Derivation DF DT

Mathematical
Problems in
Engineering
Theory, Methods, and Applications
Special Issue
Mathematical Methods Applied to the Celestial Mechanics
of Artificial Satellites
Guest Editors: Antonio F. Bertachini A. Prado, Maria Cecilia Zanardi,

Tadashi Yokoyama, Silvia Maria Giuliatti Winter, and Josep J. Masdemont
Hindawi Publishing Corporation

http://www.hindawi.com
Mathematical Methods Applied to
the Celestial Mechanics of
Artificial Satellites
Mathematical Problems in Engineering
Mathematical Methods Applied to

the Celestial Mechanics of
Artificial Satellites
Guest Editors: Antonio F. Bertachini A. Prado,
Maria Cecilia Zanardi, Tadashi Yokoyama,
Silvia Maria Giuliatti Winter, and Josep J. Masdemont
Copyright q 2012 Hindawi Publishing Corporation. All rights reserved.
This is a special issue published in “Mathematical Problems in Engineering.” All articles are open access articles distributed
under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any
medium, provided the original work is properly cited.
Editorial Board
Eihab Abdel-Rahman, Canada Elmetwally Elabbasy, Egypt Nikolaos Kazantzis, USA
Rashid K. Abu Al-Rub, USA Alex Elias-Zuniga, Mexico Farzad Khani, Iran
Salvatore Alfonzetti, Italy Anders Eriksson, Sweden Kristian Krabbenhoft, Australia
Igor Andrianov, Germany Vedat Suat Erturk, Turkey Jurgen Kurths, Germany
Sebastian Anita, Romania Qi Fan, USA Claude Lamarque, France
W. Assawinchaichote, Thailand Moez Feki, Tunisia F. Lamnabhi-Lagarrigue, France
Erwei Bai, USA Ricardo Femat, Mexico Marek Lefik, Poland
Ezzat G. Bakhoum, USA Rolf Findeisen, Germany Stefano Lenci, Italy
José Manoel Balthazar, Brazil R. A. Fontes Valente, Portugal Jian Li, China
Rasajit K. Bera, India C. R. Fuerte-Esquivel, Mexico Shihua Li, China
Jonathan N. Blakely, USA Zoran Gajic, USA Shanling Li, Canada
Stefano Boccaletti, Spain Ugo Galvanetto, Italy Tao Li, China
Daniela Boso, Italy Xin-Lin Gao, USA Ming Li, China
M. Boutayeb, France Furong Gao, Hong Kong Teh-Lu Liao, Taiwan
Michael J. Brennan, UK Behrouz Gatmiri, Iran P. Liatsis, UK
John Burns, USA Oleg V. Gendelman, Israel Shueei M. Lin, Taiwan
Salvatore Caddemi, Italy Didier Georges, France Jui-Sheng Lin, Taiwan
Piermarco Cannarsa, Italy Paulo Batista Gonçalves, Brazil Yuji Liu, China
Jose E. Capilla, Spain Oded Gottlieb, Israel Wanquan Liu, Australia
Carlo Cattani, Italy Quang Phuc Ha, Australia Bin Liu, Australia
Marcelo M. Cavalcanti, Brazil Muhammad R. Hajj, USA Fernando Lobo Pereira, Portugal
Diego J. Celentano, Chile Thomas Hanne, Switzerland Paolo Lonetti, Italy
Mohammed Chadli, France Katica R. Hedrih, Serbia Vassilios C. Loukopoulos, Greece
Arindam Chakraborty, USA C. Cruz Hernandez, Mexico Junguo Lu, China
Yong-Kui Chang, China M.I. Herreros, Spain Chien-Yu Lu, Taiwan
Michael J. Chappell, UK Wei-Chiang Hong, Taiwan Alexei Mailybaev, Brazil
Xinkai Chen, Japan J. Horacek, Czech Republic Manoranjan Maiti, India
Kui Fu Chen, China Chuangxia Huang, China Oluwole D. Makinde, South Africa
Kue-Hong Chen, Taiwan Gordon Huang, Canada Rafael Martinez-Guerra, Mexico
Jyh Horng Chou, Taiwan Yi Feng Hung, Taiwan Bohdan Maslowski, Czech Republic
Slim Choura, Tunisia Hai-Feng Huo, China Driss Mehdi, France
Swagatam Das, India Asier Ibeas, Spain Roderick Melnik, Canada
Filippo de Monte, Italy Anuar Ishak, Malaysia Xinzhu Meng, China
M. do R. de Pinho, Portugal Reza Jazar, Australia Yuri Vladimirovich Mikhlin, Ukraine
Antonio Desimone, Italy Zhijian Ji, China Gradimir V. Milovanović, Serbia
Yannis Dimakopoulos, Greece J. Jiang, China Ebrahim Momoniat, South Africa
Baocang Ding, China J. J. Judice, Portugal Trung Nguyen Thoi, Vietnam
Joao B. R. Do Val, Brazil Tadeusz Kaczorek, Poland Hung Nguyen-Xuan, Vietnam
Daoyi Dong, Australia Tamas Kalmar-Nagy, USA Ben T. Nohara, Japan
Balram Dubey, India Tomasz Kapitaniak, Poland Anthony Nouy, France
Horst Ecker, Austria Hamid Reza Karimi, Norway Sotiris K. Ntouyas, Greece
M. Onder Efe, Turkey Metin O. Kaya, Turkey Gerard Olivar, Colombia
Claudio Padra, Argentina Mohamed A. Seddeek, Egypt Yongqi Wang, Germany
Francesco Pellicano, Italy Alexander P. Seyranian, Russia Moran Wang, USA
Vu Ngoc Phat, Vietnam Leonid Shaikhet, Ukraine Yijing Wang, China
A. Pogromsky, The Netherlands Cheng Shao, China Cheng C. Wang, Taiwan
Seppo Pohjolainen, Finland Daichao Sheng, Australia Gerhard-Wilhelm Weber, Turkey
Stanislav Potapenko, Canada Tony Sheu, Taiwan Jeroen A.S. Witteveen, USA
Sergio Preidikman, USA Zhan Shu, UK Kwok-Wo Wong, Hong Kong
Carsten Proppe, Germany Dan Simon, USA Zheng-Guang Wu, China
Hector Puebla, Mexico Luciano Simoni, Italy Ligang Wu, China
Justo Puerto, Spain Grigori M. Sisoev, UK Wang Xing-yuan, China
Dane Quinn, USA Christos H. Skiadas, Greece X. Frank XU, USA
Rubén R. Garcı́a, Spain Davide Spinello, Canada Xuping Xu, USA
K. R. Rajagopal, USA Sri Sridharan, USA Jun-Juh Yan, Taiwan
Gianluca Ranzi, Australia Rolf Stenberg, Finland Xing-Gang Yan, UK
Sivaguru Ravindran, USA Changyin Sun, China Suh-Yuh Yang, Taiwan
G. Rega, Italy Jitao Sun, China Mahmoud T. Yassen, Egypt
Pedro Ribeiro, Portugal Xi-Ming Sun, China Mohammad I. Younis, USA
J. Rodellar, Spain Andrzej Swierniak, Poland Huang Yuan, Germany
R. Rodriguez-Lopez, Spain Allen Tannenbaum, USA S. P. Yung, Hong Kong
A. J. Rodriguez-Luis, Spain Cristian Toma, Romania Ion Zaballa, Spain
Ignacio Romero, Spain Irina N. Trendafilova, UK Arturo Zavala-Rio, Mexico
Hamid Reza Ronagh, Australia Alberto Trevisani, Italy Ashraf M. Zenkour, Saudi Arabia
Carla Roque, Portugal Jung-Fa Tsai, Taiwan Yingwei Zhang, USA
Mohammad Salehi, Iran Kuppalapalle Vajravelu, USA Xu Zhang, China
Miguel A. F. Sanjuán, Spain Victoria Vampa, Argentina Liancun Zheng, China
Ilmar Ferreira Santos, Denmark Josep Vehi, Spain Jian Guo Zhou, UK
Nickolas S. Sapidis, Greece Stefano Vidoli, Italy Zexuan Zhu, China
Bozidar Sarler, Slovenia Xiaojun Wang, China Mustapha Zidi, France
Andrey V. Savkin, Australia Dan Wang, China
Massimo Scalia, Italy Youqing Wang, China
Contents
Mathematical Methods Applied to the Celestial Mechanics of Artificial Satellites,
Antonio F. Bertachini A. Prado, Maria Cecilia Zanardi, Tadashi Yokoyama,
Silvia Maria Giuliatti Winter, and Josep J. Masdemont
Volume 2012, Article ID 979752, 7 pages
A Numerical Study of Low-Thrust Limited Power Trajectories between Coplanar Circular

