REPORT DOCUMENTATION PAGE Form Approved OMB No.

0704-0188
Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources,
gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this
collection of information, including suggestions for reducing this burden to Washington Headquarters Services, Directorate for Information Operations and Reports. 1215 Jefferson
Davis Highway. Suite 1204, Arlington, VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188). Washington, DC 20503
1. AGENCY USE ONLY (Leave blank) | 2. REPORT DATE 1 3. REPORT TYPE AND DATES COVERED

2010 Technical Report

4. TITLE AND SUBTITLE 5. FUNDING NUMBERS

Aerospace Applications of Adjoint Theory

6. AUTHOR(S)

Domenic Bucco
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION
REPORT NUMBER
DSTO

9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES 10. SPONSORING/MONITORING

AGENCY REPORT NUMBER
DSTO
PO Box 1500 Edinburg
South Australia 5111 Australia

11. SUPPLEMENTARY NOTES

12a. DISTRIBUTION/AVAILABILITY STATEMENT 12b DISTRIBUTION CODE

Approved for Public Release.

ABSTRACT (Maximum 200 words)

The adjoint simulation method is an efficient, computerized method for the performance analysis of linear time varying (LTV) systems
excited by deterministic and/or stochastic inputs. It is based on the impulse response function (or weighting function) of the so-called
adjoint system - an associated linear system derived from the original system by clever utilisation of the mathematical principle of duality.
Once the adjoint system is obtained, the solution process using this method is quite straight forward regardless of the nature and number of
inputs. In this report, we demonstrate the power of adjoint simulation in the context of guided missile homing loop studies and present a
MATLAB tool that can be used for its speedy implementation and execution. The tool is demonstrated by using it to analyse the miss
distance performance of a generic guided missile against an evading target.

14. SUBJECT TERMS 15. NUMBER OF PAGES

DSTO, Australia , Adjoint theory. Adjoint simulation method, Linear time varying (LTV), Mathematicla
principal of duality. Guided missile homing loop studies. Guided missile. Weapon programs. Practical
engineering approach, Laning and Battin rules. Simulation diagrams. Feedback paths, MATLAB/Simulink
16. PRICE CODE

17. SECURITY CLASSIFICATION 18. SECURITY CLASSIFICATION 19, SECURITY CLASSIFICATION 20. LIMITATION OF ABSTRACT
OF REPORT OF THIS PAGE OF ABSTRACT

CLASSIFIED UNCLASSIFIED UNCLASSIFIED UL

NSN 7540-01-280-5500 Standard Form 298 (Rev. 2-89)
Prescribed by ANSI Std. 239-18
298-102
Department of Defi'tut'
Defence Scieooe and
Techaolog) (hjNiimntion

DSTO FELLOWSHIP PROGRAM

Aerospace Applications
of Adjoint Theory
Domenic Bucco

DSTO Science and Technology for a Secure World

DSTO Fellowship Program
The DSTO Fellowship program encourages scientific innovation and creativity within
DSTO through a competitive award scheme. DSTO Fellowships recognise and support
meritorious research activity of potential future value to Defence or National Security
undertaken by our high-achieving scientists.
This program is part of the DSTO's Corporate Enabling Research Program (CERP) and
it is one of the important mechanisms at DSTO for fostering interactions with and
leveraging from industry, academia, and Australian and overseas research bodies.

Published by

Weapons Systems Division

DSTO Defence Science and Technology Organisation
PO Box 1500 Edinburgh
South Australia 5111 Australia

Telephone: (08) 8259 5555

Fax: (08) 8259 6567

© Commonwealth of Australia 2010

January 2010

APPROVED FOR PUBLIC RELEASE

 Australian Government
Department of Defence
Defence Science and
Technology Organisation

Aerospace Applications of Adjoint Theory

Domenic Bucco
Weapons Systems Division
Defence Science and Technology Organisation

ABSTRACT
The adjoint simulation method is an efficient, computerised method for the
performance analysis of linear time varying (LTV) systems excited by deterministic
and/or stochastic inputs. It is based on the impulse response function (or weighting
function) of the so-called adjoint system - an associated linear system derived from
the original system by clever utilisation of the mathematical principle of duality.
Once the adjoint system is obtained, the solution process using this method is
quite straightforward regardless of the nature and number of inputs. In this report,
we demonstrate the power of adjoint simulation in the context of guided missile
homing loop studies and present a MATLAB tool that can be used for its speedy
implementation and execution. The tool is demonstrated by using it to analyse the
miss distance performance of a generic guided missile against an evading target.

RELEASE LIMITATION
Approved for public release

4Q V(o-fl-6iq5c)
 Aerospace Applications of Adjoint Theory
Executive Summary
Adjoint theory has widely been employed by various missile companies as a pivotal
tool in support of their weapons programs. These companies have traditionally
adopted this practical engineering approach due to its simplicity, accuracy and
efficiency of use, particularly during the preliminary and conceptual stages of a new
missile design. In this approach, design parameters for the missile system can be
readily selected on the basis of sensitivity curves generated by the efficient simulation
of an associated system adjoint to the missile system under study.
However, one of the difficulties with this approach is the determination of the
associated adjoint system. Two methods exist for doing this. The adjoint system can
be obtained via a practical engineering approach or via a more rigorous mathematical
approach. Traditionally, defence engineers have used the engineering approach.
This is based on a set of adjoint construction rules developed by Laning and Battin
(7] while implementing their adjoint equations on an analogue computer. With
this approach, given the original dynamic system in the form of a simulation block
diagram with designated inputs and outputs, the adjoint system is constructed
through manual manipulation of this block diagram in accordance with the Laning
and Battin rules.
For very complex simulation diagrams with many feedback paths typically found in
missile guidance loops, manual application of the adjoint construction rules may
be extremely tedious, time consuming and prone to error. However, if the original
simulation block diagram is implemented in a commercially available graphical
simulation package such as MATLAB/Simulink, then the process required to construct
the adjoint system can be conveniently automated to minimise or eliminate these
shortcomings. A tool for doing this is outlined in this report. The tool can be used to
support preliminary studies into new concepts such as hypersonic vehicles.
The objectives of this report are to review the classical theory of the adjoint method
as applied to the guided missile problem, to present a pragmatic engineering
implementation of the theory using block diagram representation and to outline the
development of a specialised software package designed to automate the process
of adjoint system construction within the MATLAB/Simulink environment. Several
examples drawn from the open literature are used to illustrate the solution process
for typical problems encountered in Aerospace/Defence science.
 Author

