You are on page 1of 6

TP3.

Missile Endgame Analysis *


Jyh-Ching Juang, Ching-Fang Lin,
American GNC Corporation
9131 Mason Ave.,
Chatsworth, CA 91311

Johnny H. Even,and James R. Cloutier


USAF Armament Directorate
WLJMNAG
Eglin AFB, FL 32542

Abstract:

design and analysis framework is introduced. Based upon


this framework, a metric that accounts for the worst case
In missile guidance analysis, the engagement performance outcome is used to quantify endgame results. The approach
is typically quantified using the root-mean-square (RMS) is applied to compare two autopilots of the HAVE DASH
value. The underlying assumption of using the R M S value I1 system in Section 4, In Section 5 , conclusions are drawn.
is that the engagement outcome is atTected by design uncertainties that are modeled using random processes. However, many attributes that affect the engagement outcome 2
Endgame Dynamics and Uncercannot be appropriately modeled using random processes.
tainties
Examples include, but are not limited to, parameter variations, flexible dynamics, target maneuver, estimation erIn this section, the endgame dynamics are modeled using
rors, and parasitic dynamics. In this paper, instead of
a set of linear equations. Two-dimensional engagement is
looking for the average R M S performance, a new metric
assumed. The missile dynamics are modeled as
that considers the worst case performance is adopted. As a
result, a more accurate performance prediction can be ob& = zaa+q+Z66
(1)
tained. The analysis is also shown to generalize the existing
Q
=
(ma
Ama)a
m,q
m66
(2)
covariance analysis approach. The analysis framework can
be expanded for the integrated guidance and control (IGC) Here, a is the angle of attack, q is the pitch rate, and 6 is
design.
the pitch control deflection. The aerodynamic derivatives

1 Introduction
The performance of a missile at the endgame is known to
be crucial in the resulting engagement outcome. During
the endgame, several attributes interact and affect the resulting miss distance (11 [2]. However, there is a lack of
understanding of the endgame dynamics. The purpose of
this paper to provide a framework for the understanding
of the endgame dynamics and the design of an improved
guidance/control strategy. In this paper, the endgame dyn b c equations are derived and used as a framework for
the analysis of guidance/control laws. Guidancejcontrol
laws are then assessed in this framework to investigate the
worst case performance. In contrsst to covariance analysis
[2] [3] where the system is assumed to be driven by a random process, the new analysis w u m e s a bounded energy
signal and looks for the performance in the worst case scenario. The approach, similar to the way that Hw design
generalizes HZ design, uymptoticdy recovers covariance
analysis. The paper is organized into five sections. In Section 2, the endgame dynamics are reviewed. In Section 3, a

are represented by zo, 2 6 , m,, m,, and m6. The stability derivative ma, pitch moment with respect to angle of
attack, is known to determine the static stability of the
missile short-period dynamics. This parameter may vary
significantly. Thus, Ama is used to model the variation of
ma. The acceleration a, delivered by the missile is known
to be
a
, = Vm(q - d r ) = -Vm(zaa
366)

where Vm is the velocity of the missile.


The miss distance is determined by the relative motion
of the target and the missile.
Yr

= at - am

(3)

where the target acceleration at is assumed to satisfy the


first order model
at

= -&at

xtatc

(4)

A, and ate are target maneuver bandwidth and target acceleration command, respectively.

There are two sets of measurements that can be utilized

'Rescw-ch supported under USAF Contract No. F0863Lb91-C in the guidance and control design. The target informao030.
tion (range, range rate, line-of-sight angle, and line-of-sight

CH3OOO-7/9!2/0000-0744$1.0001992 IEEE
-

~-

744

rate) is obtained by processing seeker or uplink measure- where


ments. The ownship motions are sensed through an inertial
measurement unit which includes rate gyros and accelerom
eters.
The true (geometric) line-of-sight
approximated as
@true

(LOS)angle UtrUe is

Ap =

=Yr

0
0

where r is the radome boresight error slope, 0 is the pitch


attitude, and nu is the noise. The line-of-sight angular rate
measurement is assumed to satisfy

0
1

0
0
0

0
0

t.v.q,

Vctgo

where V, is the closing speed and t,, is the time-t+


intercept. The lineof-sight angle is typically contaminated
with the effects due to radome boresight error and noises.
Typical noises include clutter, fading, thermal, and target glint noises, each having a specific frequency and range
characteristic. Thus, the line-of-sight angle becomes

um =

0
Vmza

-At

0
0

0
0

O z a 1 0
O m , , O O

7,Vctgo

0
0
0
0
0

0 0
0
0
0

0 - 1

7.