Orbits in an Inverse-Square Force Field, Sandro da Silva Fernandes,
Carlos Roberto Silveira Filho, and Wander Almodovar Golfetto
Application of the Hori Method in the Theory of Nonlinear Oscillations,

Sandro da Silva Fernandes
IMU Fault Detection Based on x2 -CUSUM, Élcio Jeronimo de Oliveira, Hélio Koiti Kuga,
and Ijar Milagre da Fonseca
Low-Thrust Orbital Transfers in the Two-Body Problem, A. A. Sukhanov and A. F. B. A. Prado

Four-Impulsive Rendezvous Maneuvers for Spacecrafts in Circular Orbits Using Genetic

Algorithms, Denilson Paulo Souza dos Santos, Antnio Fernando Bertachini de Almeida Prado,
and Guido Colasurdo
Semianalytic Integration of High-Altitude Orbits under Lunisolar Effects, Martin Lara,

Juan F. San Juan, and Luis M. López
Error Modeling and Analysis for InSAR Spatial Baseline Determination of Satellite Formation
Flying, Jia Tu, Defeng Gu, Yi Wu, and Dongyun Yi
Analysis of Attitude Determination Methods Using GPS Carrier Phase Measurements,

Leandro Baroni and Hélio Koiti Kuga
Numerical Analysis of Constrained, Time-Optimal Satellite Reorientation, Robert G. Melton

Low-Thrust Out-of-Plane Orbital Station-Keeping Maneuvers for Satellites, Vivian M. Gomes

and Antonio F. B. A. Prado
An Adaptive Remeshing Procedure for Proximity Maneuvering Spacecraft Formations,

Laura Garcia-Taberner and Josep J. Masdemont
Closed Relative Trajectories for Formation Flying with Single-Input Control, Anna Guerman,
Michael Ovchinnikov, Georgi Smirnov, and Sergey Trofimov
Modified Chebyshev-Picard Iteration Methods for Station-Keeping of Translunar Halo Orbits,

Xiaoli Bai and John L. Junkins
Assessment of the Ionospheric and Tropospheric Effects in Location Errors of Data Collection
Platforms in Equatorial Region during High and Low Solar Activity Periods,
Áurea Aparecida da Silva, Wilson Yamaguti, Hélio Koiti Kuga, and Cláudia Celeste Celestino
Study of Stability of Rotational Motion of Spacecraft with Canonical Variables,

William Reis Silva, Maria Cecı́lia F. P. S. Zanardi, Regina Elaine Santos Cabette,
and Jorge Kennety Silva Formiga
Analysis of Filtering Methods for Satellite Autonomous Orbit Determination Using Celestial
and Geomagnetic Measurement, Xiaolin Ning, Xin Ma, Cong Peng, Wei Quan,
and Jiancheng Fang
The Orbital Dynamics of Synchronous Satellites: Irregular Motions in the 2 : 1 Resonance,

Jarbas Cordeiro Sampaio, Rodolpho Vilhena de Moraes, and Sandro da Silva Fernandes
Application of Periapse Maps for the Design of Trajectories Near the Smaller Primary in
Multi-Body Regimes, Kathleen C. Howell, Diane C. Davis, and Amanda F. Haapala
Unscented Kalman Filter Applied to the Spacecraft Attitude Estimation with Euler Angles,
Roberta Veloso Garcia, Helio Koiti Kuga, and Maria Cecilia F. P. S. Zanardi
Optimal Two-Impulse Trajectories with Moderate Flight Time for Earth-Moon Missions,
Sandro da Silva Fernandes and Cleverson Maranhão Porto Marinho
Comparison between Two Methods to Calculate the Transition Matrix of Orbit Motion,
Ana Paula Marins Chiaradia, Hélio Koiti Kuga,
and Antonio Fernando Bertachini de Almeida Prado
Higher-Order Analytical Attitude Propagation of an Oblate Rigid Body under