Domenic Bucco
Weapons Systems Division

Domenic Bucco received his PhD in Applied Mathematics from the University of
Adelaide. South Australia. Since 1980, he has been working at the Defence Science
and Technology Organisation (DSTO) as a research scientist in the Weapons Systems
Division. He has worked in various areas of the division including guidance and control,
modelling, simulation and analysis of guided weapons. During 1987/1988. he was
posted to the Naval Air Warfare Centre, China Lake, California, as an exchange scientist
for 18 months. While there, he worked on autopilot design and hardware in the loop
simulations. After promotion to S81T 7 in 1998, he has managed and led several groups
in the Division including the weapons modelling, simulation and analysis group, the
missile simulation group and the weapon robotics group. He has recently completed an
18 month fellowship working on weapons systems analysis and simulation techniques
using adjoint theory.
 Fellowship Publications and Presentations

1. Bucco, D. & Weiss, ill, "Further Enhancements to COVAD: A MATLAB Tool Based on
Covariance/Adjoint Methods,"A1AA M&STConference. Honolulu, Hawaii,
Aug20o8.
2. Weiss, M. & Bucco, D., "The State-Space Approach to the Method of Adjoints
for Hybrid Guidance Loop Models," AIAA GN&C Conference, Chicago, IL, USA,
Aug 2009.

3. Bucco, D. & Weiss, M., "Extending the COVAD Toolbox to Accommodate System
Non-Linearities," AIAA M&ST Conference, Chicago, IL, Aug 2009.
4. Rajagopalan A. & Bucco D. "Applications of Adjoint Theory to Problems in
Aerospace/Defence Science," EMAC Conference, Adelaide, Dec 2009.
Contents
i. Introduction 1

2. Historical Perspective 2

3. The Adjoint Simulation Method 5

3.1 Deterministic Inputs 5
3.2 Stochastic Inputs 7
3.3 Adjoint Construction Rules 9
3.4 Example of Adjoint System Construction 10

4. The Adjoint Software Tool 12

4.1 COVAD Toolbox Overview 12
4.2 Use-Case for Adjoint Analysis 13
4.3 Verification and Validation of the Toolbox 14

5. Missile Applications 15

6. Conclusion 24

Acknowledgements 25

References 25
i. Introduction
Systems analysts and engineers are often confronted with understanding the
behaviour and performance of complex, dynamic systems driven by various inputs
predominantly of a stochastic nature. Some typical applications include miss
distance studies of guided missiles, fire control problems, navigation problems and
analysis of circuits due to random noise. For these systems, the most general method
of tackling the problem and gaining critical statistical insight into the nature of the
response is via the use of the Monte Carlo simulation technique (i). However, when
the system is linear or can be approximated by a linear system, the use of the Monte
Carlo method would seem somewhat excessive and unwarranted in such cases.
Instead, the stochastically-driven linear system may be more economically and rapidly
studied using better matched linear techniques. One such technique is based on
adjoint theory [2] and is traditionally known as the adjoint simulation method (or
modified adjoint method).
The adjoint simulation method is a computerised method for the performance analysis
of linear time varying (LTV) systems. Mathematically, it is derived from the formal
theory of linear differential operators and their associated adjoints. Traditionally, the
technique has been very popular in the study of guided missile homing loops and, in
general, has been employed during the preliminary and conceptual definition stages
of many new missile programs. For this work, the LTV system under study is typically
represented in the form of a simulation block diagram. The associated adjoint system
is also represented in block diagram form. The adjoint system is constructed by
manual manipulation of the original system block diagram in accordance with a given
set of rules. These rules were first proposed by Laning and Battin [7] as part of their
adjoint simulation work on analogue computers.
In general, given an LTV system with n inputs and m outputs, the adjoint method can
be used to determine the sensitivity of any of the outputs at a fixed time to each of
the n inputs. If the inputs are stochastic in nature, then this approach can be easily
formulated to compute the Root Mean Squared (RMS) value of the system states of
interest without resorting to ensemble computations as generally required by the
Monte Carlo method [1). Hence, the technique provides the analyst with a simple but
powerful alternative to the Monte Carlo approach in those cases when linearisation
can be physically justified.
The purpose of this report is to review the theory of the adjoint simulation method as
applied to LTV systems, to present a pragmatic implementation of the theory using
block diagram representation and to outline the development of a software package
designed to automate the adjoint system construction process within the Matlab/
Simulink environment (50]. Several examples drawn from the open literature will be
used to illustrate the solution process to typical problems encountered in Aerospace/
Defence science.

2. Historical Perspective
According to Zarchan [2], the adjoint method can be traced back to the Italian
mathematician Vito Volterra, circa 1870. However, the book by Ince [62] points to
Lagrange as the discoverer of the method over a hundred years earlier. According
to Ince, while exploring ways of solving the variable coefficient ordinary differential
equation with non-zero right hand side, Lagrange was the first to mathematically
formulate the adjoint differential equation, although he did not use the term
"adjoint" to describe it. The term was later used by Fuchs in 1873 [62].
The first practical use of the adjoint method was subsequently developed by Bliss
while serving as a mathematical expert at the Army Proving Ground at Aberdeen,
Maryland, US, in 1918 [3]. Bliss used a property of the adjoint method to study the
effects of perturbations on a system of equations associated with the flight of shells.
Ballisticians then used the method in connection with their theoretical studies of
artillery hit dispersions [4].
In the early 1950s, Bennett [5] employed the adjoint method for statistical
performance analysis of linear systems on an analogue computer. Subsequently,
Laning and Battin [6] described the use of the method for the statistical analysis
of time variable networks. However, it seemed that the adjoint method was still
relatively unknown until 1956 when Laning and Battin [7] included a discussion and
proof of the method in their book on random processes. This book helped popularise
adjoint simulation and is still a highly referenced resource on the method.