0
vm 26

Bp =

26

Bw, =

m6

AAp =

- 0
0
0
0
O
0

0
0
0
0
O
0

0
0
0
0
O A
0

0
0
0
0
m ,
0

-cr o

u+ni
1+ r,s
9

where n+ is the noise and r, is the seeker time constant.


Thus,

Ap is the nominal plant model, AA, contains the perturbations due to Am,, and r , Bwl characterizes the contribution
The measurements of the ownship motion include the from exogenous signals, and Bp is the control distribution
pitch rate q and the missile acceleration am. Both are submatrix. The measurement contains the line of sight rate,
ject to noise, flexible dynamics, and sensor dynamics. Thus,
missile acceleration, and pitch rate.
the measurements become

Yr

Yr

ir

ir

at

d t a

!l
U

= (AP+AAP)

at

Q
U

+Bp6 + Bw,

.[ 2 ],
~

The above modeling can be generalized to threedimensional engagement and nonlinear dynamics. It is also
possible to incorporate the attitude information and warhead dynamics to the model.

w1

745
~

can be obtained using the robust miss distance analysis where


approach.
.

3.1 Analysis Framework


The schematic diagram for guidance analysis is depicted in
Figure 1 . Here the endgame dynamics and plant uncertainties =e related t o the miasile airframe, target dynamics,
and engagement kinematics. The plant uncertainties include seeker noise, parasitic feedback, and target maneuver.
(7) and (8) describe typical endgame dynamics and plant
uncertainties. The guidance and control system consists of
the target state estimation, tracking, guidance, autopilot,
and fuze subsystems. Uncertainties may also be assiciated
with the guidance and control system. Examples are time
to go estimation error, estimation delay, and fin limitation. For each externally supplied guidance and control
law, a closed-loop is formed. The resulting trajectories can
be analyzed to assesa the sensitivity due to heading error,
radome boresight error, passive versus active seeker design,
seeker noise, time-to-go estimation error, aerodynamic p%
rameter variations, measurement error, target maneuver,
target estimation delay, high frequency dynamics, control
limitations, and guidance/control designs.
Consider the guidance law
(9)

and the autopilot

6=kj(l+

+ k,(l+

k.

+?a,

k.

$,a,

+ L,q,

[ $0

0 0
La 0 1

contains two state variables; one is the guidance state


(estimate of the line of sight rate), another is related to the
integral of the acceleration tracking error. The exogenous
signal ~3 is used to represent the 'equivalent' estimation
error of the time-to-go. The overall system block diagram
is illustrated in Figure 2. One can easily establish the correspondence between Figure 1 and Figure 2.
zk

Combining the endgame dynamics (('3) and (8)) and the


guidance/control system ((9) and (lo)), the closed-loop system is formed. More precisely, let z b e the augmented state
z

[2]

; w be the augmented exogenous signal which

comprises of w l , w2, w3, and equivalent disturbance due to


uncertainties (due to Am, and r ) ; and the miss distance
(yr at zero time-to-go) be the performance, after some ma(10) nipulations, the system can be modeled as

i ( t ) = A(t)+(t)+ B(t)w(t)

The guidance law (9) is a standard proportional navigation


t ( t ) = C(t)z(t)
(11)
law with an effective navigation gain A and a first order filis a mod&ter whose time constant is rg. The term
The closed-loop system can be drawn as in Figure 3.
cation term to jack up the navigation gain at the endgame,
Here,
all the known dynamics are augmented in the augi.e., to gain maneuverability against maneuvering target at
s m a l l time-to-go. The parameters kl and k2 are determined mented dynamic block and the uncertainties are lumped in
experimentally or through extensive simulations. i,, is the the uncertainty block.
estimated time-tqo, which can be assumed to be
This framework is standard in robust control and analysis. However, it has not been used in systematically analyzigo= t,, Atgo
ing the endgame results. It is also noted that there are two
salient features of the endgame analysis that are considwhere At,, is the time-to-go estimation error. The autopierably different from the robustness analysis. Firstly, the
lot ( 1 0 ) is a standard three-lmp design. The gains L,, Li,
augmented plant is time-varying and possibly nonlinear. In
h a , and k l are selected to have a satiafactory command
most robustness analysis, the nominal plant is assumed linresponse and gain/phase margins.
ear time-invariant, the nonlinear and timevarying efffects
This guidance and control law can be written as
are absorbed in the uncertainty block. Secondly, the stability robustness gives way to the performance robustness in
;r,
the endgame. More precisely, the stability requirements of
i k = AkZk
Bk; ; a
-k B w , W
the missile can be relaxed for the last seco~dto go before
intercept. Indeed, the design should focus on maneuverability over stability to gain maximal agilitv to obtain a
reduced miss distance and an improved hit-to-kill. In rG
bustness analysis, stability is assumed throughout.