Gravity-Gradient Torque, Juan F. San-Juan, Luis M. López, and Rosario López
doi:10.1155/2012/979752
Editorial
Mathematical Methods Applied to the
Celestial Mechanics of Artificial Satellites
Antonio F. Bertachini A. Prado,1 Maria Cecilia Zanardi,2

Tadashi Yokoyama,3 Silvia Maria Giuliatti Winter,2
and Josep J. Masdemont4
1
INPE-DMC, Avenida dos Astronautas 1758, 12227-010 São Jose dos Campos, SP, Brazil
2
Universidade Estadual Paulista (UNESP), Campus de Guaratinguetá, CEP 12516-410 Guaratinguetá,
SP, Brazil
3
Universidade Estadual Paulista, Campus de Guaratinguetá, Caixa Postal 178, 13500-970 Rio Claro,
SP, Brazil
4
IEEC and Departament de Matemàtica Aplicada I, Universitat Politècnica de Catalunya, Diagonal 647,
08028 Barcelona, Spain
Correspondence should be addressed to Antonio F. Bertachini A. Prado, prado@dem.inpe.br
Received 9 May 2012; Accepted 9 May 2012
Copyright q 2012 Antonio F. Bertachini A. Prado et al. This is an open access article distributed
under the Creative Commons Attribution License, which permits unrestricted use, distribution,
and reproduction in any medium, provided the original work is properly cited.
Celestial mechanics is a science that comes from the more general field of astronomy. It is
devoted to the study of the motion of the planets, moons, asteroids, comets, and other celestial
bodies. Some of the main researchers involved in this field are well-known names in history,
like Johannes Kepler and Isaac Newton, who wrote the basic laws that govern these motions.
With the advances of technology, the world entered the so-called “Space Age,” where
artificial artifacts started to be built and launched into space. To perform those tasks, the
same laws used to describe the motion of the natural celestial bodies can be used to study the
motion of those manmade spacecrafts. Although it is not the beginning of the space activities,
the launch of the satellite Sputnik in 1957, by the former Soviet Union, is an important mark
of this age. This launch was then followed by the United States of America USA that, among
other space activities, landed a man on the Moon in 1969.
Nowadays, after those first historical events, the activities related to artificial satellites
are one of the most important fields in science and technology. It includes several important
applications of engineering that improved the life of everybody in the entire world.
Communications by satellite are well established for many years, and it is hard to imagine
a world without this capability. Another field that has been helped by the space activities is
the field related to the exploration of resources on Earth. Images from satellites can generate
important maps to find their locations, and this can make the difference for a sustainable use
of our resources. The use of location devices based on the GPS constellation is also everyday
2 Mathematical Problems in Engineering
more common. Several other similar services will be soon provided by Europe, Russia, and
China. The applications of this type of service include the monitoring of valuables under
transport and the search for the fast way to reach a destination. Artificial satellites are also
used to make weather forecast, which can contribute to the very important task of natural
disaster prediction. It is not only possible to help agricultural development by predicting the
behavior of the weather, but it is also possible to anticipate large-scale phenomenon such as
the occurrence of tsunamis. These activities is extremely important for everybody on Earth.
The “Mathematical Methods Applied to the Celestial Mechanics of Artificial Satel-
lites,” the title of the present issue of the journal, has a key role in those activities. One of the
tasks to be performed when planning a satellite mission is to find the most adequate orbit to
place this satellite. After this first step, it is also necessary to study practical problems related
to the maintenance of its nominal orbit and how much and when it will be modified by the
natural forces present in nature. These questions can be addressed by analyzing the position
of the spacecraft. There are also several questions regarding the orientation of the spacecraft.
It is necessary to know its orientation and how to control it in order to plan the space missions.
Those topics are all covered in the list of papers published in the present volume.
In that scope, this special issue is focused on the recent advances in mathematical
methods applied to the celestial mechanics of artificial satellites. It has a total of twenty two
papers briefly described below.
Eight of them are concerned with the problems of optimal maneuvers. “Low-thrust
orbital transfers in the two-body problem” is written by A. A. Sukhanov and A. F. B. A. Prado.
Although several methods for optimization of multirevolution orbital transfers are available
in the literature, most of them are rather complicated, with limited usage due to a variety
of assumed hypothesis. In this paper, based on a convenient linearization near a reference
orbit, the authors propose a very simple method for low-thrust transfers power limited.
The method is free from any averaging concepts, and no limit on the number of orbits around
the attracting center nor any constraint on the thrust direction is assumed. Several additional
advantages are outlined in the paper.
“Four-impulsive rendezvous maneuvers for spacecraft in circular orbits using genetic algo-
rithms” is written by D. P. S. dos Santos et al.. In this paper, the problem of rendezvous is
considered, which is necessary to transfer a spacecraft from one orbit to another, but with
the extra constraint of meeting another spacecraft when reaching the final orbit. This work
analyzes optimal rendezvous maneuvers between two coplanar circular orbits, seeking to
perform this transfer with low fuel consumption, by assuming a time-free problem and using
four burns during the process. The genetic algorithm represents a new alternative that can
be used for comparisons with the results obtained by standard procedures available in the
literature. The results show that this technique brings good results for the proposed four-
impulsive rendezvous maneuvers, when compared with the ones obtained by the traditional
impulsive method. Therefore, this approach can be used in real cases, especially when a bi-
impulsive transfer is not possible due to the limitations of the engine of the spacecraft.
Another paper concerned with the spacecraft maneuvers is “Low-thrust out-of-plane
orbital station-keeping maneuvers for satellites,” written by V. M. Gomes and A. F. B. A. Prado.
This paper considers the problem of-out-of plane orbital maneuvers for station keeping of
satellites. The main idea is to consider an Earth satellite in a disturbed orbit, and it is necessary
to perform a small amplitude orbit correction to return the satellite to the nominal orbit,
to keep it performing its mission. Optimal control is used to build an algorithm to search
for solutions for the problem of minimum fuel consumption to make the required orbital
Mathematical Problems in Engineering 3
maneuvers. The problem takes into account the accuracy in the constraint’s satisfaction.
The adjustments made in the algorithm and the used parameters allow the convergence in
most of the analyzed cases. The results show that it is possible to reduce the costs by
exploring tolerable errors in the constraint’s satisfaction. It can also be observed that the
increasing of the number of propulsion arcs decreases the fuel costs. The reason for this
fact is that increasing this number causes an increase in the degree of freedom available for
the optimization technique. Since the maneuvers have small amplitudes, this increase in the