In the 1960s, many references appeared on the applications of the adjoint method
to systems analysis using analogue computers. Books by Rogers & Connolly [8|,
Fifer [9] and Leondes [10] devoted chapters to the technique while Thorson [11]
examined the method more closely while undertaking research as part of a Master of
Science program. Peterson [12], and later Howe [13] and Tarrant [14), illustrated how
the method could be easily applied to the performance analysis of generic guided
missile systems. In fact, with regards to guided missile studies, the Peterson book
is referenced often. Several other references [15-17] discuss the adjoint simulation
approach and its application on the analogue computer. In all these applications,
the adopted methodology for the construction of the adjoint system followed that
proposed by Laning and Battin in the form of their adjoint construction rules [7].
During the same time frame, a slightly different mathematical approach applying the
adjoint operator to linear differential equations was being developed and presented
by Sussman [18]. Similar expositions using the state space representation of the
governing equations were published [19-20] relating the more mathematical approach
to the so-called modified adjoint system. State space methodologies were also
developed to handle discrete time systems [21-22]. Willems [23] used a combination
of the state space approach with the more traditional block diagram approach to
investigate the performance of a generic ground to ground missile guidance system.
The adjoint method had become a common analysis tool by the 1970s and 1980s and
was being used on preliminary design studies by various companies in the missile
industry. Advances to the basic adjoint methodology, such as the use of shaping
filters to model target manoeuvres [24-26], adjoint representation of discrete systems
[2| and applications to non-linear systems [27] extended the overall capability of
the method. The extension to non-linear systems utilised the concept of statistical
linearisation and is known as the Statistical Linearisation Adjoint Method (SLAM). An
excellent report outlining the use of the adjoint method for sensitivity analysis and
miss distance studies of generic guided missile homing loops is given by Bibel [28).
During the 1990s, the book by Zarchan [2] on "Tactical and Strategic Missile
Guidance" helped promote the use of the adjoint method as a staple engineering tool
for the performance analysis of homing missile guidance loops. Other practitioners
used the method in conjunction with the SLAM concept to explore the effects of
various noise sources on the miss distance of a radar homing missile [29-30). The
SLAM approach also featured in research undertaken as part of postgraduate degrees
[31-32]. Also during this period, an interesting development saw the introduction of
specialised software that automated the adjoint construction process using the Laning
and Battin rules [33-34]. This provided faster turn around for analysis purposes while
at the same time minimised potential errors during the adjoint construction phase.
Within the last decade, the application space for the adjoint simulation method has
expanded considerably. A new state space framework to handle continuous and
discrete system analysis by the adjoint method has been proposed by Weiss [35] and
applied to mid-course guidance problems [36-37). Raytheon has shown renewed
interest in the traditional method with applications to the preliminary analysis and
study of Ballistic Missile interceptors [38]. In addition, Raytheon researchers have
extended the basic adjoint method to allow assessment of more complex homing
guidance loops (39]. The method has also been used for comparison and evaluation
of various missile guidance laws [40-41]. Further extensions of the method include
the determination of appropriate shaping filters for estimating higher order statistics
[42], performance analysis of guidance loops under model uncertainty [43], for robust
guidance performance [44] and for conducting missile overload requirement analysis
in terminal guidance using the Li approach [45]. For non-linear homing loop analysis,
a different approach to SLAM was proposed which made use of infinite dimensional
linearisation to reduce the equations to linear form prior to applying the adjoint
process [46]. There have also been investigations carried out into non-traditional
uses of the method including applications to soft kill and survivability studies [47-48].
More recently, a new framework has been proposed for the analysis of hybrid systems
consisting of multi-rate subsystems that may be characterised by non-periodic discrete
events [49]- This facilitates the study of the effects of uplinks on the performance of a
guidance system with the use of the adjoint method.
Following [33] which had been based on the MATRIXx environment, effort has
continued on the software development front but with a major shift to the MATLAB
environment [50-53]. The last decade has also seen much interest in the method as a
topic for further research within the academic domain [54-59].
The adjoint technique has also received considerable attention in the teaching of
aerospace guidance and navigation courses |6o|.
3. The Adjoint Simulation Method
3.1 Deterministic Inputs
Consider a linear time varying (LTV) system described by the following ordinary
differential equation

at at at
where y denotes the system output u, is the deterministic system input and the
coefficients are functions of time.
According to linear systems theory [7], the solution to the above differential equation
is given by

y(t) = f u(x)w(t,x)dx (2)

where tt is the time at which the input is applied while w(r.x) denotes the system
weighting function or impulse response function. Physically, the function w(t,x),
represents the response of the system at time f to a unit impulse applied at the input at
time x. With the aid of this weighting function, the system described by equation (1)
is conveniently represented in block diagram form as shown in Figure 1.

u(t)
w(t,x ) j^V
Figure 1. Linear time varying system

While equation (2) gives the formal solution to the problem posed in Figure 1, its
practical usefulness is limited. The problem here lies in the fact that the variable of
integration corresponds to the impulse application time. Thus, in order to generate
w(f.x) required for the above solution, it becomes necessary to simulate the system
response for each impulse application time x . Once the weighting function is
obtained in this manner, it is then necessary to carry out the integration in equation
(2) numerically.
Since every linear time-varying system has a corresponding adjoint system, a better
and more efficient alternative is to reformulate equation (2) in terms of the weighting
function of the associated adjoint system. To this end, utilising the so-called modified
adjoint technique [7, 20] has the effect of requiring only one computer simulation run
to yield the desired weighting function.

