUDFWLQJ WKLV PDWUL[ IURP WKH FORVHG ORRS V\VWHP IURP WKH UHVW RI WKH WUDMHFWRU\ 6ROYLQJ IRU 3 LQ f JLYHV 3 LLf &RPSXWLQJ WKH XSSHU ERXQG LQ f JLYHV f )LJXUH VKRZV WKH SORW RI FUPD[>(Wf? 2EYLRXVO\ VLQFH DPD[>(^Wf? LV JUHDWHU WKDQ IRU PRVW RI WKH WUDMHFWRU\ WKHQ f LV QRW VDWLVILHG 7KLV SURFHGXUH ZDV UHSHDWHG E\ WDNLQJ WKH FRQVWDQW FORVHG ORRS PDWUL[ IURP HYHU\ SRLQW LQ WKH WUDMHFWRU\ DQG WKH UHVXOW ZDV WKH VDPH 7KH WLPH YDU\LQJ OLQHDU V\VWHP IURP WKLV WUDMHFWRU\ LV NQRZQ WR EH VWDEOH EHFDXVH D /\DSXQRY IXQFWLRQ H[LVWV ZKLFK FDQ VKRZ WKLV 7KH XSSHU ERXQG LQ f FDQ QRW EH HDVLO\ XVHG LI DW DOO WR VXSSRUW WKH FODLP RI VWDELOLW\ 'HVRHU >@ JLYHV D VWDELOLW\ OLPLW RQ WKH UDWH RI YDULDWLRQ RI D JLYHQ FORVHG ORRS WLPH YDU\LQJ V\VWHP :LOVRQ &ORXWLHU DQG @ JLYH D VWDELOLW\ OLPLW RQ

PAGE 13

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f[ LI D SRVLWLYH GHILQLWH PDWUL[ 3 FDQ EH IRXQG VR WKDW WKH PDWUL[ f 3$Wf $7Wf3 f LV QHJDWLYH GHILQLWH IRU DOO WLPH WKHQ WKH V\VWHP LV DV\PSWRWLFDOO\ VWDEOH 7KH VHFRQG UHVXOW DOORZV WKH ILUVW UHVXOW WR EH DSSOLHG WR D QRQOLQHDU V\VWHP *LYHQ D QRQOLQHDU V\VWHP RI WKH IRUP [ L[f ZKHUH *2f R DQG LV FRQWLQXRXVO\ GLIIHUHQWLDEOH WKHQ OHW GIW[f P G[ [ DQG DVVXPH WKDW __L[f $Lf[__

PAGE 14

DQG $ LV ERXQGHG ,I LV DQ DV\PSWRWLFDOO\ VWDEOH HTXLOLEULXP SRLQW RI ] $Wf] IRU DOO WLPH WKHQ LV D ORFDOO\ VWDEOH HTXLOLEULXP SRLQW IRU WKH V\VWHP ; r[f 7KHVH LGHDV FDQ EH DSSOLHG LQ WKH IROORZLQJ ZD\ *LYHQ D QRQOLQHDU V\VWHP [ [ZXf GHILQH $[Zf GI G[ %^[Zf [ZIL GI GX f f [Zf ZKHUH [ LV D YHFWRU FRQWDLQLQJ WKH VWDWHV DQG Z LV D YHFWRU FRQWDLQLQJ DGGLWLRQDO V\VWHP SDUDPHWHUV $W UHJLRQV QHDU WKH RSHUDWLQJ SRLQW WKH V\VWHP EHFRPHV FORVH WR $[ $[ Zf$[ %[ Zf$X f ZKHUH $[ DQG $X DUH VPDOO SHUWXUEDWLRQV EHWZHHQ WKH VWDWHV DQG LQSXWV DQG WKH RSHUDWLQJ SRLQW :H ZRXOG OLNH WR ILQG D IHHGEDFN FRQWURO ODZ X f§ .[ Zf[ f VR WKDW 3>$[ Zf %[ Zf.[ Zf@ >$[ Zf f§ %[ Zf.[ Zf@U3 f LV QHJDWLYH GHILQLWH IRU DOO [ DQG Z ZKHUH 3 LV SRVLWLYH GHILQLWH ,I VXFK D FRQWURO ODZ LV IRXQG WKHQ WKH SHUWXUEDWLRQV IURP WKH QRPLQDO WUDMHFWRU\ DUH ORFDOO\ VWDEOH IRU WKH V\VWHP [ [Z7[Zf[f f 6KDKUX] DQG %HKWDVK >@ JLYH RQH FRQWURO ODZ ZKLFK SODFHV VRPH RI WKH HLJHQYDOXHV RI f IRU WKH FDVH ZKHUH 3 +RZHYHU XQQHFHVVDU\ OLPLWDWLRQV RQ ZKHUH WKH

PAGE 15

HLJHQYDOXHV FDQ EH SODFHG H[LVW $OVR ZKLOH PDQ\ IHHGEDFN FRQWURO ODZV PDNH f QHJDWLYH GHILQLWH 6KDKUX] DQG %HKWDVK JLYH RQO\ D VPDOO VXEVHW RI WKHVH FRQWURO ODZV 3XUSRVH 7KLV SDSHU JLYHV DQ DOJRULWKP ZKLFK FDQ VWDELOL]H D OLQHDU WLPH YDU\LQJ V\VWHP E\ SODFLQJ WKH HLJHQYDOXHV RI f EHWZHHQ GHVLUHG ERXQGV DQG KDV IHZHU OLPLWDWLRQV WKDQ WKH FRQWURO ODZ LQ 6KDKUX] DQG %HKWDVK >@ /LPLWDWLRQV RQ WKH SODFHPHQW RI WKHVH ERXQGV DUH VWDWHG 7KLV DOJRULWKP DSSOLHV WR DOO V\VWHPV IRU ZKLFK D FRQVWDQW SRVLWLYH GHILQLWH 3 H[LVWV VR WKDW f FDQ EH PDGH QHJDWLYH GHILQLWH IRU DOO WLPH )LUVW WKHRU\ LV SUHVHQWHG OHDGLQJ WR WKH IRUPXODWLRQ RI WKLV DOJRULWKP 7KHQ WR GHPRQVWUDWH WKH XVHIXOQHVV RI WKLV DOJRULWKP LW LV DSSOLHG WR D OLQHDU WLPH YDU\LQJ V\VWHP ZKHUH QRUPDO SROH SODFHPHQW WHFKQLTXHV IDLO 7KLV DOJRULWKP LV WKHQ DSSOLHG WR WKH (05$$7 PLVVLOH $ QRQOLQHDU PRGHO LV PDGH DQG IURP WKLV D WLPH YDU\n LQJ OLQHDUL]HG PRGHO LV JHQHUDWHG ZKLFK LV D IXQFWLRQ RI VHYHUDO IOLJKW SDUDPHWHUV *DLQV DUH FRPSXWHG DQG WKHLU GHSHQGHQFH RQ IOLJKW SDUDPHWHUV GHWHUPLQHG $ JDLQ VFKHGXOLQJ VFKHPH LV WKHQ LPSOHPHQWHG \LHOGLQJ ORFDO VWDELOLW\ DURXQG WKH RSHUDWLQJ SRLQW 7KH UHVXOWLQJ GHVLJQ LV WHVWHG LQ D QRQOLQHDU VLPXODWLRQ )LQDOO\ WKH UHVXOWV RI WKLV WHVW DUH JLYHQ DQG WKH XVHIXOQHVV RI WKLV GHVLJQ WHFKQLTXH LV HYDOXDWHG

PAGE 16

&+$37(5 7+(25< %(+,1' 7+( '(6,*1 0(7+2' 7KH PDWHULDO LQ WKLV VHFWLRQ JLYHV WKH WKHRU\ OHDGLQJ WR WKH SURSRVHG FRQWUROOHU GHVLJQ PHWKRG 7KH ILUVW VHFWLRQ SUHVHQWV D JHRPHWULF LQWHUSUHWDWLRQ RI /\DSXQRYfV OLQHDU VWDELOLW\ WKHRUHP 7KH HIIHFWV RI FRQWURO RQ WKH YHORFLW\ ILHOG RI D V\VWHP LV VWXGLHG LQ WKH QH[W VHFWLRQ 7KH QRQHPSWLQHVV DQG FRQYH[LW\ RI WKH VHW RI DOO IHHGEDFN JDLQV ZKLFK VWDELOL]H D V\VWHP ZLWK UHVSHFW WR D JLYHQ /\DSXQRY IXQFWLRQ DUH WKHQ GLVFXVVHG )LQDOO\ DQ LWHUDWLYH SURFHGXUH WKDW ILQGV RQH HOHPHQW RI WKLV VHW LV JLYHQ $ *HRPHWULF ,QWHUSUHWDWLRQ RI /\DSXQRYfV /LQHDU 6WDELOLW\ 7KHRUHP 7R XQGHUVWDQG ZK\ WLPH LQYDULDQFH LV D QHFHVVDU\ DVVXPSWLRQ IRU HLJHQVWUXFWXUH GHVLJQ PHWKRGV WKH IROORZLQJ VHFRQG RUGHU OLQHDU WLPH YDU\LQJ V\VWHP ZDV VWXGLHG 7KH H[DPSOH JLYHQ QRZ LV IURP 9LG\DVDJDUfV H[DPSOH >@ DQG FDQ DOVR EH IRXQG LQ .KDOLO >@ *LYHQ WKH IROORZLQJ V\VWHP [ $Wf[ f ZKHUH P f§ O DFRVWf f§ D VLQLf FRVLf f§ f§ D VLQf FRVWf f§ O DVLQLf 9LG\DVDJDU >@ QRWHV WKDW WKH WUDQVLWLRQ PDWUL[ LV JLYHQ E\ rrf Hf f FRVIf H VLQLf BH'Lf VLQf HFRVf DQG WKH FKDUDFWHULVWLF HTXDWLRQ LV f f $ f§ Df$ f§ Df f§ f 7KH URRWV RI f KDYH QHJDWLYH UHDO SDUWV IRU D 7KH H[SRQHQWV LQ WKH ILUVW FROXPQ RI !L f LQGLFDWH KRZHYHU WKDW WKH V\VWHP LV XQVWDEOH IRU WKHVH YDOXHV RI D

PAGE 17

)LJXUH 7KH WUDMHFWRU\ RI Df WKH V\VWHP f DQG Ef D IUR]HQ V\VWHP 7KLV V\VWHP ZDV VLPXODWHG XVLQJ 0$7/$% ZLWK D DQG [R >@7 7KH HLJHQYDOXHV RI $Wf DUH f§ ? AM DQG f§ )LJXUH D VKRZV WKH VWDWH WUDMHFWRU\ RI WKH V\VWHP ZKHUH ;L LV DVVLJQHG WR WKH KRUL]RQWDO D[LV DQG ; LV DVVLJQHG WR WKH YHUWLFDO D[LV 7KLV SORW GHPRQVWUDWHV WKH LQVWDELOLW\ RI WKH V\VWHP ,I WKH V\VWHP ZHUH IUR]HQ LH $Wf $f WKHQ WKH VWDEOH WUDMHFWRU\ RI )LJXUH E ZRXOG UHVXOW 7KLV WUDMHFWRU\ LV VKRZQ ZLWK WKH YHORFLW\ YHFWRU ILHOG RI WKH IUR]HQ V\VWHP 7R H[SODLQ WKH LQVWDELOLW\ LQ )LJXUH D ZH ZLOO QRZ GLVFXVV WKH WLPH YDU\LQJ YHORFLW\ ILHOG RI HTXDWLRQ f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

PAGE 18

LQVLGH RQH RI WKH SRVLWLYH UHJLRQV 6LQFH WKH YHORFLW\ RI WKH V\VWHP DOZD\V KDV DQ RXWHU UDGLDO FRPSRQHQW WKH PDJQLWXGH RI WKH VWDWH YHFWRU LV DOZD\V LQFUHDVLQJ VR WKDW WKH V\VWHP f LV XQVWDEOH $ GLUHFW UHODWLRQVKLS H[LVWV EHWZHHQ WKH SRVLWLYH DQG QHJDWLYH UHJLRQ ERXQGDULHV DQG WKH HLJHQYHFWRUV RI WKH V\VWHP ,Q WKLV FDVH WKH HLJHQYHFWRUV DUH URWDWLQJ FORFNn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f 7 f ;[; ;; )RU [ WR KDYH DQ LQQHU UDGLDO FRPSRQHQW WKHQ 7KLV LV WUXH ZKHQ ZKHUH W f WW f§ f a [7[ [ $[ f f f f 6R LI [7[ LV QHJDWLYH HYHU\ZKHUH IRU DOO WLPH WKHQ WKH V\VWHP LV VWDEOH 6LQFH WKH V\VWHP LV OLQHDU LW LV VXIILFLHQW WR FKHFN WKH VLJQ RI [7$[ RQ WKH XQLW FLUFOH ,I DOO YHORFLWLHV RQ WKH XQLW VSKHURLG SRLQW LQVLGH WKDW ERXQGDU\ WKHQ WKH VDPH LV WUXH IRU

PAGE 19

E 7 7 U r 7f§A 7 M 7 W r $  M r $ L L 7 r r L L 7 7 W A L U W W U W } > f r  U 7 W / A \ 8f§;f§D / 6 ;, )LJXUH 7KH WUDMHFWRU\ DQG YHORFLW\ ILHOG RI f

PAGE 20

)LJXUH [ DQG [ DOO VSKHURLGV FHQWHUHG RQ WKH RULJLQ 7KLV OHDGV WR WKH IROORZLQJ VWDELOLW\ FRQGLWLRQ ZKLFK LV JLYHQ ZLWKRXW SURRI 6WDELOLW\ &RQGLWLRQ ,I [7[ [7$[ 9 [ [7[ 9 L IRU WKH V\VWHP [ $Lf[ WKHQ WKH RULJLQ LV D JOREDO DV\PSWRWLFDOO\ VWDEOH HTXLOLEULXP SRLQW )LJXUH LOOXVWUDWHV DQ H[DPSOH RI D VHFRQG RUGHU V\VWHP ZKLFK PHHWV 6WDELOLW\ &RQGLWLRQ 1RWH WKDW ZKLOH DOO ILJXUHV JLYHQ VR IDU UHSUHVHQW VHFRQG RUGHU OLQHDU V\VWHPV WKH DERYH VWDELOLW\ FRQGLWLRQ DSSOLHV WR OLQHDU V\VWHPV RI DQ\ RUGHU 7KH DERYH VWDELOLW\ FRQGLWLRQ FDQ EH PDGH OHVV FRQVHUYDWLYH E\ H[SDQGLQJ WKH FODVV RI VKDSHV WR HOOLSVRLGV GHILQHG LQ WKH IROORZLQJ ZD\ [73[ f ZKHUH 3 LV D V\PPHWULF SRVLWLYH GHILQLWH PDWUL[ )LJXUH VKRZV DQ HOOLSVH RI WKH IRUP f GHILQHG IRU D VHFRQG RUGHU V\VWHP 7KH VWDWH YHFWRU [ LV GUDZQ WR VRPH

PAGE 21

)LJXUH $Q H[DPSOH RI D VHFRQG RUGHU V\VWHP ZLWK QHJDWLYH UDGLDO YHORFLW\ FRPn SRQHQWV )LJXUH 2QH YHORFLW\ YHFWRU RQ WKH HOOLSVH [73[

PAGE 22

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f[ LI [73[ [73$Lf[ 9 [ [7[ 9 W IRU VRPH FRQVWDQW SRVLWLYH GHILQLWH PDWUL[ 3 WKHQ WKH RULJLQ LV D JOREDO DV\PSWRWLFDOO\ VWDEOH HTXLOLEULXP SRLQW 7KH DERYH FRQGLWLRQ DSSOLHV WR OLQHDU V\VWHPV RI DQ\ RUGHU /\DSXQRY >@ VWDWHG WKDW LI DQ\ SRVLWLYH GHILQLWH IXQFWLRQ RI WKH V\VWHP VWDWHV )[f LV DOZD\V GHFUHDVLQJ LH 9[f WKHQ WKH V\VWHP LV VWDEOH DV\PSWRWLFDOO\ VWDEOH IRU OLQHDU V\VWHPVf >@ )RU WKH OLQHDU FDVH OHW )[f [73[ f 9[f LV SRVLWLYH GHILQLWH LI DQG RQO\ LI 3 LV D SRVLWLYH GHILQLWH PDWUL[ 7DNLQJ WKH GHULYDWLYH JLYHV 9[f [73[ [73[ f [7$U3[ [73$[ f [W>$W3 3$@[ f /\DSXQRYfV VWDELOLW\ FRQGLWLRQ EHFRPHV [W>$W3 3$@[ 9 [ 9 W f