burning arcs has a limit in the savings, so the fuel consumption reaches constant values after
a certain value of the number of arcs.
“An adaptive remeshing procedure for proximity maneuvering spacecraft formations” is
written by L. Garcia-Taberner and J. J. Masdemont. This paper searches for an optimal way to
obtain reconfigurations of spacecraft formations. It uses a discretization of the time interval
and obtains local solutions by applying variational method. Applications are made for
libration point orbits, trying to find optimal meshes using an adaptive remeshing procedure
and on the determination of the parameter that governs it.
Next, “Closed relative trajectories for formation flying with single-input control” is written
by A. Guerman et al.. The authors study the problem of formation shape control under a
constrained thrust direction. It is considered a formation with two satellites and a linear
model for the relative motion including J2 is used. It is also assumed that the satellites
are equipped with a passive attitude control system that provides one-axis stabilization.
The propulsion system consists of one or two thrusters oriented along the stabilized axis.
Based on those assumptions, the authors prove that, in both cases passive magnetic attitude
stabilization and spin stabilization and for all initial relative positions and velocities of the
satellites, there are controls that guaranty their periodic relative motion.
Another paper of this topic is “Modified Chebyshev-Picard iteration methods for station-
keeping of translunar halo orbits,” written by X. Bai and J. L. Junkins. The authors present a new
modified Chebyshev-Picard iteration method for the station-keeping of a halo orbit around
the L2 libration point of the Earth-Moon system. Compared to other competing methods,
the authors emphasize some advantages of the present technique: it does not require weight
turning and computation of the state transition matrix so that, computationally, it is clearly
very efficient. Also, the final results are given in an orthogonal polynomial form.
Then, the paper “A numerical study of optimal low-thrust limited power trajectories between
coplanar circular orbits in an inverse-square force field” is written by S. Fernandes et al.. Motivated
by the use of the low-thrust propulsion system in space missions, S. Fernandes et al. presented
a numerical analysis of optimal low-thrust limited power trajectories for simple transfers
between circular coplanar orbits in an inverse square force field. In this paper, two power-
limited variable ejection velocity systems are analyzed by considering the fuel consumption
as the performance criteria. The numerical results showed a good agreement with the
analytical results derived from the linear theory. Therefore, the linear theory can be used
as a good approximation for the transfer problem with small amplitudes.
Another paper of this sequence is “Optimal two-impulse trajectories with moderate flight
time for earth-moon missions,” written by S. S. Fernandes and C. M. P. Marinho. The paper
considers two-impulse trajectories with moderate flight time for Earth-Moon missions where
the goal is to optimize the fuel consumption, expressed by the total characteristic velocity.
The well-known patched-conic approximation is used as the dynamical model. After that,
two versions of the planar circular restricted three-body problem are also considered. Two
parameters are optimized in the two-body version: the initial phase angle of the space vehicle
and the first impulse. When considering the three-body dynamics, the parameters to be
optimized are four: initial phase angle of the space vehicle, the flight time, and the first and
the second impulses.
Another topic considered in this issue is related to the determination, simulation, and
propagation of trajectories of spacecraft under several different circumstances. Concerning
this topic is the paper “Analysis of filtering methods for satellite autonomous orbit determination
using celestial and geomagnetic measurement,” written by X. Ning et al.. Orbit determination
of a satellite is a complex process which evolves the precise estimation of the ephemeris
of the satellite. In the paper by Xiaolin et al., the performance of three filters methods,
extended Kalman filter EKF, unscented Kalman filter UKF, and unscented particle filter
UPF, is analyzed under different conditions. Their simulation results showed that the UPF
performs the best while the EKF gives the worst orbit determination OD accuracy under the
same conditions. The computation cost analysis demonstrates that the UPF-based OD system
provides the best orbit determination accuracy.
Another paper of this topic is “The orbital dynamics of synchronous satellites: irregular
motions in the 2 : 1 resonance,” written by J. C. Sampaio et al.. It analyses the dynamical behavior
of a synchronous satellite trapped in a 2 : 1 resonance with the Earth under the effects of the
zonal, J20 and J40 coefficients, and the tesseral harmonics, J22 and J42 coefficients. Through a
sample of Mathieu transformations, the order of the dynamical system was reduced. Three
critical angles associated to the 2 : 1 resonance were analyzed by numerically simulating the
reduced system. Their results showed that for several different values of the initial conditions
the system presented a chaotic motion, which was confirmed by calculating the Lyapunov
exponents.
Another one in this category is “Application of periapse maps for the design of trajectories
near the smaller primary in multi-body regimes,” written by K. C. Howell et al.. This paper has
the goal of incorporating multibody dynamics into the preliminary design of a mission. This
technique adds flexibility to the problem and, in some cases, generates impact in the cost
of the maneuver. Attention is given on the development and application of design tools to
help preliminary trajectory design considering a multibody environment. Using the model
given by the circular restricted three-body problem, a trajectory near the smaller primary is
studied. Periapse Poincaré maps are then used to predict both short- and long-term trajectory
behaviors. Several trajectories are computed using this approach.
The next paper is entitled “Comparison between two methods to calculate the transition
matrix of orbit motion,” written by A. P. M. Chiaradia et al.. The goal of this paper is to
compare and choose the most suited method to calculate the transition matrix used to
propagate the covariance matrix of the position and velocity of the state estimator, within
the procedure of the artificial orbit determination. Two methods are evaluated according to
accuracy, processing time, and handing complexity of the equations for both circular and
elliptical orbits. The first method is an approximation of the Keplerian motion, providing
an analytical solution which is calculated numerically by solving the Kepler equation. It
is also optimized for an elliptical orbit. The second one is a local numerical approximation
that includes the effect of the Earth oblateness; it has no singularity and no restriction about
the kind of orbit. By analyzing the results, it is possible to observe that for short periods of
time, and when more accuracy is necessary, it is recommended the use of the second method,
since the CP time does not overload excessively the computer during the orbit determination
procedure. However, for large intervals of time and when one expects more stability on the
calculations, the use of the first method is recommended.
Another paper of this category is “Semianalytic integration of high-altitude orbits under