By definition, the differential equation that describes the modified adjoint system
corresponding to the above system but with u = 0 is given by

£z + rf-|fl..,(f)=l + ... + rffcO-M + .)z m Q (3)

dt ' di ' <lt

where 2 is a function of adjoint time f = ff - t and f/; denotes final time. Note that
the term "modified" adjoint system is used here to indicate a time reversal in the
conventional adjoint system differential equation by definition of adjoint time f'.
Now, if the adjoint system, described by the above differential equation, is subjected
to an impulsive input at timer', then the resulting system response, known as the
adjoint weighting function, will be denoted by w'(t\x').
A fundamental relationship exists between a linear time varying system and its
corresponding adjoint system [2], namely,

w'(tF-r,tr-t) = w(r,T) (4)

where T , f and f( are the impulse application, response observation and final time,
respectively, of the original system. Thus, on substitution of equation (4) into equation
(2) and, after a change of variable, 11 = f( — T we get

v(f)= Jf'"'u{t
1, -1 p-r\)w\r\ttF-t)dr\ (5)

It is clear now that the variable of integration appears as the first argument in the
adjoint weighting function and thus the solution for the output can be computed with
just one computer simulation run. For example, if the input to the system is a step of
magnitude K which is applied at time zero, and we seek the solution at final time ty,
then equation (5) becomes
 y(tF)-Kf'BFw\r),0)dr\ (6)

Consequently, the original system response at final time can now be computed in one
simulation run by integrating the weighting function of the corresponding adjoint
system. This is represented in block diagram form in Figure 2, where b(t') denotes the
standard Dirac delta function or impulse function.

&(*•) Time Varying w"(f,0)

K
y«F)
Adjoint
s
System

Figure 2. Adjoint solution for system with step input

This procedure can be extended to include many step input disturbances as well as
other types of deterministic inputs such as ramp or sinusoidal inputs. Furthermore,
since the system is linear, the superposition principle allows one adjoint computer
simulation to yield the system response to a combination of disturbances, along
with a detailed error budget showing how each disturbance influences total
system performance.

3.2 Stochastic Inputs

Suppose the linear system is driven by some noise process n(t). The system output is
now random in nature and given by

v(r) = fii(T)w(t,x)dx (7)

Squaring both sides of equation (7) and taking expectations yields the mean square
value of the output, namely.

E[y(t)] = f<Pn(T)w2{t,T)dT (8)

where we have assumed the noise input to be non-stationary white noise with
autocorrelation function R(t,x) = <J>0(f)6(7 —X), For stationary white noise input,
the output mean square value reduces to
 E[y1(t)] = %fw2(r,x)dx (9)

where the white noise power spectral density <I>M is double sided with dimension
unit-'/Hz. Again, the problem with computation of the weighting function can be
circumvented by considering the corresponding adjoint system. In this case, we have

E[y2(t)]mQ0f^[w\y\,tF-t)fdt\ (10)

If the noise input begins at f. = o and the observation time is taken as final time ff, the
mean square response simplifies to

E[y2U,)] = *o/0"[w(i1,0)],rfri (u)

Thus the mean square response of the original system due to a white noise input
can be determined by computing the weighting function of the adjoint system in
one computer simulation run. This is conveniently depicted in block diagram form
in Figure 3.

E[y\tF)]

Figure 3. Adjoint solution for white noise input

Because of the superposition principle, the above procedure can be extended to

the case of many white noise inputs. In this case, one adjoint simulation run yields
an exact statistical solution of the noise-driven system including an error budget
showing how each white noise error source contributes to the total system response.
Coloured noise inputs can be treated in a similar fashion by the use of appropriate
shaping filters.
3-3 Adjoint Construction Rules
From the above theory, it is clear that, for linear time varying systems, the
adjoint simulation method offers tremendous efficiencies over other methods
as a performance projection tool. This fact had been realised early on by many
weapons analysts and cleverly utilised in the preliminary design stages of many
current missile systems.
One of the difficulties with this approach is the construction of the adjoint system from
which the desired weighting function must be computed. Although several methods
for the construction process have appeared in the literature, by far the most popular
method has been that proposed by Laning and Battin [7). This method works directly
on the original simulation block diagram by application of a set of rules.
The adjoint construction rules are:
1. Convert all system inputs into equivalent systems driven by impulses or
white noise processes,
2. Reverse all signal flow directions,
3. Switch all system inputs to adjoint outputs and all system outputs to
adjoint inputs,
4. Change all summing points to branch points and all branch points to
summing points,
5. Replace time in all time varying elements with adjoint time f' where f" = f. - f,
6. Apply an impulse signal to the selected output of the original system,
7. At all the adjoint system stochastic output points, add the adjoint solution
sequence as illustrated in Figure 3 to yield the mean square value of the
system response.
3.4 Example of Adjoint System Construction
Traditionally, the construction of an adjoint system has been carried out manually by
following the above set of rules. As an example consider the following simple system
described by

=2- + (tF-tfy-u(t) v(0) = 2 (12)

dt

where the input is given by u(t) = s\n(t) and th denotes final time. This simple example
has two inputs and one output. Suppose we are interested in the sensitivity of y(t})
due to each of the inputs. The problem is then ideally suited to adjoint analysis.
Firstly, we re-cast the system into block diagram form. The block diagram
representation is given in Figure 4

y (0)- 2

u{t) 1 y{t)
+r ~\
- I 5

/, ,\2
VF

Figure 4. Block diagram representation of the system

After application of rule 1 in the adjoint construction process, the above block
diagram is converted to that shown in Figure 5. Here, the input 6(f) represents the
impulse function.
 6(0 »
6(0 v(0
,s-+l

it, -I)'

Figure 5 Modified block diagram following application of rule 1.

After applying the rest of the adjoint construction rules to the system block diagram
shown in Figure 5, we obtain the adjoint system associated with the original system.
This is given in block diagram form in Figure 6.

1 2
A
yitFiu 1 1 4 r i fie')
^ 2
,v +l S i

»- ,
1

Figure 6. Block diagram representation of the associated adjoint system for

deterministic input

It is clear from Figure 6 that the adjoint solution to the problem gives the sensitivity of
y(th) to each of the given inputs in one simulation run.
If, instead, the input to the system u(t) is stochastic in nature and represented as a zero
mean, white noise process characterised by the power spectral density <t>M, then the adjoint
simulation diagram has the form presented in Figure 7, where the mean and variance of
the output at time tb are are denoted by y (t ) and a2 (tF ), respectively.

y(ff)\,c

1
°,"M„
*0
1
s
<- ()- ^J A-

Figure 7. Block diagram representation of the associated adjoint system for

stochastic input

4. The Adjoint Software Tool

For very complex systems with many feedback paths, application of the adjoint rules
can be extremely tedious, time consuming and error prone. If the original simulation
block diagram is implemented in Simulink, then the process required to construct
the adjoint system can be conveniently automated to minimise these shortcomings. A
tool has been developed to achieve this. This tool, which forms part of a suite of tools
known as COVAD [50], is briefly described below.