PAGE 23

6LQFH [7 3[ [7 3$[ A[7>$7 3 3$?[ f 6WDELOLW\ &RQGLWLRQ LV HTXLYDOHQW WR /\DSXQRYfV OLQHDU VWDELOLW\ FRQGLWLRQ ,W ZRXOG EH XVHIXO WR FRPSXWH WKH ORZHU DQG XSSHU ERXQG RI WKH UDWH RI GHFD\ RI D JLYHQ SRVLWLYH GHILQLWH IXQFWLRQ 9[f [7 3[ f RI WKH VWDWHV RI D WLPH YDU\LQJ OLQHDU V\VWHP E\ HYDOXDWLQJ F PP W F PD[ W ;PLQA ?3 $Wf $7Wf3@ ;PD;?>3$Wf $7Wf3@ 6R F L[7>3$f $7Wf3@[ F 9 [ [7[ 9L f f f 7KHQ F DQG F DUH WKH ORZHU DQG XSSHU ERXQGV UHVSHFWLYHO\ RI WKH QRUPDO YHORFLW\ FRPSRQHQWV [73[ RQ WKH XQLW VSKHURLG ,I F WKHQ WKH V\VWHP LV VWDEOH $GGLQJ &RQWURO WR WKH 6\VWHP 7KH UHVXOW LQ VHFWLRQ LV DQ DQDO\WLFDO WRRO RQO\ $ GHVLJQ SURFHGXUH LV QHHGHG WR FRQWURO VWDELOLW\ IRU WKH V\VWHP [ $Wf[ %WfX f ZKHUH X LV WKH FRQWURO YHFWRU DQG [ LV WKH VWDWH YHFWRU 7KLV VHFWLRQ IRFXVHV RQ WZR TXHVWLRQV f :KDW HIIHFW GRHV X KDYH RQ WKH YHORFLW\ ILHOG RI f" f &DQ WKH QRUPDO YHORFLW\ FRPSRQHQW [73[ RQ WKH VSKHURLG [7[ EH FRQWUROOHG"

PAGE 24

)LJXUH $ YHORFLW\ YHFWRU Df ZLWKRXW FRQWURO DQG Ef ZLWK FRQWURO 7R DQVZHU TXHVWLRQ f FRQVLGHU D VHFRQG RUGHU VLQJOH LQSXW V\VWHP IUR]HQ DW VRPH LQVWDQW LQ WLPH [ $[ )"X f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f

PAGE 25

)LJXUH $ FDVH LQ ZKLFK WKH QRUPDO YHORFLW\ FRPSRQHQW FDQQRW EH FRQWUROOHG 7KHRUHP &RQVLGHU WKH V\VWHP [ $Wf[ %WfX ZKHUH GLP[f Q GLPXf P DQG $Wf DQG %Wf KDYH FRPSDWLEOH GLPHQVLRQV /HW 3 EH D FRQVWDQW SRVLWLYH GHILQLWH PDWUL[ 7KH QRUPDO YHORFLW\ FRPSRQHQW [73[ FDQ EH DUELWUDULO\ VHW WR DQ\ YDOXH ZLWK WKH ULJKW FKRLFH RI X DW D JLYHQ WLPH W DW DQ\ SRLQW [ RQ WKH VSKHURLG [7[ XQOHVV [ LV FRQWDLQHG LQ WKH VHW 6Wf 3aVSDQ%sWff IO ^[ [7[ ` f ZKHUH %sWf LV D EDVLV RI FROXPQ YHFWRUV RUWKRJRQDO WR VSDQ%Wff ,I [ f 6Wf WKHQ [73[ [73$Wf[ f DQG WKLV YHORFLW\ FRPSRQHQW FDQQRW EH FRQWUROOHG DW WLPH W 7KHRUHP JLYHV WKH SDUWV RI WKH XQLW VSKHURLG IRU ZKLFK WKH YHORFLW\ FRPSRQHQW [73[ LV XQFRQWUROODEOH DQG FDQ WKHUHIRUH EH XVHG WR GHWHUPLQH LI D OLQHDU WLPH YDU\LQJ

PAGE 26

V\VWHP LV VWDELOL]DEOH ZLWK DQ DSSURSULDWH FKRLFH RI FRQVWDQW SRVLWLYH GHILQLWH 3 ,I WKHUH H[LVWV D FRQVWDQW SRVLWLYH GHILQLWH 3 VR WKDW PD[ PD[ [7ILGLf[ [HVWf f WKH V\VWHP LV VWDELOL]DEOH 7KH DERYH FRQGLWLRQ LV HTXLYDOHQW WR UHTXLULQJ WKH IROORZLQJ PDWUL[ WR EH QHJDWLYH GHILQLWH IRU DOO WLPH f 7KLV IDFW ZDV HVWDEOLVKHG E\ )LHOGV >@ DQG LV WUXH IRU WKH IROORZLQJ UHDVRQV &RQGLn WLRQ f LV WUXH LI DQG RQO\ LI 9LA 9L f 6LQFH WKH GHQRPLQDWRU RI WKH DERYH IUDFWLRQ LV SRVLWLYH ZLWK VRPH VLPSOLILFDWLRQ WKH DERYH VWDWHPHQW LV HTXLYDOHQW WR %O^Wf>$^Wf3aO S$ULf@-%Lf[ 9A 9L f ZKLFK LV HTXLYDOHQW WR ?%O^Wf>$^Wf3aU S$7Wf@%/Wf 9L f ,I D FRQVWDQW SRVLWLYH GHILQLWH 3 H[LVWV VDWLVI\LQJ FRQGLWLRQ f WKHQ WKH JLYHQ WLPH YDU\LQJ V\VWHP FDQ EH VWDELOL]HG 7KLV UHVXOW LV PRUH JHQHUDO WKDQ D VLPLODU RQH JLYHQ E\ 6KDKUX] >@ DQG WKH WZR DUH HTXLYDOHQW ZKHQ 3 %HIRUH JLYLQJ D SURRI RI 7KHRUHP ZH JLYH DQ LQWXLWLYH H[SODQDWLRQ RI 7KHRUHP IRU WKH VHFRQG RUGHU FDVH ZLWK D VLQJOH LQSXW )LJXUH VKRZV WKH % YHFWRU IRU VXFK D V\VWHP DORQJ ZLWK VRPH HOOLSVRLG [73[ ,I D VWDWH YHFWRU [ LV GUDZQ WR VRPH SRLQW RQ WKLV HOOLSVH ZKRVH QRUPDO LV QRW SHUSHQGLFXODU WR % WKHQ WKHUH LV

PAGE 27

DOZD\V D FRQWURO X ZKLFK FDQ SODFH [ LQVLGH WKH HOOLSVRLG 7KLV LV WUXH EHFDXVH WKH YHORFLW\ ILHOG FDQ DOZD\V EH FRQWUROOHG LQ WKH % GLUHFWLRQ 7KLV LV QRW WUXH KRZHYHU IRU SRLQWV RQ WKH HOOLSVRLG ZKRVH QRUPDO LV SDUDOOHO WR VSDQ^%?Bf DV VKRZQ LQ ILJXUH 7KHVH SRLQWV RQ WKH HOOLSVRLG FDQ EH IRXQG E\ FRPSXWLQJ 3B VSDQ% BOf f ZKHUH %sB LV D EDVLV RI FROXPQ YHFWRUV RUWKRJRQDO WR VSDQ%f :H QRZ JLYH WKH SURRI RI 7KHRUHP 3URRI &DVH [ f 6Wf 6XEVWLWXWLQJ LQWR [73[ ZH KDYH [73[ [73$f[ [73ILLfX f 6LQFH [H WKHQ [ SnOL%AWff f ZKHUH O3MBeff LV DQ\ OLQHDU FRPELQDWLRQ RI %MBWf 6R [7ILILf X O3LffUS33LfX f ?%sWff7%WfX f f DQG WKHUHIRUH IURP ZH KDYH [73L [73$Lf[ f

PAGE 28

&DVH [ A 6Wf 6XSSRVH ZH ZDQW WR VHW [73[ F ZKHUH F LV VRPH DUELWUDULO\ FKRVHQ YDOXH 7KHQ ZH ZDQW F [W3$f[ [W3%fX f 6LQFH [ 6Wf WKHQ [73WfX A DQG WKHUH H[LVWV DW OHDVW RQH X VXFK WKDW LV WUXH 2QH FRXOG VROYH f IRU X WR XVH DV D FRQWURO ODZ EXW WKLV ZRXOG QRW EH SUDFWLFDO WR LPSOHPHQW EHFDXVH $ DQG % FDQ QRW HDVLO\ EH FRPSXWHG /LQHDU VWDWH IHHGEDFN LV PRUH GHVLUDEOH 7KH QH[W VHFWLRQ ZLOO GHYHORS D SURFHGXUH IRU FRPSXWLQJ WKH VHW RI IHHGEDFN JDLQV WKDW ZLOO LPSOHPHQW D IHHGEDFN FRQWURO ODZ NHHSLQJ WKH QRUPDO YHORFLW\ FRPSRQHQW [73[ ZLWKLQ VRPH VSHFLILHG OLPLW $ /LQHDU )HHGEDFN 6HW WR &RQWURO [73[ 5HFDOO IURP 6HFWLRQ WKDW WKH VWDELOLW\ RI D WLPH YDU\LQJ OLQHDU V\VWHP [ $Lf[ f FRXOG EH DQDO\]HG E\ HYDOXDWLQJ WKH ERXQGV RI [ 3$Wf[ RQ WKH VSKHURLG [7[ f 1RZ ZH ZDQW WR ILQG WKH VHW RI DOO OLQHDU VWDWH IHHGEDFN JDLQV IRU WKH FRQWURO ODZ X .Wf[ f IRU WKH V\VWHP [ $Wf[ %WfX f VR WKDW FRQGLWLRQ f KROGV IRU WKH FORVHG ORRS V\VWHP ZKHUH F DQG F DUH QRZ VSHFLILHG 7KLV VHFWLRQ ZLOO JLYH FRQGLWLRQV IRU WKH QRQHPSWLQHVV RI VXFK D VHW IROORZHG

PAGE 29

E\ D GLVFXVVLRQ RI LWV FRQYH[LW\ )LQDOO\ WKLV VHFWLRQ ZLOO SUHVHQW DQ H[KDXVWLYH VHDUFK PHWKRG IRU ILQGLQJ WKH ERXQGDU\ RI WKLV VHW )RU WKH UHPDLQGHU RI WKLV FKDSWHU $ DQG % ZLOO EH IUR]HQ DW RQH LQVWDQW LQ WLPH $ GLVFXVVLRQ IROORZV RQ KRZ WR ILQG D FRQWUROOHU RU VHW RI FRQWUROOHUV ZKLFK VDWLVILHV f DW RQH JLYHQ LQVWDQW ,I D JLYHQ V\VWHP LV WLPH YDU\LQJ WKHQ WKH UHVXOWV ZKLFK IROORZ PXVW EH DSSOLHG IRU DOO WLPH ZLWK 3 EHLQJ FRQVWDQW DQG SRVLWLYH GHILQLWH 7KHVH UHVXOWV DSSO\ WR DOO V\VWHPV ZKLFK FDQ EH VWDELOL]HG ZLWK UHVSHFW WR D /\DSXQRY IXQFWLRQ XVLQJ D FRQVWDQW 3 7KH VXFFHVV RI WKHVH PHWKRGV GHSHQGV RQ WKH H[LVWHQFH RI D FRQVWDQW SRVLWLYH GHILQLWH 3 VR WKDW S[$7Wf@%/^Wf 9L f 6LQFH ZH UHTXLUH 3 WR EH ]HUR WKHVH UHVXOWV DUH FRQVHUYDWLYH /HW & DQG & EH WKH VHW RI DOO ZKLFK VDWLVI\ WKH OHIW DQG ULJKW LQHTXDOLWLHV UHVSHFWLYHO\ RI WKH IROORZLQJ H[SUHVVLRQ F a[W>3$ $W3 3%. .7%73@[ F 9 [ [7[ f ƒM ZKHUH 3 LV SRVLWLYH GHILQLWH 7KH REMHFWLYH LV WR ILQG `& O&U?; f 7KH IROORZLQJ WKHRUHP JLYHV FRQGLWLRQV IRU WKH QRQHPSWLQHVV RI & DQG & 7KHRUHP *LYHQ WKH fQWK RUGHU PLQSXW V\VWHP [ $[ %Q f & DQG DUH QRQHPSW\ LI DQG RQO\ LI WKHUH H[LVWV D SRVLWLYH GHILQLWH 3 VR WKDW WKH IROORZLQJ FRQGLWLRQV DUH WUXH F PLQ[73$[ PLQA[7>3$ $73@[ [HV [JV f

PAGE 30

QS F PD[[ [H 3 $[ PD[ [HV W[7>L0 $W3@[ f &RPPHQWV &RQGLWLRQV f DQG f DUH UHVSHFWLYHO\ HTXLYDOHQW WR F $PLQ>A\+fO%O>$3[ 3aO $W@% A9+fa@ f DQG F $‘PD[>?n+7O%O>$3 3O$7?%A9+f@ f ZKHUH + 3%$f73a%s f 7KH PDWUL[ + LV VTXDUH IXOO UDQN DQG KDV GLPHQVLRQ Q f§ P 7KH DERYH LV WUXH IRU WKH IROORZLQJ UHDVRQV (TXDWLRQ f FDQ EH UHZULWWHQ DV F PD[ A[7>3$ $7 3@[ f PD[ fQ7%O>$3B 3aO$7@%[IL f + + L O  6LQFH + LV SRVLWLYH GHILQLWH WKHQ \+ H[LVWV LV VTXDUH DQG KDV DQ LQYHUVH %\ PDNLQJ WKH VXEVWLWXWLRQ Yfa] f EHFRPHV F PD[ 79+f%,>$3a 3aO$f%A9+fan] f ]7] L ZKLFK LV HTXLYDOHQW WR & $PD[AL9+\n%OL$3 3O$@%A9+f` f (TXDWLRQ f LV HTXLYDOHQW WR f IRU VLPLODU UHDVRQV :H QRZ JLYH D SURRI RI 7KHRUHP

PAGE 31

3URRI :H ZLOO VKRZ WKH QRQHPSWLQHVV RI & & FDQ EH VKRZQ WR EH QRQHPSW\ LQ D VLPLODU PDQQHU &DVH QU f§ PD[[ 3$[ F [H6 WKHQ [73$[ F 9 [ f 6 f f /HW N%7 3 ZKHUH N LV D VFDODU YDOXH 7KHQ [73[ A[W>3$ $W3 3%. .7%73@[ f ?[7>3$ $W3@[ AN>[73%`>%73[? f 6LQFH [W3% >%73[@7 WKHQ [W3%%W3[! 29[A6 9[7[ f ,I f LV WUXH DQG LI N LV PDGH ODUJH HQRXJK WKHQ WKH VHFRQG WHUP LQ f ZLOO GRPLQDWH 9 [ 6 DQG [73[ F 9 [ [7[ 6LQFH N%73 H ; WKHQ ; LV QRQHPSW\ &DVH PD[[7-'$[ F f [HV n 7KHQ QU [ 3 $[ F IRU VRPH [ 6 f %\ 7KHRUHP [3$[ FDQ QRW EH FRQWUROOHG 9 [ e VR & LV HPSW\

PAGE 32

5HPDUNV 7KH QRQHPSWLQHVV RI WKH LQWHUVHFWLRQ & & IO & LV QRW JXDUDQWHHG ,I WKH GHVLJQHU GLVFRYHUV WKDW QR LQWHUVHFWLRQ H[LVWV WKHQ WKH XSSHU DQG ORZHU YHORFLW\ ERXQGV ZLOO KDYH WR EH DGMXVWHG $QRWKHU XVHIXO SURSHUW\ RI & LV FRQYH[LW\ 7KLV SURSHUW\ LV YDOXDEOH LQ IRUPXODWLQJ LWHUDWLYH VHDUFK WHFKQLTXHV WR EH GHVFULEHG LQ WKH QH[W VHFWLRQ & LV FRQYH[ LI ZKHQ [ DQG \ DUH HOHPHQWV RI & WKHQ D[ f§ Df\ LV DOVR DQ HOHPHQW RI & IRU D >@ 7KHRUHP /HW EH WKH VHW RI DOO VR WKDW $PDr>?>3^$ %.f $ %.fW3`` F f /HW -& EH WKH VHW RI DOO VR WKDW $PLQ>A>3$ %.f $ %.I3@@ F f 7KHQ & & DQG & & IO & DUH FRQYH[ 3URRI ,I & DQG & DUH FRQYH[ WKHQ WKH LQWHUVHFWLRQ & LV FRQYH[ &RQYH[LW\ RI & ZLOO EH SURYHQ KHUH ,Q D VLPLODU PDQQHU WKH SURRI IRU WKH FRQYH[LW\ RI & FDQ EH ZULWWHQ /HW .L DQG EH HOHPHQWV RI & :H PXVW VKRZ WKDW D.? f§ Df. LV FRQWDLQHG LQ & ZKHQ D :H NQRZ WKDW A[W>3$ $73 3%.L .r%W3@[ F 9 [7[ f A[W>3$ $73 3%, LU3U3@[ F 9 [7[ f DQG