&

[ ]

746

3.2 H" Metric

with

Most engagement results are quantified using the rootmean-square (RMS) values [2]. The R M S values are derived
by running Monte Carlo simulations, covariance analysis,
or adjoint analysis. The underlying assumption for R M S
be modeled as stochw
analysis is that the
tic processes. Consider the system in (11). In covariance
analysis, it is assumed that the exogenous input w is a random process. The analysis is to investigate how the statistical information is propagated in the (Gauss-Markov)
process. The state covariance matrix is defined as

X ( t o )= R r l

and

y z X - ' ( t j ) > Rj

(15)

Thus, in the robust miss distance analysis, the focus is


on the Riccati equation (14). There are several advantages
of using (14) to investigate the endgame Performance.
It is clear that (14) generalizes the covariance propagation equation in (12) in the sense that when 7 is sufficiently
large and R;' = Po, (14) reduces to (12).

- -+(t)l)(z(t) - E[.(t)l)=)

It is known that the covariance at the final time, P ( t j ) ,


carries the FtMS value of the find state, e.g., RMS of the
where E is the expectation operator. Assume that tu is a misssdistance. under the assumption that A is stable, ( A ,
white, Gaussian, zero mean process with identity COVari- C)is detectable, and X(t,) = p ( t o ) ,it can be shown that
ance. The state covariance P satisfies the following equa- the matrix
dominates p , ice
tion:
X ( t ) > P(t) v t > t o
P = A P PAT B B T , P ( t o )= Po
(12)
The root-mean-square value of each state variable at
Thus, the entry of X at the terminal time can be regarded
time can be obtained from P . The covariance can also be as an upper bound of the corresponding RMS value. The
interpreted as the summation d l impulsive responses due difference between the upper bound derived from and the
to implusive inputs applied one at a time.
covariance value is due to the additional requirement that
Many uncertainties such as parameter variations, flei- the energy ratio between the output and input is bounded
ble effects, radome induced errors, target maneuvers, and by 7. A S the value Y decreases, the gap between
and p
estimation errors cannot be modeled using stochastic pro- becomes large.
In Some Case, the 'Uncertainty' is at the mercy of
some designs, where is specified to characterize some
the adversary that manipulates the uncertainty to yield a physical limitations, the performance measure to be invesworst case outcome. A S a result, using the RMS value as a tigated is the X solution in (14) rather than the P solution
measure may become conservative and misleading. A met- in (12).
an example, let 1u be the target maneuver and
ric that quantifies the worst case behavior rather than the be the missile acceleration. m e n the ~ s s i l is
e constrained
average behavior is thus desirable. Such a robust perfor- in its acceleration capability or the target is highly maneumeasure for the endgame analysis is defined as
verable, the ratio y decreases. As a consequence, the miss
distance is likely to increase. Thus, the H" metric provides
P Z d t zT(tj)Rjz(tj)
a tradeoff hetween the terminal state and the energy. This
J =sup[
w T w dt z T ( t O ) R o z ( t o )
tradeoff cannot be observed using the covariance analysis.

P ( t ) = E{(+)

+
+

st','

Ly

If

Here, the ratio between the energy of the output and that of
the exogenous input, under the influence of the worst case
input, is used as the measure. The energy is defined as
the integral of the square of state excursion from the initial
time t o to the final time t j . The matrices R, and RI are
weightings that penalize the uncertainty in the initial state
and the terminal state, respectively. This same norm has
been used in robust control of time-varying system.

The minimal 7 such that (13) is s~&fied is an indication


of the robustness of the design. One can indeed plot the
curve of y versus the terminal X to assess the tradeoff in
robustness and performance. Design insights can also be
obtained this way. For example, if the worst case miss
distance is specified, the curve can be used to show how
high the control authority should be in order to ensure the
required miss distance.