4.1 COVAD Toolbox Overview

The COVAD toolbox has been designed as a general analysis package consisting of
a Graphical User Interface (GUI) built using the Matlab 2007b GUI API. Via the GUI
one is able to interact with Simulink models, such as a missile homing loop in order
to perform a number of analysis types including the adjoint method or the Monte
Carlo method.
The GUI also provides a means to store and retrieve the results of simulations
conducted in MATLAB/Simulink by storing them within a doc block (a built-in
Simulink block) that is added into a Simulink model.
Viewing the results of the simulation is facilitated by the GUI via a link to the
Matlab 2007b graph viewing facility. The GUI toolbox allows the user to filter from
the simulated results a suitable subset for which graphs may need to be generated
for analysis and then pipes this information across to the Matlab 2007b graph
viewing facility.
It should be noted that the GUI is built with the intention of supporting both the
novice and the advanced analyst. For the novice analyst, a wizard is used to guide
the user through the steps required to configure a model and to perform the desired
analysis type. As the model is configured, a display panel on the GUI is updated so
that a quick summary of the state of the model configuration is always easily viewable
by the analyst. For the advanced user, a scripting space has been introduced within
the GUI, where the analyst can write, within some guidelines, M-code that is able to
configure the model in a more flexible way. Even with this scripting facility, it has been
broken into steps so that after a gap between analysis efforts, the time to familiarise
oneself with progress so far is not tedious. There is even the ability to mix and match
between script and wizard, thanks to the modular design under the hood, employed
when configuring the model via steps.
While the toolbox is envisaged to work across different versions of Matlab, presently it
is able to function in 2007b and upwards only, since much of it was developed within
Matlab 2007b from scratch.

4.2 Use-Case for Adjoint Analysis

1. User invokes the COVAD GUI via the MATLAB command line.
2. User then creates a project. A project is a space where multiple models on
which different sets of related analyses are packaged together in a bundle.
3. Once the project is created, the user starts to populate the project with
Simulink models.
4. For each model, the user then selects the analysis type to be applied to that
model via a list of possible analysis options. With respect to this report, the
user would select the Fwd + Adjoint analysis option.
5. This then configures the wizard to cater for the particular analysis type. The
steps of the wizard will alter depending on the analysis combination chosen.

13
 6. Now the user walks through the wizard steps and then setup of the inputs,
outputs, and model parameters of the Simulink model subject to analysis via
the GUI. The setup can involve setting up the ranges for the variables that are
parameters for the various model blocks and selective enabling of the noise
and target manoeuvre inputs to be included in the model. Via a Create Adjoint
button found on one of these types, the user will invoke a script that will then
automatically build the adjoint for the selected model in accordance with the
rules described earlier in the paper.

7. Once this is complete, the user is able to go ahead and run the simulation on
the model via the Simulate button on the GUI. This will generate run data that
will be displayed on the GUI as well as being saved into Doc Block, added to
the model, which will serve as a repository of simulation run history for the
model with associated analysis type.
8. Finally the user can take the run results and then, depending on the data
collected, be able to select columns for the x and y axis to produce relevant
plots, e.g. miss distance versus flight time, for both forward and adjoint time
simulation runs and subsequent analysis.

4.3 Verification and Validation of the Toolbox

It is envisaged that the COVAD toolbox will be verified and validated against results
given in the open literature such as those given by Zarchan [2]. The verification and
validation steps are:
a. Compare the simulink homing loop block diagrams for various configurations
to those given by Zarchan [2] in both forward and adjoint time to verify that the
models have been built correctly.
b. Compare the forward run and adjoint run simulation results with
corresponding results given by Zarchan [2] for validation purposes.

i't
5. Missile Applications
Consider the planar missile-target engagement geometry depicted in Figure 8. Here we
have a radio frequency (RF) guided missile intercepting an incoming target.

Target

Missile

>'(/)

w^-

Figure 8. Missile target engagement geometry

Both missile and target speed are assumed constant. Initially, both missile and target
are on a collision course. The governing equations describing small perturbations
about the collision course are given as follows [2] (assuming a near head-on case).

By inspection of Figure 8, the relative acceleration is approximately

(13)

Similarly, the expression for the line of sight angle o can be approximated by

o =y I R(t) (14)

For a head-on case, the closing velocity V reduces to

V = V +V (15)

The linearised range equation R(t) is approximated by the expression

'S
 R(t)-Vc(tF-t) (16)

where tf denotes the total flight time of the engagement.

In this example, the quantity of interest is the miss distance. This is defined as the
relative separation between missile and target at the end of the flight, that is.

MD = y(tF) (17)

The above equations can be represented in block diagram form as shown in Figure 9

MD = y(tF)
i<

T
aT +
J, 1 y , 1 1
-O ) '
s s
r
K(*r-t)
W

aM

Figure 9. Block diagram representation of linearised geometry equations

For homing guidance, the missile relies on its RF seeker to track the target. Thus to
complete the homing loop in Figure 9, a model of the seeker is required. The signals
from the seeker are then used by the guidance computer to generate the guidance
commands for the autopilot. The autopilot stabilises the missile and deflects the fins
to cause the desired manoeuvre. The missile then reacts aerodynamically and alters
the flight path.

In this example, the seeker, autopilot and aerodynamic response of the missile
are modelled using transfer functions. The guidance law used is the proportional
navigation guidance law and has the form ac = N' Vr(j where N is the effective
navigation ratio, V is the closing velocity, O is the rate of change of sight line angle
and a denotes the missile commanded acceleration.

16
The complete linearised homing loop model for this example is represented in block
diagram form in Figure 10.
Radar noise, which contaminates the seeker's measurement of the line of sight to the
target, is added in the model in the form of glint and receiver (fading) noise. These are
modelled as zero mean white noise processors.

Ainopilol

Figure 10. Linearised homing loop model

The target model for this example is developed as follows. Prior to intercept, the target
is assumed to execute, at random, an evasive step manoeuvre of magnitude A. The
random time T for execution of the target manoeuvre is uniformly distributed between
zero and final homing time t, with probability density function given by

/',<'> = (18)
0 , otherwise

It has been shown in [24-26I, that the above target model can be represented by a linear
shaping filter driven by white noise. This model, which is shown in Figure 11, has the
same statistical characteristics, up to second order, as that based on equation 18. The
power spectral density of the zero mean, input white noise processor is given by

'7
 K 1
a
s

Figure 11. Shaping filter equivalent of random step manoeuvre

Table 1 summarises the parameter values used in this example.