PAGE 33

7KHQ aD[7>3$ $W3 3%. ."%73@[ DF 9 [7[ f DQG AODf[7>3$$733%..A%U3@[ DfF 9[7[ f ZKHUH D $GGLQJ f WR f JLYHV s[7>3$ $73@[ ?D[7>3%.[ ."%W3@[ f§_ f§ Df[7>3M%$U .A%73@\L F 9 [7[ ZKLFK EHFRPHV f A7>3$$733%D.^ODf.fD.ODf.f7%73` F 9[7[ f 6R D.L f§ ROf. & IRU D & LV FRQYH[ 1RZ WKDW ZH KDYH FRQGLWLRQV RQ WKH QRQHPSWLQHVV DQG FRQYH[LW\ RI & DQG & LW ZRXOG EH GHVLUDEOH WR ILQG WKH ERXQGDU\ RI WKHVH VHWV :H NQRZ WKDW LI RQH RI WKH HLJHQYDOXHV RI WKH VTXDUH PDWUL[ 4 LV F WKHQ GHW>4 f§ FO@ f 7KLV LV KHOSIXO LQ XQGHUVWDQGLQJ WKH IROORZLQJ H[KDXVWLYH VHDUFK PHWKRG IRU FRPSXWn LQJ WKH ERXQGDU\ RI & $ VLPLODU SURFHGXUH H[LVWV IRU ILQGLQJ GO& f $OO HLJHQYDOXHV RI L>3$ $W3 3%. .W%W3@ FO f QHHG WR EH OHVV WKDQ RU HTXDO WR ]HUR )L[ DOO EXW RQH RI WKH HQWULHV RI /HW WKH MfWK HQWU\ EH WKH IUHH HQWU\ DQG EH UHSUHVHQWHG E\ N /HW .TM

PAGE 34

FRQWDLQ DOO WKH IL[HG HQWULHV RI DQG EH GHILQHG DV IROORZV A‘ N LM N? Q 7KHQ ZKHUH /2LM NQ N^ NPL NPM NPQ N 4LWM LI 4LM ‘f§ P R LM M Q f f f f )LQG DOO YDOXHV RI WKH IUHH HQWU\ ZKLFK PDNH GHW>3$ ƒ 3 FO 3%. ,7%73` f 7KLV SUREOHP UHGXFHV WR ILQGLQJ WKH URRWV RI D SRO\QRPLDO $IWHU VXEn VWLWXWLQJ HTXDWLRQ f LQWR WKH DUJXPHQW RI WKH DERYH GHWHUPLQDQW LW EHFRPHV N3%4LWM 4-M%W3f 3$ $W3 3%.RM .A%73f f /HW ScLMf 3%4 4O%73f f 3.Rccf ‘‘ 3$ $W3 3%.RLL .MX %U3 f

PAGE 35

7KHQ f EHFRPHV GHW> N3 3@ f 6ROYH f IRU N f 5HMHFW DOO FRPSOH[ URRWV ,I DOO WKH URRWV DUH FRPSOH[ WKHQ VNLS WKH QH[W VWHS f 7HVW WKH LQWHUYDOV EHWZHHQ WKH UHDO URRWV E\ FKHFNLQJ WR VHH LI $PD[>3$ $W3 7HO 3%. .W%W3@ f 7KH .nV WKDW ERXQG WKH LQWHUYDO ZKLFK VDWLVILHV f OLH RQ WKH ERXQGDU\ RI & &RQYH[LW\ RI & LPSOLHV WKDW QR PRUH WKDQ WZR .nV ERXQG WKLV LQWHUYDO f 5HSHDW WKH SURFHVV IRU DOO SRVVLEOH YDOXHV IRU WKH IL[HG HQWULHV LQ 7KH UHVXOW LV GO& GM& FDQ EH FRPSXWHG LQ D VLPLODU ZD\ E\ UHYHUVLQJ WKH LQHTXDOLWLHV LQ WKH DERYH SURFHGXUH DQG E\ UHSODFLQJ ?PD[ ZLWK $PQ 7R ILQG & WKH LQWHUVHFWLRQ RI & DQG & FDQ EH IRXQG LQ VWHS f 7KH IL[HG HQWULHV LQ VWHS f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

PAGE 36

([DPSOH *LYHQ WKH V\VWHP ZKHUH DQG OHW [ $[ % X f $ % FRVOErf VLQrf 3 :H ZDQW WR ILQG DOO IHHGEDFN JDLQV ZKLFK VDWLVI\ WKH IROORZLQJ FRQVWUDLQW F [ 3s F 9[[[ O f %HIRUH VSHFLI\LQJ F DQG F ZH PXVW FKHFN WKH QRUPDO YHORFLW\ FRPSRQHQWV IRU [ ( 6 DV 7KHRUHP UHTXLUHV /HW [L %O f§ VLQ r FRV r 7KHQ [I $[L f§ f f )URP 7KHRUHP ZH PXVW KDYH F F f )URP WKLV ZH FKRRVH F F f§ f f

PAGE 37

,Q WKLV H[DPSOH Q DQG P ,W ZDV GHFLGHG WR VHW L M f§ VR WKDW 4LM a 4LL f§ L n f $W?M f§ $ D f DQG $ f§ N4LS $TLL f§ N f )URP HTXDWLRQV f DQG f FRPH $F f§ a>34LL 4LL%7@ f 3LH >%4KL 4Ln%7@ f 3RF $ $7 f§ F f§ %.RLnL f§ .T;,%W f 3RF $ $7 f§ F %.RLnL .T, ;%W f 7KH URRWV RI WKH IROORZLQJ SRO\QRPLDOV DUH FRPSXWHG LQ WHUPV RI N ZKLOH LQFUHPHQWLQJ WKURXJK D ZLGH UDQJH RI YDOXHV GHW>K3LF 3RH@ f GHW>N3LF 7 7RH@ f 5HMHFWLQJ FRPSOH[ URRWV DQG FKHFNLQJ WKH UHJLRQV VHSDUDWHG E\ WKH UHDO URRWV JLYH NF.f NF.f f NF.f NF^.f f 7KH LQWHUVHFWLRQ RI WKHVH UHJLRQV DUH IRXQG N$f f§ PD[INANF@ N$nf P]QINF NF@ f f

PAGE 38

)LJXUH $ SORW RI WKH ERXQGDU\ RI & ZKLFK JXDUDQWHHV VDWLVIDFWLRQ RI WKH GHVLJQ FRQVWUDLQWV RI ([DPSOH )LQDOO\ GO& ^>N^.f .@ 9 .@ 8 ^>N.f ,@ 9 ,` f $ SORW RI G. LV VKRZQ LQ )LJXUH /HW > @ :H FDQ FKHFN WR VHH LI & & E\ HYDOXDWLQJ $L ?PWQ$ $7 %. .W%Wf f f $ ;PD;$ $W%. .W%Wf f f $V WKH IROORZLQJ VKRZV F $L f $ F f

PAGE 39

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f>$Wf3aO 3a$7Wf?%/Wf 9L f %\ 7KHRUHP WKLV FRQGLWLRQ JXDUDQWHHV WKH H[LVWHQFH RI D F VXFK WKDW .^Wf LV QRQHPSW\ IRU HYHU\ LQVWDQW LQ WLPH 7KLV SURFHGXUH DSSOLHV WR DOO V\VWHPV ZKLFK FDQ EH VWDELOL]HG ZLWK UHVSHFW WR D /\DSXQRY IXQFWLRQ JLYHQ E\ D FRQVWDQW SRVLWLYH GHILQLWH 3 7KH IROORZLQJ LV D GLVFXVVLRQ RI WKH LWHUDWLYH /\DSXQRY PHWKRG IROORZHG E\ WKH DOJRULWKP LWVHOI $Q H[DPSOH LV WKHQ JLYHQ DSSO\LQJ WKLV SURFHGXUH WR RQH RSHUDWLQJ SRLQW RI D ILIWK RUGHU OLQHDUL]HG PRGHO RI WKH (05$$7 PLVVLOH )LJXUH LOOXVWUDWHV WKH LWHUDWLYH SURFHGXUH I&I LV WKH IHHGEDFN VHW ZKLFK VDWLVILHV WKH GHVLJQHUfV SUHGHWHUPLQHG FRQVWUDLQWV IRU VRPH VSHFLILHG 3 7KH FRQVWUDLQWV DUH eI [W>3$ $W3 3%. .7%73?[ FI 9 [ [7[ f /HW ,&L EH GHILQHG DV WKH VHW RI DOO ZKLFK VDWLVILHV WKH IROORZLQJ FRQVWUDLQWV 68 [W>3$ $W3 3%. .W%W3@[ F^ 9 [ [7[ f *LYHQ .L ZH ZRXOG OLNH WR ILQG F DQG F VR WKDW LI .L A &I WKHQ .] Gf&L :H DOVR UHTXLUH WKDW ,&I & .^ ,I .] e -&I WKHQ ZH ZDQW -&^ .c 7KH IROORZLQJ GHILQLWLRQV IRU &M DQG F PHHW WKHVH UHTXLUHPHQWV /HW $ PD[ [7>3$ f§ %,&& $ f§ %.&73KW r f f

PAGE 40

)LJXUH $ JHRPHWULF YLHZ RI WKH LWHUDWLYH /\DSXQRY GHVLJQ PHWKRG

PAGE 41

DQG $ PP A[7>3 %.Wf $ %,Wf73@[ f 7KHQ OHW F PD[>FM?L@ f DQG F PLQ>FI $@ f .T LV WKH LQLWLDO JXHVV LQ WKH VHDUFK IRU .I f .I F DQG &T DUH FRPSXWHG XVLQJ f DQG f VR WKDW .T LV D PHPEHU RI G-&R DQG .I & .T 7KHQ D QHZ IHHGEDFN PDWUL[ .? LV IRXQG ZKLFK OLHV LQVLGH RI .T EXW QRW RQ WKH ERXQGDU\ 1HZ FRQVWUDLQLQJ YDOXHV DUH IRXQG LQ WKH VDPH ZD\ DV EHIRUH VR WKDW WKH ERXQGDU\ RI WKH QH[W IHHGEDFN VHW FRQWDLQV .? .b LV WKHQ IRXQG VR WKDW LW OLHV LQVLGH RI WKH SUHVHQW IHHGEDFN VHW EXW QRW RQ LWV ERXQGDU\ 7KLV SURFHVV LV FRQWLQXHG XQWLO .O .I f .I 7KH VXFFHVV RI ILQGLQJ .I GHSHQGV RQ WKH IROORZLQJ FRQGLWLRQV *LYHQ WKDW .L f G.L ZH PXVW EH DEOH WR ILQG I*L VXFK WKDW .^ .^ :H PXVW VKRZ WKDW FL F ZKHQ F F DQG FLO F ZKHQ F^ &I :H PXVW VKRZ WKDW .I & .L & .L .I PXVW EH QRQHPSW\ :H QRZ DGGUHVV WKHVH IRXU SRLQWV *LYHQ .L f G.L ZH QHHG WR ILQG D VHFRQG IHHGEDFN PDWUL[ . f .^ ZKHUH . .L 7KHQ GXH WR FRQYH[LW\ ?.L ?.D LV D PHPEHU RI .W )LJXUH D VKRZV D VHFRQG RUGHU H[DPSOH RI D SURFHGXUH IRU ILQGLQJ .LL 7KH DOJRULWKP ZLOO EH JLYHQ VKRUWO\ 7KH KRUL]RQWDO DQG YHUWLFDO D[LV DUH DVVLJQHG WR NQ DQG  UHVSHFWLYHO\ 7KH UHJLRQ .L LV HQFORVHG E\ nG. DQG G.L .L LV NQRZQ .L LV IRXQG E\ VHDUFKLQJ DORQJ

PAGE 42

WKH OLQH WKDW SDVVHV WKURXJK .^ DQG LV SDUDOOHO WR WKH NQ D[LV 3RLQW F LV IRXQG E\ FRPSXWLQJ ?.L ?.^ 6LQFH & LV FRQYH[ WKHQ F $& 7KLV VWHS LV UHSHDWHG DJDLQ E\ VHDUFKLQJ DORQJ WKH N[ D[LV 8VLQJ VLPLODU DUJXPHQWV SRLQW LV DOVR LQ $&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n DPSOH &L HTXDO WR F Gf&L ZLOO WKHQ PRYH VR WKDW SRLQW E FDQ EH IRXQG DV VKRZQ LQ ILJXUH F /DWHU WKH UHOD[HG FRQVWUDLQLQJ YDOXH FDQ WDNH RQ LWV RULJLQDO DVVLJQPHQW :H QHHG WR VKRZ WKDW F FL ZKHQ F &I ,Q WKLV FDVH &L f§ $L &I f 6LQFH .LL f ,&O DQG .L e $$ WKHQ WKH IROORZLQJ FRQGLWLRQ KROGV ZLWK VWULFW LQHTXDOLW\ A[W>3$ %.Lf $ %.Lf73@[ F 9 [ [7[ f 7DNLQJ f IRU L WKHQ A[W>3$ %.Lf $ %$nfW)@[ $L 9 [ [7[ f DQG WKHUH H[LVWV DQ [ ZKLFK VDWLVILHV WKH DERYH FRQGLWLRQ IRU HTXDOLW\ 7KHUHIRUH &L $WL DQG &L PD[?FI?L@@ F 8VLQJ VLPLODU DUJXPHQWV LW FDQ EH VKRZQ

PAGE 43

D E N; N? )LJXUH 6WHS RI WKH LWHUDWLYH /\DSXQRY GHVLJQ PHWKRG IRU Df D VHFRQG RUGHU H[DPSOH Ef KRZ LW VRPHWLPHV IDLOV DQG Ff KRZ WKLV SUREOHP LV FRUUHFWHG

PAGE 44

WKDW F FL ZKHQ FW &M :H QRZ VKRZ WKDW ,&I & -&L & & 6LQFH F PD[>F $@ WKHQ F F :H KDYH DOUHDG\ VHHQ WKDW FW $L 6R F PD[>F $@ &^? 6LPLODUO\ F F ;$ LV WKH VHW RI DOO VR WKDW FL A[7>3$ f§ %.f $ f§ 0fU3@[ &LL 9 [ [7[ f 6LQFH &L &LL DQG F F WKHQ IRU HYHU\ HOHPHQW RI WKH IROORZLQJ KROGV F L[7>3A 3$f $ 3$fW3@[ F 9 [ [7[ f 7KHUHIRUH -&L & & 6LQFH FL &M DQG FL &I WKHQ XVLQJ D VLPLODU DUJXPHQW ,&I & ,&L ,Q XVLQJ WKLV GHVLJQ SURFHGXUH 3 LV FKRVHQ VR WKDW WKH PD[LPXP XQFRQWUROn ODEOH QRUPDO YHORFLW\ FRPSRQHQW LV QHJDWLYH 7KHQ IURP 7KHRUHP F FDQ EH PDGH QHJDWLYH LQ DQ DWWHPSW WR DFKLHYH VWDELOLW\ Fc PXVW EH JUHDWHU WKDQ WKH PD[LPXP XQFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQW DQG Fc PXVW EH OHVV WKDQ WKH PLQLPXP XQFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQW )URP 7KHRUHP WKLV ZLOO JXDUDQWHH WKH QRQHPSWLQHVV RI ,&I DQG ,&I 7KH QRQHPSWLQHVV RI WKH LQWHUVHFWLRQ RI WKHVH WZR VHWV KRZHYHU LV XQNQRZQ ,I ,&I LV QRQHPSW\ WKHQ DV L EHFRPHV ODUJH $ e ,&I ,I ,&I LV HPSW\ WKHQ FO DQG F ZLOO FRQYHUJH WR YDOXHV ZKLFK GR QRW PDWFK WKH GHVLUHG FRQn VWUDLQWV DQG .^ ZLOO \LHOG D FORVHG ORRS V\VWHP WKDW PHHWV WKH FRQVWUDLQWV JLYHQ E\ F DQG FW 7KH GHVLJQHU ZLOO HLWKHU KDYH WR DFFHSW WKLV UHVXOW RU WU\ DJDLQ ZLWK D GLIIHUHQW 3 RU GLIIHUHQW FRQVWUDLQLQJ YDOXHV RU ERWK 6LQFH VWDELOLW\ LV GHVLUHG RQH DSSURDFK ZRXOG EH WR NHHS 3 DQG Fc DQG ORZHU &I XQWLO ,&I EHFRPHV ODUJH HQRXJK WR LQWHUVHFW & 7KH RXWOLQH RI WKH LWHUDWLYH /\DSXQRY GHVLJQ PHWKRG LV DV IROORZV