lunisolar effects,” written by M. Lara et al.. In this paper, the long-term behavior of high-
altitude orbits with lunisolar perturbations and about a noncentral Earth including J2 and J3
potential terms is approached in a semianalytical way using canonical perturbation theory.
The authors take advantage of the different scales of time to do the averaging in the extended
phase space. In addition, up to second-order terms in the Earth’s oblateness coefficient,
the averaging has been computed in closed form of the eccentricity, and therefore, the
semianalytical integration can be applied to an orbit of arbitrary eccentricity.
The next set of papers is related to the attitude of the satellites, including maneuvering,
determination, and propagation of attitude. A paper related to that is “Analysis of attitude
determination methods using GPS carrier phase measurements,” written by L. Baroni and H. K.
Kuga. Global navigation satellite systems GNSS can be used to determine the orientation
of a platform. By using at least three GPS antennas, properly mounted on a platform, the
determination of the baseline vectors, formed between the antennas, is the prerequisite for
the attitude determination of this platform. L. Baroni and H. K. Kuga implemented and tested
two algorithms LSAT and LAMBDA in order to obtain an accurate attitude determination in
real time, by using data provided by the GPS receivers. As a result, they found that although
both algorithms had a similar performance, the processing time of the LSAT method was
larger. Although the inclusion of the quaternions increases the processing time, they are
necessary to avoid singularities.
Another one of this topic is “Numerical analysis of constrained, time-optimal satellite
reorientation,” written by R. G. Melton. The author studies on time-optimal satellite slewing
maneuvers with one satellite axis required to obey multiple path constraints. The paper
considers four cases in which this axis is either forced to follow a finite arc of the constraint
boundary or has initial and final directions that lie on the constraint boundary. It is shown
that the precession created by the control torques, moving the axis away from the constraint
boundary, provides faster slewing maneuvers. The paper also proposes a two-stage process
for the generation of optimal on-board solutions in less time.
Another one is the paper “Study of stability of rotational motion of spacecraft with canonical
variables,” written by W. R. Silva et al.. In this work, the main problem of the stability of
the rotational motion of a satellite is revisited. The most adequate set of variables Andoyer
canonical set and Kovalev-Savchenko theorem are adopted to analyze the stability of the
equilibrium points. Some important optimization on the techniques involved determination
of the equilibrium points, normalization and application of Kovalev-Savchenko theorem is
obtained. Examples considering real satellites with very large number of equilibrium points
are analyzed.
Then, the paper “Unscented Kalman filter applied to the spacecraft attitude estimation with
Euler angles” is written by R. V. Garcia et al.. In this paper, an algorithm is used to make real-
time estimation of the attitude of an artificial satellite. Some real data obtained by the CBERS-
2 satellite China-Brazil Earth Resources Satellite is used. The estimator used for attitude
determination is the unscented kalman filter, which is a new alternative to the extended
Kalman filter. This algorithm is capable of carrying out estimation of the states of nonlinear
systems, and it does not require the linearization of the functions present in the model. It uses
a transformation to generate a set of vectors that, after a nonlinear transformation, preserves
the same mean and covariance of the random variables obtained before the transformation.
A comparison between the unscented Kalman filter and the extended Kalman filter is
performed using real data.
Another of this category is the paper “Higher-order analytical attitude propagation of an

oblate rigid body under gravity-gradient torque,” written by J. F. San-Juan et al.. This paper is
related to the rotation of an oblate rigid body under gravity-gradient torque, a nonintegrable
problem that may be analyzed through the perturbation theory. The Andoyer variables, used
to describe the rotational motion, simplify the solution of the partial differential equations
that provide the generating function of the Lie transforms. It is shown that when a higher-
order perturbation theory is used to study the rotation of a uniaxial satellite under gravity-
gradient torque, this shows that known special configurations of the attitude dynamics at
which the satellite rotates, on average, as in a torque-free state, are only the result of an early
truncation of the secular frequencies of motion. The higher order of the analytical solution
can be compared to the long-term numerical integration of the equations of motion.
Other four papers are related to specific applications, including error analysis of
space missions, locations of platforms, and solutions of mathematical equations. One of
this type is the paper “Error modeling and analysis for InSAR spatial baseline determination of
satellite formation flying,” written by J. Tu et al.. Close formation flying satellites equipped
with synthetic aperture radar SAR antenna have become the focus of space technology.
The analysis of the errors introduced by spatial baseline measurement is important for the
performance of the SAR satellite system. J. Tu et al. carefully analyzed these errors and
classified them into two groups: the errors related to the baseline transformation and
those related to the GPS measurements. Through a set of simulations, they verified that
the errors related to the GPS measurements are the principal influence on the space baseline
determination, while the carrier phase noise of the GPS observation and fixing error of the
GPS receiver antenna are the main factors of the errors related to the GPS measurements.
Another one is “IMU fault detection based on χ2-CUSUM,” written by É. J. de Oliveira et
al.. Combining chi-square cumulative sum and media filter techniques, an efficient algorithm
is presented for the problem involving fault detection and isolation on inertial measurement
units. In particular, the success and feasibility of the method is clearly demonstrated in the
case of low-level step fault.
The paper “Assessment of the ionospheric and tropospheric effects in location errors of data
collection platforms in equatorial region during high and low solar activity periods,” written by
Á. A. da Silva et al., provides an assessment of ionospheric effects using IRI international
reference ionosphere and IONEX IONosphere map exchange models and tropospheric
delay compensation using climatic data provided by the National Climatic Data Center—
NOAA Satellite and Information Service. Two selected DCPs Data Collection System are
used in conjunction with the Brazilian Data Collection Satellite SCD2 during high- and
low-solar activity periods. Results show that the ionospheric effects on transmission delays
are significant in equatorial region and should be considered to reduce the DCP location
errors, mainly in high solar activity periods. It is also shown that the platform location errors
can be reduced when the ionospheric and the tropospheric effects are properly considered.
Another is the paper “Application of the Hori method in the theory of nonlinear oscillations,”
written by S. S. Fernandes. In this paper, the Hori method is applied to the theory of nonlinear
oscillations. Two algorithms to determine and generate functions and the new system of
differential equations are derived, based on a more general algorithm proposed by Sessin
1983. The solutions are not uniquely determined because the algorithms involve arbitrary
functions. Different choices of them can be made to simplify the new system of differential
equations and to define near-identity transformations. These algorithms are then applied to
determine second-order asymptotic solutions of two well-known equations in the theory of
nonlinear oscillations: van der Pol and Duffing equations.
Acknowledgments
The Guest Editors would like to thanks all the reviewers, the authors, and all the staff of this
journal involved in the preparation of this special issue for the opportunity to publish the
papers related to this topic.
Antonio F. Bertachini
A. Prado
Maria Cecilia Zanardi
Tadashi Yokoyama
Silvia Maria Giuliatti Winter
Joseph J. Masdemont
doi:10.1155/2012/168632
Research Article
A Numerical Study of Low-Thrust Limited Power
Trajectories between Coplanar Circular Orbits in
an Inverse-Square Force Field
Sandro da Silva Fernandes,1 Carlos Roberto Silveira Filho,2