Table 1. Nominal values of system parameters

Parameter Nominal Parameter Nominal Value

Value
Time of Flight, f, 5S Tgt Man Mag, A 3g
Seeker Time Const., Ts 0.05 s Glint Noise PSD, <H(. 2 m'/Hz
Filter Time Const., 7"„ 0.1 s Fading Noise PSD, <I>( io' rad'/Hz
Autopilot Time Const., TA 0.1 s PN Ratio, N 4

After inclusion of the target model into the homing loop block diagram of Figure 10
and entering the parameter values, given in Table i, into COVAD, we obtain the GUI
information displayed in Figure 12.

M11MMWM
11 «» ir
NomScucat
PSD Te M 0S—
(0 m
I—<J»»J«"~ EJOMWM 2 0
©T»tl O'TOJ 1. [_»«_]
O'IP"
ONOI»
I "* A(m«) ^•VX 0
"1 " 1 "1 ' 1
* 1 » |»a| or |
'- ••••

In 0.

Block Diagram
QHM**Errei
[ |RM»Hrr.
m K.

II (i [X]
[W]
mfnrift] | WO ] on
©w O*-*

3 C

Figure 12. Parameter values entered into COVAD

18
The preview pane in the figure displays a functional representation of the entered data
and is there to provide the user with an indication of the structure of the simulation
model that will be built in block diagram form in Simulink. Once the Next button is
pressed, the Simulink model representation of this data is automatically generated.
This is shown in Figure 13.

•rfTlW Ed^:
->©- -M5-
•9-
JJ Miss Distance

-•CD

Figure 13. Simulink implementation of the missile homing loop

The adjoint toolbox in COVAD may now be invoked to automatically convert the
Simulink model of the original system into an associated Simulink adjoint system for
further analysis. The Simulink adjoint system is shown in Figure 14.

Note that the impulse response of the adjoint system has been computed by imposing
an initial condition of unity on integrator in the diagram. Standard Simulink tools
can now be utilised for further analysis of the system.

Figure 14. Corresponding adjoint system in Simulink

19
Figure 15 compares the results from the adjoint simulation with those generated using
a Monte Carlo method. The results give the RMS miss distance of the engagement as
a function of flight time. The adjoint results were obtained in one simulation run of
the Simulink system. The Monte Carlo results are based on the physical model of the
target manoeuvre (as opposed to the shaping filter approach required for the adjoint
model) and required 200 runs for each value of flight time ff considered.

i 3 MC set 200
Adjoint

0
0 12 3
tF (s)

Figure 15. RMS miss distance comparison - Adjoint vs Monte Carlo results

In addition to the total RMS miss distance results, the adjoint solution also provides,
within the same simulation run, an error budget consisting of the contributions that
each input has on total system performance. The error budget information is displayed
in Figure 16.

4
?
~ Tgt Man
S3 G lint Noise
5 Fading Noise

0
0 12 3
tF (s)

Figure 16. Adjoint solution generates error budget

It is clear from the error budget plot that, in this example, glint noise is the biggest
contributor to miss distance.

As another example, let us now use the software to investigate the performance
of the missile against other possible target manoeuvres. If instead of the random
step manoeuvre, the target executes a sinusoidal manoeuvre at some time prior to
intercept, where, as previously, the time Tat which the manoeuvre is initiated is
assumed to be uniformly distributed over the interval | o, t, | with probability density
function given by equation (18), then such a manoeuvre can be modelled as shown
in Figure 17.

u. CO
a,
2 2
S +0)

Figure 17. Random sinusoidal target manoeuvre model.

The parameter <o appearing in the figure denotes the frequency, in rad/s, of the
sinusoidal target manoeuvre. The RMS miss distance adjoint simulation results for
this form of the target manoeuvre are summarised in Figure 18 for selected values
of the parameter eo . Note that all other nominal values for the system parameters as
given by table 1 remain the same here.

1 6
u= 1
1.4 (0=2
03= 3
1.2

1
1
I 0.8
M
1 0.6
0.4

0 2

0
12 3 4 5
tF(s)

Figure 18. Random sinusoidal target manoeuvre effects on miss distance

21
As a final example, we investigate the performance of the interceptor when the
target executes a random telegraph manoeuvre. The random telegraph manoeuvre
represents a policy, starting at time zero, in which the target executes either a
maximum positive or negative acceleration such that the number of sign changes
per second follows a Poisson distribution, and the average number of sign changes
is u per second [26]. A typical realisation of a 3g Poisson target manoeuvre is shown
in Figure 19.

40

20

-20

-40

Time (s)

Figure 19. Random telegraph target manoeuvre

The shaping filter equivalent of a Poisson target manoeuvre can be represented by
white noise through a low pass filter as shown in Figure 20 [61].

U. + •"> >
1
J • 2v
s

Figure 20. Shaping filter equivalent of random telegraph target manoeuvre

In the Figure, u represents the average number of sign changes while |} denotes the
magnitude of the target manoeuvre, in this case, 3g. Also in the figure, the white noise
input has spectral density

<*>„=£- (19)
V

Using the same parameter values as previously, the adjoint block diagram can be
readily constructed using COVAD. The results of the adjoint simulation for selected
values of the parameter u are given in Figure 21.

1.4
v= .1
1.2 v= .3
v=3
v=.5

_
I OS

£ 0.6

0.4

0
0 12 3 4
tF(s)

Figure 21, Random telegraph manoeuvre effects on miss distance

»3
 6. Conclusion
Mathematicians have long been paying attention to the adjoint operator and duality
theorems in their studies of the existence of solutions to differential equations. In this
report, the adjoint simulation method has been employed to investigate the response
and sensitivity of linear time varying systems to a combination of deterministic
and stochastic inputs. MATLAB/Simulink software has been developed to facilitate
applications of the method to problems encountered in Aerospace/Defence science.