PAGE 45

f &KRRVH 3 &I DQG F 7KLV VHOHFWLRQ PXVW REH\ 7KHRUHP ,W VKRXOG EH QRWHG WKDW 7KHRUHP JXDUDQWHHV WKH QRQHPSWLQHVV RI W&I DQG ,&I EXW QRW WKHLU LQWHUVHFWLRQV ,I WKH V\VWHP LV WLPH YDU\LQJ WKHQ LQ RUGHU WR XVH WKLV DOJRULWKP 3 PXVW EH IRXQG VR WKDW ?%O^Wf>$^Wf3aO S$7Wf?%/^Wf 9L f 2WKHUZLVH WKLV DOJRULWKP FDQQRW JXDUDQWHH VWDELOLW\ f &RPSXWH F DQG F VR WKDW G.R DQG -&I & I&R ZLOO EH WKH LQLWLDO JXHVV LQ WKH VHDUFK IRU .I f /HW L f /HW L L 7 f )LQG .L VR WKDW .^ DQG .^ f &RPSXWH FW DQG F VR WKDW .O G.[ DQG .c & RU VR WKDW ,&I f 5HSHDW VWHSV f f DQG f XQWLO RQH RI WZR HYHQWV RFFXU F &M DQG F FM F f§ FLBf DQG F f§ FBLf EHFRPH YHU\ VPDOO 5HPDUNV ,I HYHQW RFFXUV WKHQ f&I LV QRQHPSW\ DQG .W f&M ,I HYHQW RFFXUV WKHQ ,& LV HPSW\ DQG .L \LHOGV D FORVHG ORRS V\VWHP WKDW PHHWV WKH FRQVWUDLQWV FRUUHVSRQGLQJ WR F DQG F :H QRZ H[SODLQ KRZ WR SHUIRUP VWHSV f f DQG f 6WHS f

PAGE 46

7KH ORZHU DQG XSSHU ERXQG RI [73[ IRU WKH RSHQ ORRS V\VWHP L WLYHO\ $R f§ APLQ>aA^3 $ ƒ) 3f? $R $PD[>f§3 $ $7 3f` 7KHQ F PLQ>FI $T@ F PD[>FI $@ 6WHS f 7KH DOJRULWKP LV QRZ JLYHQ .L )RU M WR P )RU N WR Q 3R MN 3 NM4MN 3L 3%4MN 4ON%73f 3R 3$ %.RMnNf $ %.RNf73 FO 6ROYH GHWIN3L 3@ LQ WHUPV RI N IRU F F DQG F (HMHFW DOO YDOXHV ZKLFK GR QRW PHHW WKH FRQVWUDLQWV IURP VWHSV f DQG f 7KH UHVXOW LV WZR LQWHUYDOV ZKRVH ORZHU ERXQG LV NLB DQG N DQG ZKRVH XSSHU ERXQG LV N? DQG  f )LQG WKH LQWHUVHFWLRQ RI WKHVH LQWHUYDOV E\ HYDOXDWLQJ UHVSHF f f f f

PAGE 47

N PD[>NX N? N f§ PLQ>NL @ )LQG WKH PLGSRLQW RI WKH LQWHUYDO E\ FRPSXWLQJ N MN ?NNf /HW ,? NMN4MN G .TMN 1H[W N 1H[W M .LL 6WHS f /HW $ ;PLQ>?3$ $733%.L.7%73f? f $ $PD[>L3$ $733%.L.M%73f@ f t 6R &M PLQ>FIL $@ f &L PD[>F$@ f 7KH IROORZLQJ H[DPSOH LOOXVWUDWHV WKH IHDVLELOLW\ LQ DSSO\LQJ WKLV PHWKRG WR WKH (05$$7 PLVVLOH ([DPSOH :H ZRXOG OLNH WR DSSO\ WKH LWHUDWLYH /\DSXQRY GHVLJQ PHWKRG WR WKH (0n 5$$7 PLVVLOH 7KH PLVVLOH ZDV IORZQ LQ D VLPXODWLRQ WKURXJK D WUDMHFWRU\

PAGE 48

XVLQJ DQRWKHU DXWRSLORW GHVLJQ 7KH PRGHO ZDV OLQHDUL]HG DQG WKH IROn ORZLQJ V\VWHP ZDV WDNHQ DW VHFRQGV LQWR WKH IOLJKW ; $[ %X f ZKHUH n n $ f n % DQG [ >DS"U@U f f 6WHS f /HW 3 7KH PLQLPXP DQG PD[LPXP XQFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQWV ZHUH IRXQG E\ FRPSXWLQJ WKH OLPLWLQJ YDOXHV LQ HTXDWLRQV f DQG f 6LQFH 3 f§ WKHVH WHUPV VLPSOLI\ DQG DUH HYDOXDWHG DV IROORZV PLQ [7 >3$ $ 3KL ;6 DQG PD[ [7>3$ $7 3?[ [HV $ UPQ>?%O$ $A%@ f f ?PD;>?%O$ $7f%/@ f f

PAGE 49

7KHRUHP JXDUDQWHHV WKH QRQHPSW\QHVV RI & DQG & LI Fc f§ DQG FM f§ :LWK WKLV LQ PLQG ZH OHW Fc f§ DQG f§ 6LQFH WKH LQWHUVHFWLRQ RI & DQG & LV QRW JXDUDQWHHG &M ZDV FKRVHQ WR EH YHU\ QHJDWLYH WR LQFUHDVH WKH SUREDELOLW\ RI JHWWLQJ DQ DQVZHU 6WHS f 6R $ f§ APLQ >A$ $Uf@ f f f A $7f@ f f f F PPF\ $f f f &R PD[FI $f f f 6WHSV f f $ SURJUDP ZDV ZULWWHQ IRU 0$7/$% WR FDUU\ RXW WKH LWHUDWLRQV LQ VWHSV f f 7KH SURJUDP ZRXOG WHUPLQDWH LI & M ZDV IRXQG RU ZKHQ _&L f§ FWBL_ H f

PAGE 50

.c ZDV IRXQG EHIRUH 7KH SURJUDP UDQ LWHUDWLRQV 7KH ILQDO UHVXOW LV n ,, f B 1RZ WR FKHFN WKH UHVXOW ,, A APLQ > \ A $U %.I f f ;I $U %.M .M%7f@ f f§ f 7KLV PHHWV WKH GHVLUHG FRQVWUDLQW &I ;I ?I Fc

PAGE 51

&+$37(5 $ 7,0( 9$5<,1* 6(&21' 25'(5 (;$03/( 7KH IROORZLQJ SUREOHP JLYHV D FDVH ZKHQ SROH SODFHPHQW VXFFHHGV LQ JLYLQJ HLJHQn YDOXHV ZLWK QHJDWLYH UHDO SDUWV EXW IDLOV WR VWDELOL]H WKH V\VWHP 7KH /\DSXQRY GHVLJQ PHWKRG LV WKHQ HPSOR\HG DQG WKH UHVXOWLQJ FORVHG ORRS V\VWHP LV VKRZQ WR EH DV\PSWRWLFDOO\ VWDEOH IRU DOO WLPH :H ZRXOG OLNH WR ILQG D IHHGEDFN FRQWURO ODZ WKDW VWDELOL]HV WKH V\VWHP [ $Lf[ %Wf X f ZKHUH $Wf f§ FRVf>FRVf FRV 7 f@ f§ VLQf>FRVf FRV W WWf@ f§ f§ FRVf>VLQf VLQ UfM f§ VLQWf>VLQLf VLQW WWf@ f DQG %Wf &26 LW UOf f§ VLQW Wf f 7KH HLJHQYDOXHV RI $Wf DUH DQG IRU DOO WLPH 7KH IROORZLQJ FRQWURO ODZ LV SURSRVHG X > FRVLf f§VLQWf @[ f 7KH UHVXOWLQJ FORVHG ORRS V\VWHP FDQ EH IRXQG LQ H[DPSOH E\ 9LG\DVDJDU >@ DQG DOVR LQ .KDOLO >@ 7KH HLJHQYDOXHV IRU WKH UHVXOWLQJ FORVHG ORRS V\VWHP DUH $ f§ s M 6LQFH WKH HLJHQYDOXHV KDYH QHJDWLYH UHDO SDUWV RQH ZRXOG H[SHFW WKH FORVHG ORRS V\VWHP WR EH VWDEOH +RZHYHU 9LG\DVDJDU VKRZV WKDW WKH WUDQVLWLRQ PDWUL[ LV r f Hn FRVf H r VLQWf f§HRVLQf HBFRVf f

PAGE 52

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f 6LQFH 3 LV WKH LGHQWLW\ PDWUL[ ZH DUH LQWHUHVWHG LQ WKH QRUPDO YHORFLW\ FRPSRQHQW ZKLFK LV RQ WKH SDUW RI WKH XQLW FLUFOH ZKRVH WDQJHQW LV SDUDOOHO WR % 6R ZH OHW [ %O > HU f 7KH XQFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQW DW W LV [73[ A[W>3$f Uf3@[ f

PAGE 53

7KH XQFRQWUROODEOH YHORFLW\ IRU WKLV H[DPSOH LV FRQVWDQW IRU DOO WLPH ,Q VHOHFWLQJ WKH FRQVWUDLQLQJ YDOXHV ZH PXVW KDYH F F f 7KH IROORZLQJ DVVLJQPHQWV DUH PDGH F f§ f F f§ f 6LQFH WKH V\VWHP LV VHFRQG RUGHU DQG KDV RQO\ RQH LQSXW LW LV SRVVLEOH WR SORW WKH VHW RI DOO IHHGEDFN JDLQV ZKLFK VDWLVI\ WKH IROORZLQJ F A[7>3$Wf f§ %Wf.Wff $Wf f§ %Wf.Wff73@[ F 9[ [7[ 9L f 7KLV WLPH YDU\LQJ IHHGEDFN VHW LV VKRZQ LQ ILJXUH 6LQFH WKH WLPH YDU\LQJ QDWXUH RI WKH V\VWHP LV SHULRGLF DQG IURP LQVSHFWLRQ RI ILJXUH WKH IROORZLQJ FRQWURO ODZ LV FKRVHQ X >FRVLf f§ VLQLf@[ f 7KH IHHGEDFN PDWUL[ LQ f LV VKRZQ WR EH LQVLGH WKH PRYLQJ IHHGEDFN VHW LQ ILJXUH 7KH HLJHQYDOXHV RI WKH GHULYDWLYH RI WKH UHVXOWLQJ /\DSXQRY IXQFWLRQ LV $>$f %^Wf.^Wf $7^Wf .7^Wf%7^Wf` 9L f 7KHUHIRUH WKH FORVHG ORRS V\VWHP LV VWDEOH )LJXUH VKRZV D WUDMHFWRU\ RI WKH /\DSXQRY EDVHG FORVHG ORRS V\VWHP ZKHUH WKH LQLWLDO FRQGLWLRQ LV [ > @7

PAGE 54

)LJXUH 7KH WLPH YDU\LQJ IHHGEDFN VHW ZKLFK VDWLVILHV WKH GHVLJQ FRQVWUDLQWV

PAGE 55

7UDMHFWRU\ RI WKH /\DSXQRY EDVHG FORVHG ORRS WLPH YDU\LQJ V\VWHP )LJXUH $ WUDMHFWRU\ RI WKH FORVHG ORRS V\VWHP XVLQJ WKH /\DSXQRY GHVLJQ PHWKRG ;T > @7 7KH V\VWHP LV VWDEOH

PAGE 56

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

PAGE 58

]HUR $OVR WKH ZHLJKW RI WKH PLVVLOH LV VPDOO LQ FRPSDULVRQ WR WKH DHURG\QDPLF IRUFH VR JUDYLW\ LV QHJOHFWHG 7KH DHURG\QDPLF IRUFH LQ WKH [ GLUHFWLRQ LV PXFK VPDOOHU WKDQ WKH DHURG\QDPLF IRUFH LQ WKH \ DQG GLUHFWLRQV DQG WKHUHIRUH ZLOO EH LJQRUHG IRU WKH UHVW RI WKH SDSHU 1HZWRQfV VHFRQG ODZ RI PRWLRQ LPSOLHV WKH IROORZLQJ )\ P>Y UX f§ SZ@ f )] P>Z SY f§ TX? f )\ DQG )] DUH WKH DHURG\QDPLF IRUFHV LQ WKH \ DQG GLUHFWLRQV 7KH TXDQWLWLHV LQVLGH WKH EUDFNHWV DUH WKH WRWDO DFFHOHUDWLRQV LQ WKH \ DQG GLUHFWLRQV :H NQRZ WKDW 9 >X Z`r f 6LQFH Y DQG Z DUH PXFK VPDOOHU WKDQ X WKHQ 9 a X $QJOH RI DWWDFN DQG VLGHVOLS DUH JLYHQ E\ D DUFWDQ IW DUFWDQ ,I ZH DVVXPH WKDW D DQG IW DUH VPDOO WKHQ Z D a f§ X DQG X (TXDWLRQV f DQG f FDQ EH UHZULWWHQ DV )\ PX 9 Z E USf§ X X f f f f 6f f

PAGE 59

)] f§ PX nZ Y f§ 3 /X X ZKLFK VLPSOLILHV WR )\ P9 )] P9 nY f§I U f§ SD X nZ f§ S T / X :H DVVXPH WKDW WKH IRUZDUG YHORFLW\ FKDQJHV VORZO\ VR WKDW a 7KHQ DQG 6R f DQG f EHFRPH Z D a f§ X 3 a X P9 )]B P9 U f§ SHW f§ •/ S T 7KH DHURG\QDPLF IRUFHV DUH JLYHQ E\ )\ 46 >F
PAGE 60

IRUP DQG DUH JLYHQ LQ WKH ILUVW DSSHQGL[ DORQJ ZLWK RWKHU GDWD SHUWDLQLQJ WR WKH (05$$7 PLVVLOH 6XEVWLWXWLQJ f DQG f LQWR f DQG f DQG VROYLQJ IRU £ DQG JLYHV f§ SRW f§ U >&\ &\SS &\UU &\6SS &\U67 DQG f 46&1D 46 A m f§ RW TSS f§c &1DD &1TT &1VTT D P9 46&1£ Lf§O U P9 P9 46 T S f§\ &1DD &1TT &1LT6T f f f (TXDWLRQV f DQG f DUH WZR QRQOLQHDU VWDWH HTXDWLRQV ,Q RUGHU WR VHSDUDWH SLWFK G\QDPLFV IURP UROO\DZ G\QDPLFV LV DVVXPHG FORVH WR ]HUR LQ f 7KXV D 46&1R P9 L 4V ? I 46 &Q f D f§ &MY f T P9 P9 46 P9 D 16T rT f (TXDWLRQ f FDQ EH ZULWWHQ DV rt}f rfr‘r rf2rf r f :H QRZ WXUQ WR WKH PRPHQW HTXDWLRQV RI WKH PLVVLOH 6LQFH WKUXVW LV ]HUR DQG VLQFH JUDYLW\ GRHV QRW FRQWULEXWH DQ\ PRPHQW WR WKH PLVVLOH WKHQ WKH PRPHQWV DURXQG WKH [ \ DQG D[LV DUH JLYHQ E\ P DQG Q UHVSHFWLYHO\ WKH PRPHQWV GXH WR DHURG\QDPLF SUHVVXUH 7KH\ DUH JLYHQ E\ O f§ 46G >&LS &LSS &LUU &LS6S &LWW P f§ 46G &UULD4 f &PW& 7 &PTT a &PTT f f DQG Q f§ 46G &QSIO &QS3 &QUU &Q6SS &Q6U f

PAGE 61

ZKHUH G LV WKH PLVVLOH GLDPHWHU $JDLQ WKH DHURG\QDPLF FRHIILFLHQWV FRPH IURP ZLQG WXQQHO WHVWLQJ DQG DUH WDEXODWHG LQ WKH ILUVW DSSHQGL[ (XOHUfV WKUHH PRPHQW HTXDWLRQV FDQ EH ZULWWHQ DV IROORZV ‘ O n P f 3 n T Q U ZKHUH +STUf ,[[ f§,[< f§,[] f§,[\ ,\\ f§,\] f§,[] f§,\] ,]] f f DQG ,\\ ,]]fTU :]U Tf ,[YST ,[\US + a,]] ,[[fUS ,[]S a Uf ,[\TU ,\]ST f ,[[ O<P ,]] a ,[[f US@ >Q ,[[ ,\\f ST@ f f f 6XEVWLWXWLQJ f f DQG f LQWR f f DQG f JLYHV 46G T S 46G K ,[[ ?SOS3 &OS3 &OUU &OSGS &OU67 << aWa &PTT 6-PFcT ,]] ,[[ ,\\ US f f DQG 46G U ,]] &QJ &QS3 &QUU &QV S &QcM6‘ ,[[ f§ ,\\ f§M ST L]]