and Wander Almodovar Golfetto3
1
Departamento de Matemática, Instituto Tecnológico de Aeronáutica, 12228-900 São José dos Campos, SP,
Brazil
2
EMBRAER S. A., Divisão de Ensaio em Voo, 12227-901 São José dos Campos, SP, Brazil
3
Subdepartamento Técnico do Departamento de Ciência e Tecnologia Aeroespacial, 12228-900 São José dos
Campos, SP, Brazil
Correspondence should be addressed to Sandro da Silva Fernandes, sandro@ita.br
Received 15 November 2011; Accepted 19 January 2012
Academic Editor: Silvia Maria Giuliatti Winter
Copyright q 2012 Sandro da Silva Fernandes et al. This is an open access article distributed under
the Creative Commons Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
A numerical study of optimal low-thrust limited power trajectories for simple transfer no
rendezvous between circular coplanar orbits in an inverse-square force field is performed by
two different classes of algorithms in optimization of trajectories. This study is carried out by
means of a direct method based on gradient techniques and by an indirect method based on the
second variation theory. The direct approach of the trajectory optimization problem combines the
main positive characteristics of two well-known direct methods in optimization of trajectories:
the steepest-descent first-order gradient method and a direct second variation second-order
gradient method. On the other hand, the indirect approach of the trajectory optimization
problem involves two different algorithms of the well-known neighboring extremals method.
Several radius ratios and transfer durations are considered, and the fuel consumption is taken
as the performance criterion. For small-amplitude transfers, the results are compared to the ones
provided by a linear analytical theory.
1. Introduction
The main purpose of this work is to present a numerical study of optimal low-thrust limited
power trajectories for simple transfers no rendezvous between circular coplanar orbits in an
inverse-square force field. This study has been motivated by the renewed interest in the use
of low-thrust propulsion systems in space missions verified in the last two decades due to the
development and the successes of space missions powered by ionic propulsion; for instance,
Deep Space One and SMART 1 missions. Several researchers have obtained numerical and
sometimes analytical solutions for a number of specific initial orbits and specific thrust
profiles 1–6. Averaging methods are also used in such researches 7–11.
Two idealized propulsion models have most frequently been used in the analysis of
optimal space trajectories 12: the limited power variable ejection velocity systems—LP
systems—are characterized by a constraint concerning with the power there exists an upper
constant limit for the power, and the constant ejection velocity-limited thrust systems—
CEV systems—are characterized by a constraint concerning with the magnitude of the thrust
acceleration which is bounded. In both cases, it is usually assumed that the thrust direction
is unconstrained. The utility of these idealized models is that the results obtained from them
provide good insight into more realistic problems.
In the study presented in this paper only LP systems are considered. The fuel con-
sumption is taken as the performance criterion and it is calculated for various radius ratios
ρ rf /r0 , where r0 is the radius of the initial circular orbit O0 , and rf is the radius of the
final circular orbit Of and for various time of flight tf − t0 . The optimization problem associ-
ated to the space transfer problem is formulated as a Mayer problem of optimal control with
Cartesian elements—components of position and velocity vectors—as state variables. Trans-
fers with small, moderate, and large-amplitudes are studied, and the numerical results are
compared to the results provided by a linear theory given in terms of orbital elements 12–
15.
Two different classes of algorithms are applied in determining the optimal trajectories.
They are computed through a direct approach of the trajectory optimization problem based
on gradient techniques, and through an indirect approach based on the solution of the two-
point boundary value problem obtained from the set of necessary conditions for optimality.
The direct approach involves a gradient-based algorithm which combines the main
positive characteristics of the steepest-descent first-order gradient method and of a direct
method based upon the second variation theory second-order gradient method. This
algorithm has two distinct phases: in the first one, it uses a simplified version of the steepest-
descent method developed for a Mayer problem of optimal control with free final state
and fixed terminal times, in order to get great improvements of the performance index in
the first iterates with satisfactory accuracy. In the second phase, the algorithm switches to
a direct method based upon the second variation theory developed for a Bolza problem
with fixed terminal times and constrained initial and final states, in order to improve the
convergence as the optimal solution is approached. This kind of algorithm for determining
optimal trajectories is well known in the literature 16, and the version used in this paper is
quite simple, since it uses a simplified version of the steepest-descent method, as mentioned
before, with terminal constraints added to the performance index by using a penalty function
method see Section 2.2. This procedure simplifies the algorithm, providing a solution with
satisfactory accuracy, and can avoid some of typical divergence troubles of the classical
steepest-descent method as discussed in McDermott and Fowler 17.
The indirect approach involves the solution of the two-point boundary value problem
through two different algorithms of the neighboring extremals method. The formulation of
the neighboring extremals method, as presented herein, is associated with a Bolza optimal
control problem with fixed initial and final times, fixed initial state and constrained final
state 18, 19. Basically, the method consists in iteratively determining the initial values of the
adjoint variables and the Lagrange multipliers associated to the final constraints. It involves
the linearization, about an extremal solution, of the nonlinear two-point boundary value
problem defined by the application of Pontryagin Maximum Principle 20 to the optimi-
zation problem. The linearized problem has been solved through the state transition matrix,
and through the generalized Riccati transformation 16, 27. The algorithms based on the
state transition matrix and the Riccati transformation are well known in the literature, and
the version used in this paper has a slight modification as described in da Silva Fernandes
and Golfetto 15.
A brief description of the versions of the algorithms used in this paper can be found in
da Silva Fernandes 21. Finally, note that the results presented herein complete and extend
the results previously obtained 15, 21, 22.
2. Optimal Low-Thrust Limited Power Trajectories