The worst case performance is obtained by finding a 7


Note that only forward integration is needed in the robust miss distance analysis. The terminal penalty function
such that
RI is free during this integration process. Thus, if one
z T ( t ) p C z ( tdt
) z T ( t j ) R j z ( t j )< y z
(13) regards Rj as a free parameter, the best terminal performance can be evaluated after the forward integration. This
wT(t)w(t)dt zT(tO)Roz(to)
will involve a spectral test (15). The terminal state can
for any permissible tu. After some manipulations, it Can be then be ensured to lie within an absolute bound which is
shown that it suffices to find a positive definite matrix
established by and ~ ( t , ) .
such that
One advantage of this Hw metric is that multiple, siX = AX X A T BBT y - z X C T C X
(14) multaneous parameter variations can be assessed. Recall

hy

Jy

747

that in the covariance analysis, the contributions due to


different uncertainty sources satisfy an additive property.
u;,lulj. This is equivalent to requirThat is, U:, =
ing that each uncertanty applies one at a time. Since the
endgame outcome is affected by the interactions of many
uncertainty sources in a highly complex way, the analysis
should account for the combination of various uncertainties in the worst manner. The effect of the interaction can
be assessed using the Hm metric. This is done simply by
combining all the uncertainties in the B and C matrices in
(11). To reduce conservatism, diagonal scaling techniques
can be applied. However, the scaling factors are required to
be static and commutative with the uncertainty structure.

cj

Application

Concluding Remarks

An endgame analysis and design framework is described.


This framework can help understanding of the endgame
dynamics, identifying critical design parameters, and bringing insights to the guidance/control design. A worst caae
analysis approach is introduced and applied to the HAVE
DASH I1 system. The analysis approach generalizes the
covariance analysis method and provides a more accurate
prediction on the engagement performance.
Based upon the analysis framework and performance
measure, robust control design methods can be applied
to obtain a robust integrated guidance and control design.
The results will be presented in another paper.

References

The above approach is applied to analyze the guidance and [l] J. R. Cloutier, J. H. Evers, and J . J. Feeley. An assesscontrol designs of the HAVE DASH I1 system. Two autopiment of air-to-air missile guidance and control technollots are assessed in the followings. Autopilot 1 is designed
ogy. IEEE Control System Magnztne, 27-34, 1989.
using the Ha approach. Autopilot 2 is designed through
the quantitative feedback theory. Both autopilots exhibit 121 C. F. Lin. Modern Navigation, Guidance, and Control
satisfactory gain/phase margins and
response at
Processing. Prentice-Hall, 1991.
all flight conditions. The guidance system is assumed to be
a proportional navigation type with the effective naviga- [3] F. W. Nesline and P. Zarchan. Miss distance dynamics
in homing missiles. In Proc. AIAA guidance and control
tion ratio A = 4.0. The target is assumed to be a first order
conference, pages 84-98, 1984.
= 1.Osec. The maneuver amsystem with time constant
plitude is normalized by 9 g. The guidance time constant
rgis assumed to be 0.2 sec. The matrix R, is assumed
to be a diagonal matrix. The 'variances' corresponding to
gr, yr, and uc are assumed to be 10 ft, 5 ft/sec, and 10
ft/sec2, respectively. In the analysis, the exogenous input
is the target maneuver amplitude act and the performance
measure is the control activity 6.
The covariance analysis leads to the I M S trajectory p r e
files for the miss distance and the control deflection, which
are shown in Figures 4 and 5. Autopilot 1 (13.5457 ft) has
a smaller terminal RMS miss distance than the second autopilot (17.7137 ft). However, Autopilot 1uses more excessive control deflections in correcting initial heading error.

The properties of the X solution and the P solution are


illustrated in Figure 6. Here the value y is selected to be
10. The square roots of the entries corresponding to the yr
state are plotted. The robust analysis, which considers an
additional constraint on energy, leads to a larger estimate
than the covariance analysis.
The robustness versus performance results for the two
autopilots is plotted in Figure 7. Autopilot 1 has a smaller
RMS miss distance than Autopilot 9 Eowever, Autopilot
1 is less robust than Autopilot 2. That is, against a highly
maneuverable target, the former is more likely to saturate
than the latter. This information c a Le
~ used to refine the
guidance/control law or to resize the control surface.

748

Figure 1. Guidance Analysis Schematic Diagram

. _

Figure 5: Covariance Analysis: RMS Control Deflection


najectory

Figure 2: Guidance Analysis Block Diagram

Figure 3: Analysis Framework

~i~~~ 4: covariance ~ ~ ~ lms


~ ~Mia4
i Distance
s :
jectorv

n& Figure

mance

749

7: Tradeoff in Robustneas and Terminal Perfor-

You might also like