PAGE 62

,I J\URVFRSH HIIHFWV DUH FRQVLGHUHG VPDOO WKHQ f f f f DQG f FDQ EH ZULWWHQ DV WZR VHSDUDWH V\VWHPV WKH SLWFK G\QDPLFV DQG WKH UROO\DZ G\n QDPLFV 7KH SLWFK G\QDPLFV DUH DV IROORZV D 46&QD Of§O U P9 46U ? \ p6U ? f§&1D f D W\&9 P9 ? P9 46 P9 &16T f WLJ 46G L\\ 7 &W &PcT 7 &UILLTAT f f 7KH UROO\DZ G\QDPLF HTXDWLRQV DUH n ‘ : rmbm!f rrrfA r f f 3 U 46G ,[[ 46G >FKS FOSS &OUU &KSS &WU6U L]] ?&QSIL &QS3 &Q7U &QVSS &QHUU f 7KH VWDWHV RI WKH SLWFK PRGHO DUH D DQG T ZLWK 6T DV LWV LQSXW )RU WKH UROO\DZ PRGHO WKH VWDWHV DUH c S DQG U DQG WKH LQSXWV DUH 3 DQG 6U 7KH /LQHDU 0RGHO 7KH SUHYLRXV VHFWLRQ JDYH D PRGHO RI WKH SLWFK G\QDPLFV DQG WKH UROO\DZ G\QDPn LFV LQ WKH IROORZLQJ IRUP >T T `7 IJDT6Tf f > S U `7 ISU3SUSUf f :H ZRXOG OLNH WR KDYH WZR OLQHDUL]HG PRGHOV IRU XVH ZLWK WKH SURSRVHG GHVLJQ PHWKRG 7KH UHVXOW ZLOO EH WZR V\VWHPV LQ WKH IROORZLQJ IRUP [ $[ %Q f

PAGE 63

ZKHUH [ FRQWDLQV WKH VWDWHV DQG X FRQWDLQV WKH LQSXWV $ DQG % DUH PDWULFHV ZKLFK DUH IXQFWLRQV RI VHYHUDO WLPH YDU\LQJ IOLJKW SDUDPHWHUV DQG DUH FRPSXWHG DV IROORZV $ O [Z ff GI f RPLQDO "f AI: f QRPLQDO ZKHUH Z FRQWDLQV DGGLWLRQDO IOLJKW SDUDPHWHUV 1RWH WKDW [ DQG X DUH QRZ SHUWXUn EDWLRQV IURP WKH SRLQW DURXQG ZKLFK WKH OLQHDUL]DWLRQ LV WDNHQ :H OLQHDUL]H WKH SLWFK PRGHO ILUVW )URP LQVSHFWLRQ RI f ZH VHH WKDW DT?O f§ 46&1D P9 O 46 & DLO 0AU P9 46 1D ETOO f§ P9 4VFA\ I46 f§`&1T P9 P9 P9 &Q 7R OLQHDUL]H f ZH PXVW ILUVW VXEVWLWXWH f LQ IRU £ 7KHQ ZH GLIIHUHQWLDWH DV LQ f DQG f DT f§ DT f§ 46G ,\\ 46G ,\\ 46G ,\\ &PD &P£ 46&1£ L &Pc &Pr L P9 46&1£ P9 46 & P9 46 1D P9r1T &PT &P£ I 46&1£ P9 L 46U OY&1m 7KH UHVXOWLQJ OLQHDUL]HG SLWFK PRGHO LV &W TOO 8TO D ETOO NT DT DT T f 7KH VDPH SURFHGXUH LV DSSOLHG WR WKH UROO\DZ PRGHO $VVXPLQJ WKDW D LV FRQVWDQW LQ f DQG IURP LQVSHFWLRQ RI HTXDWLRQV f WR f WKH IROORZLQJ UHVXOWV WWSUOO f§ 46&\ P9 `f§L DSUO f§ D 46&
PAGE 64

SU E SU f§ 46&\6S P9 f 46G&LS &/SU ;; ESUO f§ ,[[ 46&\U P9 &OSU f§ f SU f§ 46G&Q 46G&OS B 46G&>U ,[[ ESU f§ ,[[ nQS B 46G&QS 46G&QU GSU M L tSU f§ M tSU f§ ,]] fS7 O 46GFQVS / AQU M ff 9SU f§ == ,]] f )-r ,]] 46G&QU ,]] 7KH OLQHDUL]HG PRGHO LV JLYHQ E\ n GSUOO GSUO ASUO f-n ESUOO ESUL S U tSU GSU GS7 3 SU ESU U &WSU GSU GSU U ESU ESU f

PAGE 65

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f 0 f§ f 9 626

PAGE 66

'HSHQGHQW )OLJKW 3DUDPHWHUV )LJXUH $ EORFN GLDJUDP RI WKH V\VWHP\ XVHG WR GHWHUPLQH WKH GHSHQGHQFH RI JDLQV RQ IOLJKW SDUDPHWHUV

PAGE 67

ZKHUH 9VRD LV WKH VSHHG RI VRXQG %RWK S DQG 9VRV DUH IXQFWLRQV RI DOWLWXGH 3 I$Kf f 9VRV IVKf f ZKHUH K LV DOWLWXGH LQ IHHW DERYH VHD OHYHO +HUH IS DQG IV DUH IXQFWLRQV EDVHG RQ DWPRVSKHULF WDEOHV DQG DUH LPSOHPHQWHG E\ OLQHDU LQWHUSRODWLRQ 6ROYLQJ f IRU 9 DQG VXEVWLWXWLQJ WKH UHVXOW LQWR f JLYHV 4 f :H JHQHUDWH D WKLUG WDEOH LQ WKH IROORZLQJ ZD\ 0Kf I3KfIrKf S9VRV f 6R 4 f 7KH IXQFWLRQ LV D RQHWRRQH IXQFWLRQ VR WKDW LWV LQYHUVH FDQ EH IRXQG E\ UHDGLQJ WKH WDEOH EDFNZDUGV :LWK WKLV LQ PLQG ZH FDQ VROYH f IRU K )URP f DQG f 9 0966 0IVKf 6XEVWLWXWLQJ f LQWR f WR HOLPLQDWH K JLYHV $OVR IURP f f f f f

PAGE 68

(TXDWLRQV f DQG f DUH XVHG WR HOLPLQDWH S DQG 9 LQ WKH OLQHDU PRGHO $V D UHVXOW WKH OLQHDUL]HG PRGHO FDQ QRZ EH JHQHUDWHG IURP WKH IROORZLQJ VHYHQ IOLJKW SDUDPHWHUV 0 4 D c S T U 7KH )OLJKW 3DUDPHWHU *HQHUDWRU f $ VHULHV RI IOLJKW FRQGLWLRQV DUH PDGH DQG XVHG WR JHQHUDWH PDQ\ OLQHDU PRGHOV )HHGEDFN JDLQV DUH JHQHUDWHG IRU HDFK FRQGLWLRQ 7KH ILUVW RI WKH VHULHV LV FDOOHG WKH QRPLQDO IOLJKW FRQGLWLRQ 7KH YDOXHV RI WKH SDUDPHWHUV IRU WKH QRPLQDO IOLJKW FRQGLWLRQ DUH 0 4 SVI D r r S rV T rV U r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

PAGE 69

7DEOH nKH LQLWLDO SRLQW DQG UDQJH RI FKDQJLQJ IOLJKW SDUDPHWHUV ,QLWLDO SRLQW 0LQLPXP YDOXH 0D[LPXP YDOXH 0 4 SVI SVI SVI D r r r 3 r r r 3 rV rV rV 4 rV +} R R f rV U rV rV rV 7KH LQLWLDO IHHGEDFN JDLQV DUH IRXQG E\ XVLQJ D SROH SODFHPHQW DOJRULWKP $W WKH QRPLQDO IOLJKW FRQGLWLRQ WKH OLQHDU PRGHOV DUH JLYHQ E\ %T $ f§ U/USU f§ / / f f %SU f§ f f 7KH GHVLUHG HLJHQYDOXHV IRU WKH FORVHG ORRS SLWFK G\QDPLFV KDYH EHHQ FKRVHQ WR EH f§sMO )RU WKH FORVHG ORRS UROO\DZ PRGHO WKH GHVLUHG SROH ORFDWLRQV DUH f§ sM f§ 7KH UHVXOWLQJ IHHGEDFN JDLQV DUH .T .S7 f§ f )RU ERWK FORVHG ORRS V\VWHPV 3 PXVW EH IRXQG VR WKDW [73[ LV D /\DSXQRY IXQFWLRQ 7KH IROORZLQJ SUREOHP LV VWDWHG *LYHQ D VWDEOH OLQHDU V\VWHP [ $[ ILQG D SRVLWLYH GHILQLWH IXQFWLRQ 9 [73[ VR WKDW 9 [W3$ $73f[ LV D QHJDWLYH GHILQLWH IXQFWLRQ /HW $ EH SXW LQWR -RUGDQ FDQRQLFDO IRUP $ 6-6 f

PAGE 70

ZKHUH 6 LV DQ LQYHUWLEOH PDWUL[ 7KH GLDJRQDOV RI DUH WKH UHDO SDUWV RI WKH HLJHQn YDOXHV RI O DQG LPDJLQDU\ SDUWV RI WKH HLJHQYDOXHV OLH LQ VNHZ V\PPHWULF ORFDWLRQV RII RI WKH GLDJRQDOV )RU H[DPSOH LI WKH HLJHQYDOXHV RI $ DUH D s ME F WKHQ D f§E E D F /HW f§ -GLDJ aW -VNHZ f f ZKHUH -GLDJ LV WKH V\PPHWULF SDUW RI DQG -VNHZ LV WKH VNHZ V\PPHWULF SDUW RI ,Q WKH DERYH H[DPSOH D n n E n D -VNHZ f§ E F -GLDJ f§ 0DNLQJ WKH IROORZLQJ WUDQVIRUPDWLRQ RQ WKH V\VWHP [ O[ OHW [ 6] 7KHQ 6L $6] DQG f f f L f§ 6 $6] ] f /HW 3] f§-GLDJ‘ :H ZLVK WR FKHFN WKH YHORFLWLHV RI WKH V\VWHP ] -] RQ WKH HOOLSVRLG n7Ga a7U f =73]= ]7-GLDJ] 7KH QRUPDO WR WKH HOOLSVRLG DW ] LV f§ -GLDJ= 7KH SURMHFWLRQ RI ] -] RQWR WKH QRUPDO LV -GLDJ= +HUH WKH QRUPDO YHORFLW\ FRPSRQHQW KDV WKH VDPH PDJQLWXGH EXW RSSRVLWH GLUHFWLRQ WR WKH QRUPDO RI WKH HOOLSVRLG ,I WKH V\VWHP KDV QR FRPSOH[ HLJHQYDOXHV WKHQ WKH YHORFLWLHV DUH RUWKRJRQDO WR WKH HOOLSVRLG HYHU\ZKHUH )RU WKLV UHDVRQ WKH

PAGE 71

FKRLFH RI M3 -GLDJ LV WKH EHVW FKRLFH IRU D SRVLWLYH GHILQLWH IXQFWLRQ IRU WKH V\VWHP ] ] 9]f a]7-GLDJ] f 0DNLQJ WKH IROORZLQJ WUDQVIRUPDWLRQ LQWR WKH [ FRRUGLQDWH V\VWHP JLYHV ] 6 [ f ZKLFK LPSOLHV 9[f [7>6a@7 -GLDJ6nr f 2XU FKRLFH RI 3 LV 3 >6n@ -G]DJ6a f )RU WKH QRPLQDO IOLJKW FRQGLWLRQ WKH -RUGDQ FDQRQLFDO IRUP RI $T f§ %T.T DQG $SU f§ %S7.SU LV IRXQG DQG IURP f 3m 3 f§ SU f§ n f T f§ .T%M3Tf DW WKH QRPLQDO SRLQW DUH f [ /LNHZLVH IRU WKH FORVHG ORRS UROO\DZ V\VWHP WKH HLJHQYDOXHV DUH [ f 7KH ,WHUDWLYH /\DSXQRY 'HVLJQ 0HWKRG 7KH LWHUDWLYH /\DSXQRY GHVLJQ PHWKRG JHQHUDWHV IHHGEDFN JDLQV VR WKDW [73[ DQG [73SU[ DUH /\DSXQRY IXQFWLRQV IRU HDFK FORVHG ORRS V\VWHP 7KH DOJRULWKP UHTXLUHV WKH LQLWLDO JXHVVHV .TR DQG .SU IRU WKH ILUVW SRLQW DQG WKH SRVLWLYH GHILQLWH PDWULFHV 3T DQG 3S7 $V D JLYHQ IOLJKW FRQGLWLRQ FKDQJHV WKH IHHGEDFN JDLQV IURP

PAGE 72

WKH SUHYLRXV SRLQW EHFRPH WKH LQLWLDO JXHVV IRU WKH SUHVHQW SRLQW 7KH UHVXOW RI WKLV DOJRULWKP LV D VHULHV RI JDLQV RQH VHW IRU HDFK IOLJKW FRQGLWLRQ JHQHUDWHG 7KH QH[W SRUWLRQ RI WKLV VHFWLRQ GLVFXVVHV VRPH PRGLILFDWLRQV PDGH WR WKH DOJRULWKP IURP VHFWLRQ 7KH PDWHULDO ZKLFK IROORZV GHVFULEHV KRZ WKH GHVLJQ FRQVWUDLQWV DUH VHOHFWHG 7KH GHVLJQ PHWKRG RI VHFWLRQ ILQGV D VR WKDW WKH HLJHQYDOXHV RI _3$$73f IDOO EHWZHHQ Fc DQG F ZKHUH $ LV WKH FORVHG ORRS V\VWHP 7KH GHVLJQ DOJRULWKP XVHG LQ )LJXUH KDV EHHQ PRGLILHG VR WKDW WKH UHVXOWLQJ VDWLVILHV WKH IROORZLQJ UHTXLUHPHQWV 4L $L $R f§ ?>3$ %.f $ %.fW3@ ?>3^$ %.f $ %.fW3` F[ F f FQ $Q >?>3^$%.f $%.fW3`? FQ ZKHUH $L LV WKH VPDOOHVW HLJHQYDOXH $ LV WKH VHFRQG VPDOOHVW HLJHQYDOXH DQG VR RQ 7KH YDOXHV RI WKH FfV DUH VXSSOLHG E\ WKH GHVLJQHU 7KLV PRGLILFDWLRQ KDV EHHQ PDGH ZLWK WKH EHOLHI LI PRUH GHVLJQ FRQVWUDLQWV DUH PDGH WKHQ WKH SHUIRUPDQFH ZLOO YDU\ OHVV ZLWK FKDQJHV LQ WKH IOLJKW FRQGLWLRQV 7KH PRGLILHG DOJRULWKP LV DV IROORZV f &KRRVH 3 F[M FL FQI DQG FQI 7KH VHOHFWLRQ RI DQG FQc PXVW REH\ 7KHRUHP +RZHYHU 7KHRUHP GRHV QRW JXDUDQWHH WKH QRQHPSWLn QHVV RI f &RPSXWH F&LR FQFQ VR WKDW WKH LQLWLDO JXHVV .R e DQG .` & & f /HW I f /HW L f§ L f &RPSXWH .W VR WKDW .^ L DQG ƒn A $7BL f &RPSXWH FX FL FQ FQ VR WKDW .W G,&L DQG & &W

PAGE 73

f 5HSHDW VWHSV f f DQG f XQWLO RQH RI WZR HYHQWV RFFXU &MM 4LIL FX &?`L L 4QL 4QI AQG &QL f§ &QI FOW f§ FBf Ff f§ FPB f EHFRPH YHU\ VPDOO (YHQW LPSOLHV WKDW WKH ILQDO DQVZHU KDV EHHQ IRXQG (YHQW LPSOLHV WKDW ,&I LV HPSW\ DQG .^ VDWLVILHV WKH FRQVWUDLQWV FRUUHVSRQGLQJ WR FX FX FQ FQ 6WHSV f f DQG f DUH DFFRPSOLVKHG LQ WKH IROORZLQJ ZD\ 6WHSV f DQG f /HW 7KHQ HL $L ?^3$ $73 3%.L .M%U3f ?^3$ $W3 3%.L .M%73f? f FX PLQ>FOI H[@ FX PD[>FLHL@ 4QL "L=L>FQ\ HQ@ &MLL f§ f@ f 6WHS f /HW .L )RU M WR P )RU N WR Q 3L a3%4LN 47N%73f 3R 3$ %.MNf $ f§ %.DMWf73 FO .MN NMN4MN