In this section, the optimization problem concerning with optimal low-thrust limited power
trajectories is formulated. Application of each one of the proposed algorithms is also present-
ed. For completeness, a very brief description of the linear theory is included.
2.1. Formulation of the Optimization Problem

Low-thrust limited power propulsion systems are characterized by low-thrust acceleration
level and a high specific impulse 12. The ratio between the maximum thrust acceleration
and the gravity acceleration on the ground, γmax /g0 , is between 10−4 and 10−2 . For such
system, the fuel consumption is described by the variable J defined as
t
1
J γ 2 dt, 2.1
2 t0
where γ is the magnitude of the thrust acceleration vector γ , used as control variable. The
consumption variable J is a monotonic decreasing function of the instantaneous mass m of
the space vehicle:

1 1
J Pmax − , 2.2
m m0
where Pmax is the maximum power, and m0 is the initial mass. The minimization of the final
value Jf is equivalent to the maximization of mf or the minimization of the fuel consumption.
The optimization problem concerning with simple transfers no rendezvous between
coplanar orbits is formulated as: at time t, the state of a space vehicle M is defined by the
radial distance r from the center of attraction, the radial and circumferential components of
the velocity, u and v, and the fuel consumption J. In the two-dimensional formulation, the
state equations are given by 23:
du v2 μ
− 2 R,
dt r r
dv uv
− S,
dt r
2.3
dr
u,
dt
dJ 1 2
R S2 ,
dt 2
where μ is the gravitational parameter, R and S are the radial and circumferential components
of the thrust acceleration vector, respectively. The optimization problem is stated as: it is
proposed to transfer a space vehicle M from the initial state at the time t0 0:
u0 0 v0 1 r0 1 J0 0, 2.4
to the final state at the prescribed final time tf :

μ
u tf 0 v tf r tf rf , 2.5
rf
such that Jf is a minimum, that is, the performance index is defined by:

IP J tf . 2.6
Equations 2.4 and 2.5 are given in canonical units, and they define the initial and final
circular orbits. utf , vtf , and rtf denote the state variables at the prescribed final time

tf , and 0, μ/rf , and rf are the prescribed values defining the final circular orbit. Similar
definition applies at the initial time t0 0 see 2.4. For LP system, it is assumed that there
are no constraints on the thrust acceleration vector 12.
In the formulation of the optimization problem described above, the variables are
written in canonical units, such that the gravitational parameter μ is equal to 1.
2.2. Application of the Gradient-Based Algorithm

As described in da Silva Fernandes 21, the first phase of the gradient-based algorithm
involves a simplified version of the steepest-descent method, which has been developed for
a Mayer problem of optimal control with free final state and fixed terminal times. So, the
optimal control problem defined by 2.3–2.6 must be transformed into a new optimization
problem with final state completely free. In order to do this, the penalty function method
24, 25 is applied. The new optimal control problem is then defined by 2.3 and 2.4, with
the new performance index obtained from 2.5 and 2.6:
2
2 1 2
IP J tf k1 u tf k2 v t f − √ k3 r tf − rf , 2.7
rf
where k1 , k2 , k3 1. The penalty function method involves the progressive increase of the
penalty constants; but, for simplicity, they are taken fixed in the gradient-based algorithm,
since the steepest-descent is used only to provide a convex nominal solution as starting
solution for the second order gradient method.
According to the algorithm of the simplified version of the steepest-descent method,
the adjoint variables λu , λv , λr , and λJ are introduced, and the Hamiltonian H is formed using
2.3 21, 26:

v2 μ uv 1
H λu − 2 R λv − S λr u λJ R2 S2 . 2.8
r r r 2
From the Hamiltonian H, one finds the adjoint equations:
dλu v
λv − λr ,
dt r
dλv v u
−2 λu λv ,
dt r r

dλr v2 μ uv
2
− 2 3 λu − 2 λv ,
dt r r r
dλJ
0,
dt
2.9
and, from the performance index defined by 2.7, we get the terminal conditions for the
adjoint equations:

λu tf −2k1 u tf ,

1
λv tf −2k2 v tf − √ ,
rf 2.10

λr tf −2k3 r tf − rf ,

λJ tf −1.
The algorithm also requires the partial derivatives of the Hamiltonian H with respect
to the control variables. These partial derivatives are given by:
∂H ∂H
λu RλJ , λv SλJ . 2.11
∂R ∂S
The second phase of the gradient-based algorithm involves the second order gradient
method, developed for a Bolza problem with fixed terminal times and constrained initial
and final states, which requires the computation of the first order derivatives of the vector
function ψ containing the terminal constraints and the scalar function Φ, corresponding to
the augmented performance index, and the computation of the second order derivatives of
the Hamiltonian H with respect to all arguments. The partial derivatives of the Hamiltonian
function are given in a matrix form by:

λJ 0
Hαα ,
0 λJ
⎡ ⎤
1 0
⎢ ⎥
⎢ 0 1⎥
⎢ ⎥
Hλα ⎢ ⎥,
⎢ 0 0⎥
⎣ ⎦
R S
Hxα 04 × 2 null matrix,