24
Acknowledgements
This study was carried out under the Defence Science and Technology Organisation
(DSTO) Fellowship Program. The author wishes to acknowledge the following for
their support and encouragement: Dr Martin Weiss of TNO, The Netherlands, for
his constructive comments and advice on aspects of the adjoint method, Mr Arvind
Rajagopalan of Weapons Systems Division (WSD), for software development support,
DSTO (including Chief Defence Scientist Advisory Committee (CDSAC), Chief Weapons
Systems Division (CWSD) and Task Manager - Dr Gulay Mann) for the opportunity and
support to carry out this work.

References
[1] Taylor, J.H., "Statistical Performance Analysis of Nonlinear Stochastic Systems
by the Monte Carlo Method," Mathematics & Computers in Simulation, Vol 23,
1981.

[2] Zarchan, P., "Tactical and Strategic Missile Guidance," 1s1 Ed. Progress in
Astronautics and Aeronautics, Vol. 124, AIAA Inc. Washington DC, 1990.
[3] Moulton, F.R.,"New Methods in Exterior Ballistics," The University of Chicago
Press, Chicago, Illinois, 1926.
[4] Bliss, G.A. "Mathematics for Exterior Ballistics," John Wiley & Sons,
Inc., New York, 1944.
[5I Bennett, R.R., "Analog Computing Applied to Noise Studies," Proceedings of
the I.R.E., Oct 1953.
|6| Laning, J.H. & Battin, R.H., "An Application of Analog Computers to the
Statistical Analysis of Time and Variable Networks," I.R.E. Transactions - Circuit
Theory, Mar 1955.
[7] Laning, J.H. & Battin, R.H., "Random Processes in Automatic Control,"
McGraw-Hill Book Company, New York, NY, 1956.
[8] Rogers, A.E. & Connolly, T.W., "Analog Computation in Engineering Design,"
McGraw-Hill Book Company, Inc., NY, i960.
[9] Fifer, S., "Analogue Computation, Theory, Techniques and Applications,"
McGraw-Hill Book Company, Inc., NY, 1961.
 [IO] Leondes, C.T., "Computer Control Systems Technology," Ch. 6, "Analog-
Computer Theory," by Pfeffer, I., McGraw-Hill Book Company, Inc., NY, 1961.

[11] Thorsen, R.A., "The Adjoint Analog ComputingMethod," Master's Thesis,

San Diego State College, Jan 1966.

[12] Peterson, E.L., "Statistical Analysis and Optimization of Systems,"

lohn Wiley & Sons, Inc., NY, 1961.

[13) Machol, R.E., "System Engineering Handbook," Ch. 19, "Guidance" by Howe,
R.M., McGraw-Hill Book Company, New York, NY, 1965.

[14] Tarrant, G.A., " The Method of Adjoint Systems and its Application to Guided
Missile Noise Studies," Report PB186191, EASAMS, Camberly, UK, 1969.
[15] Huskey, H.D. & Korn, G.A., "Computer Handbook," Ch. 5 "Part 9: Random
Process Studies," by Rideout, V.C., McGraw-Hill Book Company, Inc., NY, 1962.
[16] Levine, L., "Methods for Solving Engineering Problems Using Analog
Computers," McGraw-Hill Book Company, Inc., NY, 1964.
[17] Korn, G.A., "Random Process Simulation and Measurements,"
McGraw-Hill Book Company, Inc., NY, 1966.
[18] Sussman, R., "A Method of Solving Linear Problems by Using the Adjoint
System," Memo No. UCB/ERL M-2, Electronics Research Laboratory, University
of California, Berkeley, CA, Feb 1962.
[19] DeRusso, P.M., Roy, R.J. & Close, CM., "State Variables for Engineers,"
John Wiley & Sons, Inc., NY, 1967.
[20] Kailath, T., "Linear Systems," Prentice-Hall, Inc., Englewood Cliffs, NJ, 1980.
[21] Sarachik, P.E. & Kreindler, E., "Concerning Adjoints of Discreet-Time Systems,"
IEEE Transactions on Automatic Control, Vol. AC-10, pp. 350-352, Jul 1965.
[22] Brogan, W.L., "Performance Analysis of Continuous, Sampled and Multirate-
Sampled Systems with Random Inputs," IEEE Transactions on Automatic
Control, pp. 599-601, Oct 1967.
[23] Willems, G.C., "An Improved Approach to the Adjoint Method Applied to the
Statistical Analysis of a Homing Missile," Report No. RE-TR-67-8, US Army
Missile Command, Redstone Arsenal, AL, Aug 1967.

26
[24] Fitzgerald, R.J. & Zarchan, P., "Shaping Filters for Randomly Initiated Target
Maneuvers," A1AA Guidance & Control Conference, Paper No. 78-1304, Palo Alto,
CA, 1978.

[25] Fitzgerald, R.J., "Shaping Filters for Disturbances with Random Starting Times,"
A1AA lournal of Guidance & Control, Vol. 2, No. 2,1979.

[26] Zarchan, P., "Representation of Realistic Evasive Maneuvers by the Use of

Shaping Filters," AIAA Journal of Guidance & Control, Vol. 2, No. 4,1978.

I27I Zarchan, P., "Complete Statistical Analysis of Nonlinear Missile Guidance

Systems - SLAM," AIAA Journal of Guidance & Control, Vol. 2, No. 1,1979.

[28] Bibel, J.E., "Application of the Adjoint Method in Homing Missile Performance
Analysis," NSWC Technical Report TR 88-311, May 1990.

[29] Itohara, H. & Miwa, S., "Non-Gaussian Clutter Effect on the Miss Distance of a
Radar Homing Missile," AIAA Conference, AIAA-92-4536-CP, 1992.

[30] Miwa, S., "Radome Effect on the Miss Distance of a Radar Homing Missile,"
Electronics & Communications in Japan, Part 1, Vol. 81, No. 7, pp. 516-523,1998.

[31] Fourie, J.P., "On the Conceptual Design of the Control of a Nonlinear System,"
Master of Engineering, University of Pretoria, Pretoria, Nov 1991.

[32] Eltohamy, K., "Adjoint Methods for Missile Guidance Systems," Master of
Science, Arizona State University, Arizona, Dec 1993.