PAGE 74

6ROYH GHW>N3L 3@ IRU LQ WHUPV RI N IRU F F FQ 5HMHFW DOO YDOXHV RI N ZKLFK GR QRW PHHW WKH FRQVWUDLQWV IURP VWHSV f RU f 7KH UHVXOW LV Q LQWHUYDOV ZKRVH ORZHU ERXQGV DUH GHVLJQDWHG E\ NO NQ DQG ZKRVH XSSHU ERXQGV LV QDPHG NL NQ )LQG WKH LQWHUVHFWLRQ RI WKHVH LQWHUYDOV E\ HYDOXDWLQJ N PD[>NDNQ@ N PmQ>NLNf@ )LQG WKH PLGSRLQW RI WKH LQWHUYDO E\ FRPSXWLQJ NMOIF LN Nf /HW f§ NMN4MN ) $RMN 1H[W N 1H[W M .L 5HPDUNV /HW WKH IHHGEDFN VHWV &O & L -& $) fff M&Q DQG .Q EH GHILQHG UHVSHFWLYHO\ DV WKH VHW RI DOO VR WKH FRQVWUDLQWV f DUH PHW 7KHRUHP SURYLGHV FRQGLWLRQV IRU WKH QRQHPSWLQHVV RI DQG O&Q %XW FRQGLWLRQV IRU WKH QRQHPSWLQHVV RI WKH UHPDLQLQJ VHWV DUH VWLOO XQNQRZQ $OVR f&Q DUH QRW FRQYH[ LQ JHQHUDO 7KHVH DUH WKH OLPLWDWLRQV RI XVLQJ WKH PRGLILHG GHVLJQ DOJRULWKP :H QRZ WXUQ WR VHOHFWLQJ FRQVWUDLQLQJ YDOXHV IRU WKH HLJHQYDOXHV RI A>3$ f§ %.fA $ f§ %.f73@ ,W LV QHFHVVDU\ WR HYDOXDWH WKH XQFRQWUROODEOH QRUPDO YHORFLW\ FRPSRn QHQWV IRU HDFK IOLJKW FRQGLWLRQ WKDW ZLOO EH JHQHUDWHG LQ )LJXUH 7DEOHV DQG VKRZ IRU ERWK PRGHOV WKH PLQLPXP DQG PD[LPXP XQFRQWUROODEOH QRUPDO YHORFLWLHV IRU HDFK FKDQJLQJ IOLJKW SDUDPHWHU

PAGE 75

7DEOH 8QFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQWV IRU WKH SLWFK PRGHO FKDQJLQJ SDUDPHWHU PLQLPXP XQFRQWUROODEOH YHORFLW\ PD[LPXP XQFRQWUROODEOH YHORFLW\ 0 4 D 3 3 U 7DEOH 8QFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQWV IRU WKH UROO\DZ PRGHO FKDQJLQJ SDUDPHWHU PLQLPXP XQFRQWUROODEOH YHORFLW\ PD[LPXP XQFRQWUROODEOH YHORFLW\ 0 4 D 3 T U )RU WKH SLWFK PRGHO WKH XQFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQWV UDQJH IURP f§ WR f§ 7KHRUHP UHTXLUHV WKDW F" f ,Q DGGLWLRQ IURP f ZH ZDQW &TO ; &TL &T &f &T f )URP WKLV WKH FRQVWUDLQLQJ YDOXHV IRU WKH SLWFK PRGHO KDYH EHHQ FKRVHQ WR EH F" [ TFT [ f &T f§ &T f§ f§ /LNHZLVH IRU WKH UROO\DZ PRGHO ZKHQ ORRNLQJ DW 7DEOH 7KHRUHP UHTXLUHV WKDW &SU 6SUO f

PAGE 76

7KH HLJHQYDOXHV LQ f VXJJHVW WKH IROORZLQJ 4SUO [ &SUO f§SU &SU f§SU &SU f 7KH IROORZLQJ FRQVWUDLQLQJ YDOXHV KDYH EHHQ FKRVHQ IRU WKH UROO\DZ PRGHO FSUO FS7O FSU FSW f 4SW] f§ &SU f§ )RUPXODWLRQ RI D 6WDWH 7UDFNHU 7KH DXWRSLORW RI WKH (05$$7 PLVVLOH ZLOO EH D VWDWH WUDFNHU 7KDW LV ZH ZDQW WR EH DEOH WR FKDQJH WKH ORFDWLRQ RI WKH HTXLOLEULXP SRLQW LQ RUGHU WR FRQWURO WKH YDOXHV RI VRPH RI WKH VWDWHV 7KH IROORZLQJ VKRZV KRZ WKLV ZLOO EH DFFRPSOLVKHG *LYHQ WKH OLQHDU V\VWHP [ $[ %X f \ &[ f ZH ZRXOG OLNH WR ILQG D FRQWURO ODZ X f§.[ $7H9 f VR WKDW \ WKH RXWSXW WUDFNV Y WKH UHIHUHQFH LQSXW DV\PSWRWLFDOO\ :H UHTXLUH \ Y ZKHQ [ :KHQ [ WKHQ $[ f§ %.[ %.7HI9 f 6LQFH LV FKRVHQ VR WKDW WKH V\VWHP LV VWDEOH WKHQ $ f§ %. LV LQYHUWLEOH DQG [ $ %.fO%.UH6Y f $OVR 9 \ &[ f§&$ %.faO%.UHLY f

PAGE 77

%HFDXVH f LV WUXH IRU DQ\ Y WKHQ f§&$ %.faO%.UHI f &RQWUROODELOLW\ RI WKH V\VWHP LPSOLHV WKDW &$f§ %.f % LV LQYHUWLEOH 6ROYLQJ f IRU .UHI JLYHV .UHI a>&$ %.fO%@aO f 7KH (0 5 $ $7 PLVVLOH KDV WKUHH LQSXWV DQG WKHUHIRUH RQO\ WKUHH VWDWHV FDQ EH WUDFNHG &RQWUROOLQJ D LQ WKH SLWFK PRGHO DQG c DQG S LQ WKH UROO \DZ PRGHO LV GHVLUDEOH )RU WKH SLWFK PRGHO \ D LPSO\LQJ WKDW &J > @ f DQG WKHUHIRUH .UHIT a>&T$T %T.Tf %T@aO f )RU WKH UROO\DZ PRGHO DQG WKXV %FISU f§ >&UfUI $SU %SUKSUf % SU f f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r 6R ZH VHW DR HTXDO WR r

PAGE 78

7DEOH ([WUHPH YDOXHV DQG UDQJH RI WKH SLWFK FKDQQHO FRQWURO WHUPV DV LQGLYLGXDO IOLJKW SDUDPHWHUV YDU\ *DLQV GHSHQG PRVWO\ RQ 0 4 DQG D 0 4 D 3 U FRQWURO PLQ PLQ PLQ PLQ PLQ PLQ PLQ WHUP PD[ PD[ PD[ PD[ PD[ PD[ PD[ GLII GLII GLII GLII GLII GLII GLII NTQDR NTX4R NUHI TOOAF2 DQG ORRN DW NTQFLR :H VHW T DQG HTXDO WR rV DQG r UHVSHFWLYHO\ VR WKDW ZH FDQ ORRN DW NTLT DQG NUHI QDF 7KH VXP RI WKHVH WKUHH WHUPV DUH IHG LQWR WR WKH HOHYDWRU )RU WKH WHUPV ZKLFK DUH IHG LQWR WKH UHPDLQLQJ LQSXWV WKH IROORZLQJ DVVLJQPHQWV DUH PDGH b tR r 3R 3FR OrV f U r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

PAGE 79

NTO fDOSKD2 KTrDOSKD2 NTLUDMSKD2 f§ A

PAGE 80

7DEOH ([WUHPH YDOXHV DQG UDQJH RI WKH UROO\DZ FKDQQHO FRQWURO WHUPV DV LQGLn YLGXDO IOLJKW SDUDPHWHUV YDU\ *DLQV GHSHQG PRVWO\ RQ 0 4 DQG D 0 4 D 3 3 U FRQWURO PLQ PLQ PLQ PLQ PLQ PLQ PLQ WHUP PD[ PD[ PD[ PD[ PD[ PD[ PD[ GLII GLII GLII GLII GLII GLII GLII NSUOOILR IFSUO34 ASUOA NSU OA NSU3R NSU A NUHI SUOOAF2 NUHI SUO3F2 2 2 &2 NUHI SUOILF NUHI SU3F2

PAGE 81

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

PAGE 82

7DEOH 9DOXHV RI D LQ WKH SLWFK FRQWURO D 3LWFK 5ROO
PAGE 83

SU )LJXUH )RUPXODWLRQ RI WKH 'HVLJQ &RQVWUDLQWV 7DEOH 8QFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQWV IRU WKH SLWFK PRGHO DGHJUHHVf PLQLPXP XQFRQWUROODEOH YHORFLW\ PD[LPXP XQFRQWUROODEOH YHORFLW\ 7DEOH 8QFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQWV IRU WKH UROO\DZ PRGHO DGHJUHHVf PLQLPXP XQFRQWUROODEOH YHORFLW\ PD[LPXP XQFRQWUROODEOH YHORFLW\

PAGE 84

DW WKH QRPLQDO IOLJKW FRQGLWLRQ IURP WKH SUHYLRXV FKDSWHU ZHUH IRXQG WR EH f§ [ DQG f§ 7KHRUHP UHTXLUHV WKDW FT FTO f f %XW ZH DOVR ZDQW FTO f§ [ &TL FT f§ FT VLQFH ZH GHVLUH WKH HLJHQYDOXHV RI f WR EH FORVH WR WKRVH DW WKH QRPLQDO IOLJKW FRQGLWLRQ )URP WKLV WKH FRQVWUDLQLQJ YDOXHV ZHUH FKRVHQ WR EH &TO ; &TO FT &T /LNHZLVH IRU WKH UROO\DZ PRGHO WKH XQFRQWUROODEOH QRUPDO YHORFLW\ FRPSRQHQWV UDQJH IURP f§ WR f§ )URP 7KHRUHP ZH PXVW KDYH f§ FSU FSQ f :LWK WKH HLJHQYDOXHV RI 3S7 $SU % SU .SU f fK ASU %SU .S7f7 3S7@ FRQGLWLRQ EHLQJ f§ [ DQG f§ WKH f§SUO [ &SU f§SU &SU f /SU &SU :LWK WKLV LQ PLQG WKH IROORZLQJ VHOHFWLRQV ZHUH PDGH FSUL f§ FSUL f§ &SU &SU f§SU f§ &S7

PAGE 85

*HQHUDWLQJ WKH /RRN8S 7DEOH :LWK WKH GHVLJQ FRQVWUDLQWV VHW IHHGEDFN JDLQV DQG IHHGIRUZDUG JDLQV DUH JHQn HUDWHG IRU HDFK JULG SRLQW )LJXUH JLYHV D EORFN GLDJUDP RI WKH V\VWHP XVHG WR DFFRPSOLVK WKLV )RU WKH SLWFK PRGHO WKH LQLWLDO JXHVV FRPHV IURP RQH RI WKH JDLQV WKDW ZDV JHQHUDWHG LQ WKH SUHYLRXV FKDSWHU ZKHQ GHWHUPLQLQJ WKH GHSHQGHQFH EHn WZHHQ JDLQV DQG IOLJKW SDUDPHWHUV 7KH RSHUDWLQJ SRLQW IURP ZKLFK WKLV LQLWLDO JXHVV RULJLQDWHV LV 0 4 SVI D r f DQG LV RQH RI WKH H[WUHPH YDOXHV OLVWHG LQ 7DEOH 7KH UHVXOW RI WKH ILUVW SRLQW LV XVHG WR VWDUW DGMDFHQW SRLQWV ZKLFK LQ WXUQ VWDUW QHZ DGMDFHQW SRLQWV XQWLO JDLQV KDYH EHHQ FRPSXWHG IRU WKH HQWLUH JULG 7KH ORRNXS WDEOH IRU WKH UROO\DZ PRGHO LV PDGH LQ WKH VDPH ZD\ 1XPHULFDO SUREOHPV ZHUH HQFRXQWHUHG LQ SDUWV RI WKH ORRNXS WDEOH IRU WKH UROO \DZ PRGHO 7KH\ ZHUH VLPLODU WR WKH SUREOHPV WKDW ZHUH SUHGLFWHG LQ VWHS f RI WKH LWHUDWLYH GHVLJQ PHWKRG JLYHQ LQ &KDSWHU 7R RYHUFRPH WKHVH GLIILFXOWLHV VRPH RI WKH FRQVWUDLQLQJ YDOXHV ZHUH UHOD[HG IRU D QXPEHU RI LWHUDWLRQV DQG ZHUH ODWHU UHWXUQHG WR WKHLU RULJLQDO DVVLJQPHQWV LQ WKH DOJRULWKP (YHQWXDOO\ WKH GHVLUHG IHHGEDFN PDWUL[ ZDV IRXQG )LJXUH VKRZV NTX YHUVHV PDFK QXPEHU DQG G\QDPLF SUHVVXUH ZKHQ D r $V WKLV ILJXUH ZRXOG LQGLFDWH WKH JDLQV JHQHUDWHG IURP WKH LWHUDWLYH /\DSXQRY PHWKRG DUH VPRRWK ZLWK UHVSHFW WR WKH GHSHQGHQW IOLJKW SDUDPHWHUV 7KLV IDFW JLYHV KRSH WKDW WKH JDLQ VFKHGXOLQJ VFKHPH ZLOO EH HDV\

PAGE 86

*DLQ 6FKHGXOHV )LJXUH )RUPXODWLRQ RI WKH ORRNXS WDEOH

PAGE 87

! )LJXUH NT8 YHUVHV 0 DQG 4 ZKHQ D r

PAGE 88

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r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

PAGE 89

0 0 )LJXUH $ SORW RI WKH NTX YHUVHV 0 ZLWK Df YDULRXV FRQVWDQW YDOXHV RI 4 DQG D r DQG Ef D OHDVW VTXDUHV WKLUG RUGHU SRO\QRPLDO ILW IRU WKH SORW ZKHUH 4 SVI DQG D r RI DGMDFHQW JULG SRLQWV /LQHDU PRGHOV ZHUH DOVR PDGH DW HDFK WHVW SRLQW ZKHUH S T DQG U ZHUH VHW WR ]HUR 7KH HLJHQYDOXHV RI ?>3$ a %.f $ %.I3` f DQG A $SU 6SnS-?SUnM aI %SU .SU f 3SU @ f ZHUH FRPSXWHG WR VHH LI WKH\ UHPDLQHG ZLWKLQ WKH GHVLUHG OLPLWV 7DEOH VKRZV WKH PD[LPXP HLJHQYDOXH RI f IRU DOO RI WKH WHVW SRLQWV DW D r 7KHVH YDOXHV DUH SORWWHG DJDLQVW WKHLU LQGLFHV LQ )LJXUH 0RVW RI WKH HLJHQYDOXHV RI 7DEOH OLH ZLWKLQ WKH GHVLUHG OLPLWV RI f§ DQG f§ 7KH HLJHQYDOXH LQ VHYHQWHHQWK URZ VHFRQG FROXPQ KRZHYHU LV f§ WKH ZRUVW YDOXH IRXQG RXW RI DOO RI WKH WHVW SRLQWV $OWKRXJK WKH GHYLDWLRQ LV KLJK WKLV YDOXH LV VWLOO QHJDWLYH LQGLFDWLQJ VWDELOLW\ IRU WKH FORVHG ORRS V\VWHP )RU WKH SLWFK FRQWUROOHU WKH DFWXDO OLPLWLQJ YDOXHV RI DQG $ DW DOO WKH WHVW SRLQWV DUH [ $L [ $" f

PAGE 90

)LJXUH $ WHVW RI WKH FXUYH ILWWLQJ URXWLQHV XVHG WR LPSOHPHQW WKH DXWRSLORW

PAGE 91

7DEOH 7KH PD[LPXP HLJHQYDOXHV RI f IRU DOO WHVW SRLQWV DW D f§ r 40 LQGLFHV )LJXUH $ SORW RI WKH HLJHQYDOXHV IURP 7DEOH

PAGE 92

/LNHZLVH IRU WKH UROO\DZ FRQWUROOHU WKH OLPLWLQJ YDOXHV DW DOO WKH WHVW SRLQWV DUH [ V ;SUO [ n ;SU ;SU f 6RPH RI WKHVH YDOXHV GLIIHU VLJQLILFDQWO\ IURP WKH GHVLUHG FRQVWUDLQWV KRZHYHU VLQFH WKHVH YDOXHV DUH VWLOO QHJDWLYH WKHVH GHYLDWLRQV DUH DFFHSWDEOH DQG LQGLFDWH WKDW WKH FORVHG ORRS V\VWHP ZLOO EH VWDEOH ,W VKRXOG DOVR EH QRWHG WKDW PRVW HLJHQYDOXHV UHPDLQ ZHOO ZLWKLQ WKHLU GHVLUHG FRQVWUDLQWV DV VKRZQ LQ 7DEOH

PAGE 93

&+$37(5 121/,1($5 6,08/$7,216 $ QRQOLQHDU VLPXODWLRQ KDV EHHQ XVHG WR WHVW WKH SURSRVHG DXWRSLORW IRU WKH (0 5$$7 PLVVLOH )LUVW D VHFWLRQ IROORZV JLYLQJ DQ RYHUYLHZ RI WKH QRQOLQHDU VLPXODWLRQ $ WHVW PRGXOH LV WKHQ PDGH WR JHQHUDWH VWDWH FRPPDQGV LQ RUGHU WR HYDOXDWH WKH DXWRSLORWf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

PAGE 94

)LJXUH $ %ORFN 'LDJUDP RI WKH 1RQOLQHDU 0LVVLOH 6LPXODWLRQ

PAGE 95

D9F ZLOO EHFRPH VPDOO DQG D=F ZLOO EHFRPH SRVLWLYH DF DQG & DUH FRPSXWHG LQ DQ DWWHPSW WR PDWFK D=F DQG D9F UHVSHFWLYHO\ 7KH DXWRSLORW LPSOHPHQWV WKH FRQWURO ODZ 8 f§ .; .UHI9 f ZKHUH [ LV D YHFWRU FRQWDLQLQJ WKH DFWXDO VWDWHV DQG Y FRQWDLQV WKH VWDWH FRPPDQGV IURP WKH EDQNWRWXUQ %77f ORJLF 7KH VWDWHV FRPH IURP H[DFW PHDVXUHPHQWV LQ WKH VLPXODWLRQ ,I WKLV DXWRSLORW ZHUH WR EH LPSOHPHQWHG LQ DQ DFWXDO PLVVLOH WKH VWDWHV ZRXOG EH PHDVXUHG XVLQJ DQ LQHUWLDO SODWIRUP 7KH JDLQV DQG .UHI FRPH IURP WKH JDLQ VFKHGXOH LPSOHPHQWHG ZLWK D FRPELQDWLRQ RI SRO\QRPLDOV DQG LQWHUSRODWLRQ ,Q WKLV VLPXODWLRQ WKHUH LV QR GHOD\ LQ WKH JDLQ VFKHGXOH DQG DQG .UHc DUH SURGXFHG LQVWDQWDQHRXVO\ 7KH RXWSXW RI WKH DXWRSLORW LV WKH FRQWURO VXUIDFH DQJOHV 6S T DQG U /LQHDU DQG DQJXODU DFFHOHUDWLRQV DUH FRPSXWHG E\ WKH PLVVLOH G\QDPLFV PRGXOH RI WKH SURJUDP 7KH VLPXODWLRQ XVHV WKH RXWSXW RI WKH PLVVLOH G\QDPLFV WR FRPSXWH DOO RI WKH IOLJKW YDULDEOHV LQFOXGLQJ WKH SRVLWLRQ DQG YHORFLW\ RI WKH PLVVLOH $ 7HVW RI 6WDWH 7UDFNLQJ 7KH PRGHO IRU WKH (05$$7 PLVVLOH KDV ILYH VWDWHV DQG WKUHH LQSXWV 7KH DXWRSLn ORW LV GHVLJQHG WR WUDFN WKUHH VWDWH FRPPDQGV DF & DQG SF %HIRUH UXQQLQJ PLVVLOH WDUJHW VFHQDULRV LW ZDV GHFLGHG WR WHVW WKH DXWRSLORWfV WUDFNLQJ DELOLW\ 7KH %77 ORJLF ZDV GLVFRQQHFWHG DQG WKH IROORZLQJ FRPPDQGV ZHUH DSSOLHG WR WKH UHIHUHQFH LQSXWV RI WKH DXWRSLORW r IRU V W V B r IRU V W V DF a r IRU V W V r IRU V W r IRU V W V & r IRU V W V r IRU V W f

PAGE 96

7DEOH 7KH ULVH WLPHV RI HDFK FRPPDQGHG VWDWH $OWLWXGHIWf 0DFK bVf LUVf USVf 722 rV IRU V W V b IRU V W V b IRU V W V b IRU V W V b IRU V W V b IRU V W f 7KH H[SHULPHQW ZDV SHUIRUPHG RQFH DW DQ LQLWLDO DOWLWXGH RI IW ZLWK DQ LQLWLDO PDFK QXPEHU RI DQG D VHFRQG WLPH DW DQ LQLWLDO DOWLWXGH RI IW ZLWK DQ LQLWLDO PDFK QXPEHU RI )LJXUH VKRZV WKH UHVXOWV RI WKH ILUVW WHVW ,Q DGGLWLRQ WR WKH FRPPDQGHG VWDWHV ILJXUH DOVR VKRZV WKH UHPDLQLQJ WZR VWDWHV T DQG U 7KH ULVH WLPH DV GHILQHG DV WKH WLPH QHHGHG WR DFKLHYH b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r LV URWDWHG LQWR VLGHVOLS 7KH VDPH HIIHFWV DUH SUHVHQW LQ WKH VHFRQG H[SHULPHQW DW IW )LJXUH G DQG H VKRZ WKH WZR VWDWHV WKDW KDYH QR FRPPDQGV 7KH VSLNHV LQ WKH SLWFK UDWH RFFXU ZKHQ DF FKDQJHV YDOXH EHFDXVH D SLWFKLQJ PDQHXYHU LV UHTXLUHG WR FKDQJH WKH DQJOH RI DWWDFN 7KH WZR VSLNHV LQ WKH \DZ UDWH RFFXU IRU WKH VDPH

PAGE 97

)LJXUH $ WHVW RI VWDWH WUDFNLQJ DW DQ LQLWLDO DOWLWXGH RI IW ZLWK DQ LQLWLDO PDFK QXPEHU RI 7KH VWDWHV VKRZQ DUH Df D Ef Ff S Gf T DQG Hf U

PAGE 98

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n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f§r OLPLW DV ILJXUH F LQGLFDWHV

PAGE 99

Df 7RS 9LHZ RI 0LVVLOH f DQG 7DUJHW f§f 7UDMHFWRU\ )LJXUH Df 7RS YLHZ DQG Ef VLGH YLHZ RI PLVVLOH DQG WDUJHW WUDMHFWRULHV RI D VFHQDULR ZKLFK RFFXUUHG DW IW 7KH PLVV GLVWDQFH LV IW ,QLWLDO PDFK QXPEHUV RI WKH WDUJHW DQG PLVVLOH DUH DQG UHVSHFWLYHO\

PAGE 100

Df &RPPDQGHG ] $FFHOHUDWLRQ f DQG $FWXDO ] $FFHOHUDWLRQ f§f )LJXUH &RPPDQGHG DQG DFWXDO Df ] DFFHOHUDWLRQV DQG Ef \ DFFHOHUDWLRQV IRU WKH VFHQDULR LQ ILJXUH

PAGE 101

Df &RPPDQGHG $OSKD f DQG $FWXDO $OSKD f )LJXUH &RPPDQGHG DQG DFWXDO VWDWHV Df D Ef DQG Ff S IRU WKH VFHQDULR LQ ILJXUH

PAGE 102

Df 7RS 9LHZ RI 0LVVLOH f DQG 7DUJHW f§f 7UDMHFWRU\ )LJXUH Df 7RS YLHZ DQG Ef VLGH YLHZ RI PLVVLOH DQG WDUJHW WUDMHFWRULHV RI D VFHQDULR ZKLFK RFFXUUHG DW IW 7KH PLVV GLVWDQFH LV IW ,QLWLDO PDFK QXPEHUV RI WKH WDUJHW DQG PLVVLOH DUH DQG UHVSHFWLYHO\

PAGE 103

Df 7RS 9LHZ RI 0LVVLOH f DQG 7DUJHW f§f 7UDMHFWRU\ )LJXUH Df 7RS YLHZ DQG Ef VLGH YLHZ RI PLVVLOH DQG WDUJHW WUDMHFWRULHV RI D VFHQDULR ZKLFK RFFXUUHG DW IW 7KH PLVV GLVWDQFH LV IW ,QLWLDO PDFK QXPEHUV RI WKH WDUJHW DQG PLVVLOH DUH DQG UHVSHFWLYHO\

PAGE 104

7KH PLVV RFFXUUHG GXH WR WKH UHGXFHG HOHYDWRU HIIHFWLYHQHVV IURP WKH ORZ DLU GHQVLW\ DW KLJK DOWLWXGHV $QRWKHU PLVV LV VKRZQ LQ ILJXUH )LJXUH VKRZV WKDW WKH GHVLUHG ] DFn FHOHUDWLRQ ZDV QRW DFKLHYHG EHFDXVH WKH FRPPDQGHG DQJOH RI DWWDFN ZDV DW LWV r OLPLW 7KH FRPPDQGHG QRUPDO DFFHOHUDWLRQ ZDV WRR JUHDW WR EH UHDOL]HG $JDLQ WKH DXWRSLORW SHUIRUPHG DV ZHOO DV LW FRXOG EXW WKH PLVV RFFXUUHG EHFDXVH RI SK\VLFDO OLPLWDWLRQV RI WKH PLVVLOH 7KH UHVXOWV RI WKHVH VLPXODWLRQV LQGLFDWH WKDW WKH DXWRSLORW GRHV LWV MRE DQG LQ UHDVRQDEOH VFHQDULRV DOORZV WKH PLVVLOH WR LQWHUFHSW WKH WDUJHW

PAGE 105

Df 7RS 9LHZ RI 0LVVLOH f DQG 7DUJHW f 7UDMHFWRU\ )LJXUH Df 7RS YLHZ DQG Ef VLGH YLHZ RI PLVVLOH DQG WDUJHW WUDMHFWRULHV RI D VFHQDULR ZKLFK RFFXUUHG DW IW 7KH PLVVLOH PLVVHG WKH WDUJHW E\ IW ,QLWLDO PDFK QXPEHUV RI WKH WDUJHW DQG PLVVLOH DUH DQG UHVSHFWLYHO\

PAGE 106

Df &RPPDQGHG ] $FFHOHUDWLRQ f DQG $FWXDO ] $FFHOHUDWLRQ f§f )LJXUH &RPPDQGHG DQG DFWXDO Df DFFHOHUDWLRQV LQ WKH GLUHFWLRQ Ef FRPn PDQGHG DQG DFWXDO DQJOHV RI DWWDFN DQG Ff WKH HOHYDWRU GHIOHFWLRQ DQJOH IRU WKH VFHQDULR VKRZQ LQ ILJXUH 6DWXUDWLRQ RI WKH HOHYDWRU DQJOH ZDV WKH FDXVH RI WKH PLVV

PAGE 107

Df 7RS 9LHZ RI 0LVVLOH f DQG 7DUJHW f§f 7UDMHFWRU\ )LJXUH Df 7RS YLHZ DQG Ef VLGH YLHZ RI PLVVLOH DQG WDUJHW WUDMHFWRULHV RI D VFHQDULR ZKLFK RFFXUUHG DW IW 7KH WDUJHW ZDV PLVVHG IW ,QLWLDO PDFK QXPEHUV RI WKH WDUJHW DQG PLVVLOH DUH DQG UHVSHFWLYHO\

PAGE 108

Df &RPPDQGHG ] $FFHOHUDWLRQ f DQG $FWXDO ] $FFHOHUDWLRQ f Ff (OHYDWRU 'HIOHFWLRQ $QJOH )LJXUH &RPPDQGHG DQG DFWXDO DFFHOHUDWLRQV Df LQ WKH GLUHFWLRQ Ef FRPn PDQGHG DQG DFWXDO DQJOHV RI DWWDFN DQG Ff HOHYDWRU GHIOHFWLRQ DQJOH IRU WKH VFHQDULR VKRZQ LQ ILJXUH 6DWXUDWLRQ RI WKH D FRPPDQG UHVXOWHG LQ WKH PLVV

PAGE 109

&+$37(5 &21&/86,21 $ FRQWUROOHU GHVLJQ DOJRULWKP KDV EHHQ PDGH ZKLFK FDQ VWDELOL]H WLPH YDU\LQJ OLQHDU V\VWHPV 7KH PRWLYDWLRQ EHKLQG WKLV ZRUN ZDV WKH QHHG WR DFKLHYH ORFDO VWDn ELOLW\ RI D QRQOLQHDU V\VWHP E\ VWDELOL]LQJ D OLQHDUL]HG PRGHO ZKLFK LV D IXQFWLRQ RI WLPH YDU\LQJ V\VWHP SDUDPHWHUV 7KLV ZRUN EHJDQ E\ VWXG\LQJ VHFRQG RUGHU OLQHDU WLPH YDU\LQJ V\VWHPV WKDW ZHUH VWDEOH ZKHQ IUR]HQ DW DQ\ JLYHQ WLPH EXW ZHUH XQn VWDEOH ZKHQ DOORZHG WR YDU\ ZLWK WLPH *HRPHWULF WHFKQLTXHV ZHUH XVHG WR UHGHULYH /\DSXQRYf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n OLQHDU VLPXODWLRQ 7KH PLVVLOH ZDV DEOH WR LQWHUFHSW WKH WDUJHW LQ UHDVRQDEOH VFHQDULRV DQG IOLJKW FRQGLWLRQV 7KH VLJQLILFDQFH RI WKH SURSRVHG GHVLJQ DOJRULWKP LV WKDW QR DVVXPSWLRQ LV PDGH RQ WLPH LQYDULDQFH RI WKH V\VWHP +RZHYHU DV VWDWHG LQ FKDSWHU WKLV DOJRULWKP FDQ RQO\ EH XVHG WR VWDELOL]H D V\VWHP LI WKHUH H[LVWV D FRQVWDQW SRVLWLYH GHILQLWH 3 VXFK WKDW f>$f3B 3aO$7Wf@%M/Wf LV QHJDWLYH GHILQLWH IRU DOO WLPH ,I WKLV LV WUXH WKHQ D .Wf FDQ EH IRXQG VR WKDW [73[ LV D /\DSXQRY IXQFWLRQ IRU WKH WLPH

PAGE 110

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n>Wf>$Wf3a 3aO $7Wf@%MBWf LV SRVLWLYH GHILQLWH )XUWKHU VWXG\ LQ WKLV DUHD ZRXOG EH ZDUUDQWHG

PAGE 111

$33(1',; $ $(52'<1$0,& '$7$ )25 7+( (0 5 $ $7 $,5)5$0( 7KLV DSSHQGL[ JLYHV DHURG\QDPLF GDWD IRU WKH H[WHQGHG PHGLXP UDQJH DLUWRDLU WHFKQRORJ\ (05$$7f DLUIUDPH 7KH DHURG\QDPLF FRHIILFLHQWV DUH JLYHQ LQ WDEXODU IRUP $OO FRHIILFLHQWV WKDW FRUUHVSRQG WR DQJOH UDWHV KDYH QR GLPHQVLRQV DQG PXVW EH PXOWLSOLHG E\ $OO FRHIILFLHQWV WKDW FRUUHVSRQG WR DQJOH SRVLWLRQV DUH JLYHQ LQ SHU GHJUHHV DQG PXVW EH PXOWLSOLHG E\ A 0LVVLOH 5HIHUHQFH 'LDPHWHU 0LVVLOH 5HIHUHQFH $UHD G IW 6 IW 7DEOH $O 7DEXODU GDWD IRU &Y D0 r r r r r r r r r

PAGE 112

7DEOH $ 7DEXODU GDWD IRU &P D0 r r r r r r r r 2 &0 7DEOH $ 7DEXODU GDWD IRU &AT D0 r r r r r r r r r

PAGE 113

7DEOH $ 7DEXODU GDWD IRU & n D0 r r r r r r r r R 2 7DEOH $ 7DEXODU GDWD IRU &P D0 r r r r r r r r R 2

PAGE 114

7DEOH $ 7DEXODU GDWD IRU &\ D0 r r r r r r r r r 7DEOH $ 7DEXODU GDWD IRU &QU D0 r r r r r r r r r

PAGE 115

7DEOH $ 7DEXODU GDWD IRU &LU D0 r r r r r r r r R 2 0 7DEOH $ 7DEXODU GDWD IRU &\6U D0 r r r r r r r r R 2

PAGE 116

7DEOH $ 7DEXODU GDWD IRU &L D0 r r r r r r r r r 7DEOH $OO 7DEXODU GDWD IRU &c6S D0 r r r r r r r r r

PAGE 117

7DEOH $ 7DEXODU GDWD IRU & D0 r r r r r r r r R 2 7DEOH $ 7DEXODU GDWD IRU &\6S D0 r r r r r r r r r

PAGE 118

7DEOH $ 7DEXODU GDWD IRU &c RO0 r r r r r r r r r 7DEOH $ 7DEXODU GDWD IRU &\S D0 r r r r r r r r R 2