[33] Bucco, D., "Adjoints Revisited: A Software Tool to Facilitate Their Application,"
AIAA GNC Conference, New Orleans, USA, Aug 1997.

[34] Bucco, D., "Computerised Techniques for Assessing the Terminal Performance of
RF Guided Missiles," Third International Conference on Modeling & Simulation,
MS'97, Melbourne, Australia, Oct 1997.

[35] Weiss, M., "Adjoint Method for Missile Performance Analysis on State-Space
Models," Journal of Guidance, Control and Dynamics, Vol. 28, No. 2, pp. 236-248,
2005.

136] Weiss, M. & Bucco, D., "Handover Analysis for Tactical Guided Weapons Using the
Adjoint Method," AIAA GN&C Conference, San Francisco, CA, USA, Aug 2005.

[37] Weiss, M. & Bucco, D., "Evaluation Method for Dual Phase Guided Weapons
Based on the Adjoint Method," AIAA GN&C Conference, Hilton Head, South
Carolina, USA, Aug 2007.

*7
 [38] Weibel, S.P. & Cottrell, R.G., "Reviving Adjoints as an Analysis Tool in Missile
Defence Interceptor Requirements Definition," Second AAAF International
Conference on Missile Defence, 2004.

[39] Moorman, M.J., Warkomski.E.J., Lam, M.Q. & Elkanick, M.E., "Extending Adjoint
Simulation Beyond its Traditional Role: Accounting for Kalman Tracker Dynamic
Interaction & Beyond," AIAA GNC Conference, San Francisco, USA, Aug 2005.

[40] Lin, CM., Hsu, C.F., Chang, S.K. & Wai, R.J., "Guidance Law Evaluation for
Missile Guidance Systems," Asian Journal of Control, Vol. 2, No. 4, pp. 243-250,
Dec 2000.

[41] Rogers, S., "Missile Guidance Comparison," AIAA GN&C Conference,

Providence, Rhode Island, Aug 2004.

[42] Weiss, M. & Van De Beld, D., "A Generalized Shaping Filter Method for Higher
Order Statistics," Probabilistic Engineering Mechanics, Vol. 22, pp. 313-319, 2007.
[43] Weiss, M., Rol, M., Falkena, W. & Scherer, C, "Guidance Performance Analysis
In the Presence of Model Uncertainties," AIAA GN&C Conference, Hilton Head,
South Carolina, USA, Aug 2007.
[44] Weiss, M., "Robust Guidance Performance Analysis Against Weaving Targets,"
AIAA GN&C Conference, Honolulu, Hawaii, USA, Aug 2008.
[45] Denggao, J., Fenghua, H. & Yu, Y., "Finite Time Li Approach for Missile Overload
Requirement Analysis in Terminal Guidance," Chinese Journal of Aeronautics,
Vol. 22, pp. 413-418, 2009.
[46] Lum, K.Y., "Infinite-Dimensional Linearization and Extended Adjoint Method for
Nonlinear Homing Loop Analysis," AIAA GN&C Conference, Austin, Texas, USA,
Aug 2003.

[47] Weiss, M., "A Novel Concept for Scheduling and Effect Assessment of Soft Kill
Against an Antiship Missile Based on the Adjoint Method," Defense, Security &
Safety, TNO, Rijswijk, The Netherlands, Private Communication, Feb 2009.

[48] Vermeulen, A. & Maes, G., "Missile Avoidance Manoeuvres with Simultaneous
Decoy Deployment," AIAA GNC Conference, Chicago, IL, Aug 2009.
[49] Weiss, M. & Bucco, D., "The State-Space Approach to the Method of Adjoints for
Hybrid Guidance Loop Models," AIAA GN&C Conference, Chicago, IL,
USA, 2009.

28
[50] Bucco, D. & Weiss, M., "Development of a Matlab/Simulink Tool to Facilitate
System Analysis and Simulation via the Adjoint and Covariance Methods," AIAA
M&ST Conference, Hilton Head, South Carolina, Aug 2007.
[51] Bucco, D. & Weiss, M., "Further Enhancements to COVAD: A MATLAB Tool Based
on Covariance/Adjoint Methods," AIAA M&ST Conference, Honolulu, Hawaii, Aug
2008.

[52] Bucco, D. & Weiss, M., "Extending the COVAD Toolbox to Accommodate System
Non-Linearities," AIAA M&ST Conference, Chicago, IL, Aug 2009.
[53] Rajagopalan, A. & Bucco, D., "Applications of Adjoint Theory to Problems
in Aerospace/Defence Science," Engineering Mathematics and Applications
Conference, Adelaide, Australia, Dec 2009.
[54) Swee, J.C.S., "Missile Terminal Guidance and Control Against Evasive Targets,"
MSc Thesis, Naval Postgraduate School, Monterey, CA, Mar 2000.
[55I Pathirana, P.N., "Modern Robust and Computer Control Applied to Guidance,"
PhD Thesis, The University of Western Australia, WA, Australia, Nov 2001.
[56] Lee, K.H., "Tactical Evasive Maneuver of Missile for Survivability Enhancement,"
ME Thesis, Korea Advanced Institute of Science and Technology, FDCL MT-033,
Korea, Feb 2003.
[57] Han, P., "Adjoint Method and Missile Guidance Control," MSc Thesis, Royal
Melbourne Institute of Technology University, Melbourne, Australia, Nov 2005.
[58] Doff, E.)., "The Development of an Adjoint Toolbox for Matlab," Thesis, Royal
Netherlands Naval Academy, The Netherlands, Nov 2006
[59] Rol, M., "Performance Analysis of Guidance Loops under Model Uncertainty,"
MSc Thesis, Delft University of Technology, The Netherlands, )ul 2007.
[6o| Zarchan, P., "Fundamentals of Tactical & Strategic Missile Guidance Short
Course," AIAA Professional Studies Series, Washington DC, US, 1995.
(61] Zarchan, P., "Tactical and Strategic Missile Guidance," 5th Ed. Progress in
Astronautics and Aeronautics, Vol. 124, AIAA Inc. Washington DC, 2007.
[62] Ince, E.L., "Ordinary Differential Equations," Dover Publication, Inc.,
New York, 1956.
Page classification: UNCLASSIFIED

2<)