PAGE 119

7DEOH $ 7DEXODU GDWD IRU &8S D0 r r r r r r r r r 7DEOH $ 7DEXODU GDWD IRU &LY 0 FQT 7DEOH $ 7DEXODU GDWD IRU &PT 0 &PT

PAGE 120

7DEOH $ 7DEXODU GDWD IRU &Y£ 0 &1 7DEOH $ 7DEXODU GDWD IRU &P£ 0 FP 7DEOH $ 7DEXODU GDWD IRU &\U 0 &\U

PAGE 121

7DEOH $ 7DEXODU GDWD IRU &8U 0 FQU 7DEOH $ 7DEXODU GDWD IRU &?S 0 &LS

PAGE 122

$33(1',; % ,1(57,$/ '$7$ )25 7+( (05$$7 $,5)5$0( 7KLV DSSHQGL[ JLYHV WKH ZHLJKW DQG PRPHQWV DQG SURGXFWV RI LQHUWLD RI WKH (0 5$$7 PLVVLOH ZLWK QR IXHO 0LVVLOH :HLJKW : OE 0RPHQWV RI ,QHUWLD ,[[ VOXJIW ,\\ VOXJIW ,]] VOXJIW 3URGXFWV RI ,QHUWLD ,[\ VOXJIW ,[] VOXJIW ,\] f§ f§ VOXJIW

PAGE 123

5()(5(1&(6 >@ 5REHUW ) :LOVRQ -DPHV 5 &ORXWLHU DQG 5 @ 5RJHU / 6PLWK $Q $XWRSLORW 'HVLJQ 0HWKRGRORJ\ IRU %DQNWR7XUQ 0LVVLOHV $LU )RUFH $UPDPHQW /DERUDWRU\ (JOLQ $)% $XJXVW >@ -HII 6 6KDPPD DQG 0LFKDHO $WKDQV f$QDO\VLV RI *DLQ 6FKHGXOHG &RQWURO IRU 1RQOLQHDU 3ODQWVf ,((( 7UDQVDFWLRQV RQ $XWRPDWLF &RQWURO 9RO 1R SS >@ -HII 6 6KDPPD DQG 0LFKDHO $WKDQV f*DLQ 6FKHGXOLQJ 3RWHQWLDO +D]DUGV DQG 3RVVLEOH 5HPHGLHVf ,((( &RQWURO 6\VWHPV 0DJD]LQH 9RO 1R SS >@ -HII 6 6KDPPD DQG 0LFKDHO $WKDQV f6WDELOLW\ DQG 5REXVWQHVV RI 6ORZO\ 7LPH 9DU\LQJ /LQHDU 6\VWHPVf 3URFHHGLQJV RI WKH WK &RQIHUHQFH RQ 'HFLVLRQ DQG &RQWURO /RV $QJHOHV 'HFHPEHU >@ -HII 6 6KDPPD DQG -DPHV 5 &ORXWLHU f*DLQ6FKHGXOHG 0LVVLOH $XWRSLORW 'Hn VLJQ 8VLQJ /LQHDU 3DUDPHWHU 9DU\LQJ 7UDQVIRUPDWLRQVf -RXUQDO RI *XLGDQFH &RQWURO DQG '\QDPLFV 9RO 1R SS >@ .HYLQ $ :LVH %DUU\ & 0HDUV DQG .DPHVKZDU 3RROOD f0LVVLOH $XWRSLORW 'Hn VLJQ 8VLQJ +RR 2SWLPDO &RQWURO ZLWK 6\QWKHVLVf 3URFHHGLQJV RI WKH $PHULFDQ &RQWURO &RQIHUHQFH 9RO 6DQ 'LHJR >@ 0DULR $ 5RWHD DQG 3UDPRG 3 .KDUJRQHNDU f6LPXOWDQHRXV ++ 2SWLPDO &RQWURO ZLWK 6WDWH )HHGEDFNf 3URFHHGLQJV RI WKH $PHULFDQ &RQWURO &RQIHUHQFH 9RO 6DQ 'LHJR >@ &KDUOHV $ 'HVRHU f6ORZO\ 9DU\LQJ 6\VWHP [ $If[f ,((( 7UDQVDFWLRQV RQ $XWRPDWLF &RQWURO 9RO $& SS 'HFHPEHU >@ 0 9LG\DVDJDU 1RQOLQHDU 6\VWHPV $QDO\VLV 3UHQWLFH+DOO (QJOHZRRG &OLIIV 1>@ :ROIJDQJ +DKQ 6WDELOLW\ RI 0RWLRQ 6SULQJHU9HUODJ 1HZ @ 6 0 6KDKUX] DQG 6 %HKWDVK f'HVLJQ RI &RQWUROOHUV IRU /LQHDU 3DUDPHWHU 9DU\LQJ 6\VWHPV E\ WKH *DLQ 6FKHGXOLQJ 7HFKQLTXHf -RXUQDO RI 0DWKHPDWLFDO $QDO\VLV DQG $SSOLFDWLRQV 9RO SS -XO\

PAGE 124

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
PAGE 125

%,2*5$3+,&$/ 6.(7&+ .XUW :DOWHU .RHQLJ ZDV ERUQ RQ -XO\ LQ 5ROOD 0LVVRXUL +H JUDGXDWHG ZLWK KLJK KRQRUV IURP WKH 8QLYHUVLW\ RI )ORULGD ZLWK D EDFKHORUfV GHJUHH LQ HOHFWULFDO HQJLQHHULQJ LQ 'HFHPEHU RI +H UHFHLYHG D )ORULGD *UDGXDWH 6FKRODUVKLS LQ WR SXUVXH D PDVWHUfV GHJUHH LQ HOHFWULFDO HQJLQHHULQJ DW WKH 8QLYHUVLW\ RI )ORULGD ZKLFK KH UHFHLYHG LQ 'HFHPEHU RI +H ZDV DGPLWWHG LQWR WKH 3DODFH .QLJKW SURJUDP LQ DQG UHFHLYHG VXSSRUW IURP WKH 8QLWHG 6WDWHV $LU )RUFH WR ZRUN RQ D 3K' LQ HOHFWULFDO HQJLQHHULQJ DW WKH 8QLYHUVLW\ RI )ORULGD

PAGE 126

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

PAGE 127

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


27
Remarks:
The nonemptiness of the intersection 1C = 1C fl 1C is not guaranteed. If the designer
discovers that no intersection exists, then the upper and lower velocity bounds will
have to be adjusted.
Another useful property of 1C is convexity. This property is valuable in formulating
iterative search techniques to be described in the next section. 1C is convex if when
x and y are elements of 1C, then ax + (1 a)y is also an element of 1C for 0 < a < 1
[16]-
Theorem 3 Let K- be the set of all K so that
Ama*[\[P{A BK) + (A BK)tP}} < c (2.61)
Let JC be the set of all K so that
Amin[^[P(A BK) + (A BKfP]] > c (2.62)
Then /C, 1C, and 1C = 1C fl 1C are convex.
Proof:
If 1C and 1C are convex, then the intersection 1C is convex. Convexity of
1C will be proven here. In a similar manner the proof for the convexity of
1C can be written.
Let Ki and K2 be elements of 1C. We must show that aK\ + (1 a)K2
is contained in 1C when 0 < a < 1. We know that
^xt[PA + ATP PBKi K*BtP]x < c V xTx = 1 (2.63)
^xt[PA + ATP PBI<2 /i2rPrP]x < c V xTx = 1 (2.64)
and


CHAPTER 3
A TIME VARYING SECOND ORDER EXAMPLE
The following problem gives a case when pole placement succeeds in giving eigen
values with negative real parts, but fails to stabilize the system. The Lyapunov
design method is then employed, and the resulting closed loop system is shown to be
asymptotically stable for all time.
We would like to find a feedback control law that stabilizes the system
x = A(i)x + B(t) u
(3.1)
where
A(t) =
1 + 1.5cos()[cos() + cos( + 7T /18)] 1 1.5 sin()[cos() + cos (t + tt/18)]
1 1.5cos()[sin() + sin( + 7r/18)j 1 + 1.5sin(t)[sin(i) + sin(t + tt/18)]
(3.2)
and
B(t) =
COS it + 7r/l8)
sin(t + 7t/18)
(3.3)
The eigenvalues of A(t) are -.4889 and 1.4661 for all time. The following control law
is proposed.
u = [ 1.5cos(i) 1.5sin(t) ]x
(3.4)
The resulting closed loop system can be found in example 5.3,109 by Vidyasagar
[10] and also in Khalil [13]. The eigenvalues for the resulting closed loop system are
A = .25 j.6614. Since the eigenvalues have negative real parts, one would expect
the closed loop system to be stable. However, Vidyasagar shows that the transition
matrix is
$(*, 0)
e'5 cos() e sin(t)
e-osin() e_cos()
(3.5)
46


50
Trajectory of the Lyapunov based closed loop time varying system
Figure 3.3. A trajectory of the closed loop system using the Lyapunov design method.
Xq = [1 0]T. The system is stable.


69
Solve det[k.Pi + P0] = 0 for in terms of k for c = c1, ... cn.
Reject all values of k which do not meet the constraints from
steps 2) or 6). The result is 2n intervals whose lower bounds are
designated by kl5 ... k2n and whose upper bounds is named ki, ..., k2n.
Find the intersection of these intervals by evaluating
k = max[ka,...,k2n]
k = mn[ki,k2].
Find the midpoint of the interval by computing
kjlfc = i(k + k).
Let K kj,kQj,k F Aoj,k
Next k
Next j
Ki+1 = K
Remarks:
Let the feedback sets /Cl5 1C i, JC2, AF2, jCn, and Kn be defined respectively as the
set of all K so the constraints (5.31) are met. Theorem 2 provides conditions for the
nonemptiness of and lCn. But conditions for the nonemptiness of the remaining
sets are still unknown. Also, K, 1, ... ,)Cn are not convex in general. These are the
limitations of using the modified design algorithm.
We now turn to selecting constraining values for the eigenvalues of ^[P(A BK)^-
(A BK)TP]. It is necessary to evaluate the uncontrollable normal velocity compo
nents for each flight condition that will be generated in Figure 5.1. Tables 5.2 and 5.3
show for both models the minimum and maximum uncontrollable normal velocities
for each changing flight parameter.


CHAPTER 5
THE DEPENDENCE OF GAINS ON FLIGHT PARAMETERS
This chapter describes the procedure that was used to determine which flight
parameters would be scheduled against the gains. Figure 5.1 gives a block diagram
of the system used for this procedure. The feedback gains are computed using the
iterative Lyapunov design method described in Chapter 2. The gains depend on
the linear model which in turn depends on seven flight parameters. Six of these
parameters are held constant while the seventh one changes. The resulting gains are
checked to see if they depend on this changing parameter. The process is repeated
for each flight parameter. It is found that the gains depend on angle of attack, mach
number, and dynamic pressure. This information will be used in the gain scheduling
process. The next section will discuss the atmospheric tables used to generate p and
V from M and Q. The flight parameter generator will then be presented. The third
section describes the initializer. The details of the iterative Lyapunov design method
are given in the fourth section. Finally, the results of the comparison between the
gains and flight parameters will be given in the fifth section.
5.1 Generating o and V from M and 0
By inspection, the linear models from Chapter 4 clearly depend on a, j3, p, q, r,
p, V, and the aerodynamic coefficients. The coefficients, however, can be eliminated
from the list because they depend on mach number and angle of attack. It is desirable
to replace p and V with M and Q since the later two can be easily measured on the
missile. We know the following.
Q = \pV2 (5.1)
1/
M = (5.2)
V SOS
60


the size of a time varying uncertainty. Both of these tests were found to be too
conservative when applied to an already existing stable autopilot, and the results
were inconclusive. In addition, these tests are analytic tools. As yet, no design
procedure exists which can give a closed loop linear system that will pass either
test. It would be desirable to formulate a design method that can yield a closed
loop time varying linear system which satisfies a less conservative stability condition.
Vidyasagar [10] gives two important results. The first can also be found in Hahn [11]
and says that given a time varying system
x = A(t)x
if a positive definite matrix P can be found so that the matrix
(1.17)
PA(t) + AT(t)P
(1.18)
is negative definite for all time, then the system is asymptotically stable.
The second result allows the first result to be applied to a nonlinear system. Given
a nonlinear system of the form
x = /(i,x)
where
/(GO) = o
and / is continuously differentiable, then let
df(t,x)
m =
dx
x=0
and assume that
||/(i,x) A(i)x||


12
Figure 2.1. The trajectory of (a) the system (2.1) and (b) a frozen system.
This system was simulated using MATLAB with a = 1.5 and xo = [1,0]T. The
eigenvalues of A(t) are \ + ^j and Figure 2.1a shows the state trajectory
of the system 2.1 where Xi is assigned to the horizontal axis, and X2 is assigned to
the vertical axis. This plot demonstrates the instability of the system. If the system
were frozen, i.e. A(t) = A(0), then the stable trajectory of Figure 2.1b would result.
This trajectory is shown with the velocity vector field of the frozen system.
To explain the instability in Figure 2.1a, we will now discuss the time varying
velocity field of equation (2.1). The plots in Figure 2.2 show the trajectory of the
time varying system in the state plane for eight instances in time. For each plot, the
velocity vector field for that instant in time is superimposed on to the trajectory. Each
plot in Figure 2.2 shows the boundaries of four pie-shaped regions which this paper
refers to as positive and negative regions. Two positive regions exist that are defined
as the set of all points whose velocities have an outer radial component, i.e. each of
these velocities have a component pointing directly away from the origin. Likewise,
there are two negative regions containing velocities with inner radial components.
The plots show that the boundaries separating the positive and negative regions
rotate in a clockwise direction. Also, the current position in the trajectory remains


87
Likewise, for the roll-yaw controller, the limiting values at all the test points are
-2.03 x 10s
<
Xprl
<
-9.28 x 10'
-2392.7
<
Xpr2
<
-182.6
-26.27
<
Xpr3
<
-3.96
(7.4)
Some of these values differ significantly from the desired constraints; however, since
these values are still negative, these deviations are acceptable and indicate that the
closed loop system will be stable. It should also be noted that most eigenvalues
remain well within their desired constraints as shown in Table 7.1.


90
aVc will become small and aZc will become positive. ac and (3C are computed in an
attempt to match aZc and aVc respectively.
The autopilot implements the control law
U = KX + KrefV (8.1)
where x is a vector containing the actual states and v contains the state commands
from the bank-to-turn (BTT) logic. The states come from exact measurements in the
simulation. If this autopilot were to be implemented in an actual missile, the states
would be measured using an inertial platform. The gains K and Kref come from the
gain schedule implemented with a combination of polynomials and interpolation. In
this simulation there is no delay in the gain schedule and K and Kre¡ are produced
instantaneously. The output of the autopilot is the control surface angles Sp, 6q, and
6r. Linear and angular accelerations are computed by the missile dynamics module
of the program. The simulation uses the output of the missile dynamics to compute
all of the flight variables including the position and velocity of the missile.
8.2 A Test of State Tracking
The model for the EMRAAT missile has five states and three inputs. The autopi
lot is designed to track three state commands: ac, /3C, and pc. Before running missile
target scenarios it was decided to test the autopilots tracking ability. The BTT logic
was disconnected, and the following commands were applied to the reference inputs
of the autopilot.
10 for 0 s < t < .5 s
_ 0 for .5 s < t < 2.75 s
ac ~ 10 for 2.75 s < t < 3.75 s
, 0 for 3.75 s < t
( 0 for 0 s < t < 2 s
3C = < 5 for 2 s < t < 2.5 s
( 0 for 2.5 s < t
(8.3)


I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
Thomas E. Bullock, Chairman
Professor of Electrical
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy,
Jacob Hamfner
Professor of Electrical
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
Michel A. Lynch
Lecturer of Electrical
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
John Staudhammer
Professor of Electrical
Engineering


77
Table 6.1. Values of a in the pitch control
a
Pitch
Roll-Yaw
-3.0
-1.0
1.0
1.0
4.0
2.5
8.0
4.0
12.0
8.0
16.0
12.0
20.0
16.0
20.0
er look-up table.
made more restrictive and returned to their original assignments. This became a
tedious process for some parts of the grid. When the number of iterations exceeded
100 it was decided to add more grid points so that the desired feedback gains could
be found in fewer iterations.
6.2 Formulation of the Design Constraints
Before using the iterative Lyapunov design algorithm, the uncontrollable normal
velocity components for the entire M-Q-a grid must be determined. This information
is needed to formulate the design constraints. This process is described by the block
diagram in Figure 6.1. Table 6.2 presents the minimum and maximum uncontrollable
normal velocity components of the pitch model throughout the M Q grid for each
value of a. Here P = Pq, the matrix computed during the initializing procedure
of the last chapter. Table 6.3 gives the same result for the roll-yaw model where
P = Ppr- The extreme values of uncontrollable normal velocities for the pitch model
are 49.5985 and 37.3480. Also the eigenvalues of
5IP,{A, B,K,) + (A, B,K,)tP,}
(6.1)