⎡ ⎤
v v2 μ
⎢ 0 2 r − r2 2 r3 0⎥
⎢ ⎥
⎢ v u uv ⎥
⎢− − 0⎥
Hλx ⎢
⎢ r r r2
⎥,
⎥
⎢ ⎥
⎢ 1 0 0 0⎥
⎣ ⎦
0 0 0 0
⎡ ⎤
λv v
⎢ 0 − λv 0⎥
⎢ r r2 ⎥
⎢ λ ⎥
⎢− v λu v u
−2 2 λu 2 λv 0⎥
⎢ 2 ⎥
Hxx ⎢
⎢ r r r r ⎥.
⎥
⎢ 2 ⎥
⎢ v v u v μ uv ⎥
⎢λv 2 −2 2 λu 2 λv 2 3 − 6 4 λu − 2 3 λv 0⎥
⎣ r r r r r r ⎦
0 0 0 0
2.12
α denotes the control vector αT R S, and it has been introduced to avoid confusion with
the state variable u, x is the state vector xT u v r J, and λ is the adjoint vector λT
λu λv λr λJ .
From 2.5 and 2.6, one defines the functions ψ and Φ:
⎡ ⎤
u tf
⎢ ⎥
⎢ 1 ⎥
ψ ⎢v tf − √ ⎥, 2.13
⎣ rf ⎦

r tf − rf

1
Φ J tf Λ1 u tf Λ2 v tf − √ Λ3 r tf − rf , 2.14
rf
where Λi , i 1, 2, 3 are Lagrangian multipliers associated to the final constraints defined by

2.13. The partial derivatives of ψ and Φ are then given by:
⎡ ⎤
1 0 0 0
⎢ ⎥
ψx ⎢ ⎥
⎣0 1 0 0⎦,
2.15
0 0 1 0
Φxx 0. 4 × 4 null matrix.
The results of the gradient-based algorithm to the optimization problem described

above are presented in Section 3.
2.3. Application of the Neighboring Extremals Algorithms

Let us to consider the Hamiltonian function defined by 2.8. Following the Pontryagin Max-
imum Principle 20, the control variables R and S must select from the admissible controls
such that the Hamiltonian function reaches its maximum along the optimal trajectory. Thus,
λu λv
R∗ − , S∗ − . 2.16
λJ λJ
The adjoint variables λu , λv , λr , and λJ must satisfy the adjoint differential equations
and the transversality conditions. Therefore, from 2.3–2.5 and 2.16 one finds the
following two-point boundary value problem for the transfer problem defined by 2.3–2.6:
du v2 μ λu
− 2− ,
dt r r λJ
dv uv λv
− − ,
dt r λJ
dr
u,
dt
dJ 1
2 λ2u λ2v ,
dt 2λJ
dλu v
λv − λr ,
dt r
dλv v u
−2 λu λv ,
dt r r

dλr v2 μ uv
−2 3 λu − λv ,
dt r2 r r2
dλJ
0,
dt
2.17
with the boundary conditions:
u0 0,
v0 1,
r0 1,
J0 0,
2.18
u tf 0,

μ
v tf ,
rf

r tf rf ,

λJ tf −1.
The neighboring extremals algorithms are based on the solution of a linearized two-
point boundary value problem that involves the derivatives of the right-hand side of 2.3
with respect to the state and adjoint variables 16, 21, 27. These equations can be put in the
following form:
dδx dδλ
Aδx Bδλ, Cδx − AT δλ, 2.19
dt dt
with δxt xn 1 t − xn t and δλt λn 1 t − λn t, where n denotes the iterate, and A, B,
and C are matrices given by:
⎡ 2v ⎤
0 −a 0
⎢ ⎥
⎢ v ru uv ⎥
⎢− − 0 ⎥
A⎢
⎢ r r r2 ⎥ ⎥,
⎢ 1 0 0 0⎥
⎣ ⎦
0 0 0 0
⎡ ⎤
1 λu
⎢− λJ 0 0 λ2 ⎥
⎢ J ⎥
⎢ ⎥
⎢ 1 λv ⎥
⎢ 0 − 0 ⎥
⎢ λJ λ2J ⎥
B⎢ ⎥,
⎢ 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ λ ⎥
⎣ u λv 0 − c ⎦
2 2 3
λJ λJ λJ
⎡ ⎤
λv vλv
⎢ 0 − 2 0⎥
⎢ r r ⎥
⎢ λv 2λu ⎥
⎢ − b 0⎥
C⎢
⎢ r r ⎥,
⎥
⎢ vλv ⎥
⎢− b d 0⎥
⎣ r2 ⎦
0 0 0 0
2.20
where
v2 μ
a 2
− 2 3,
r r
2v u
b λu − 2 λv ,
r2 r
2.21
c λ2u λ2v ,

2v2 6μ 2uv
d − 3 4 λu 3 λv .
r r r
The results of the neighboring extremals algorithms to the optimization problem de-
scribed above are presented in Section 3.
2.4. Linear Theory

For completeness, a very brief description of a first-order analytical solution for the problem
of optimal simple transfer no rendezvous between close quasicircular coplanar orbits in
an inverse-square force field is presented. This approximate solution, also referred as linear
theory, is expressed in nonsingular orbit elements, and it is valid for orbits with very small
eccentricities. According to Marec 12 or da Silva Fernandes and Golfetto 15, for transfers
between circular orbits, only Δα is imposed, and assuming that the initial and final positions
of the vehicle in orbit are symmetric with respect to x-axis of the inertial reference system,
that is,
f −
0 Δ
/2, the linear solution can be written as:

u na h sin
− k cos
,

v na 1 h cos
k sin
,
a
r ,
1 h cos
k sin

2.22
1 a5
Jt 4
−
0 λ 2
α 8 sin
− sin
0 λ α λh − 8 cos
− cos
0 λα λk
2 μ3
3
5 3

−
0 sin 2
− sin 2
0 λ2h − cos 2
− cos 2
0 λh λk
2 4 2
3
5

−
0 − sin 2
− sin 2
0 λ2k ,
2 4
with

5
a
αt α0 4 3

−
0 λα sin
− sin
0 λh − cos
− cos
0 λk , 2.23
μ

5
a 5 3
ht h0 4 sin
− sin
0 λα
−
0 sin 2
− sin 2
0 λh
μ3 2 4
2.24
3
− cos 2
− cos 2
0 λk ,
4

5
a 3
kt k0 3
−4 cos
− cos
0 λα − cos 2
− cos 2
0 λh
μ 4
2.25
3
5

−
0 − sin 2
− sin 2
0 λk ,
2 4
⎧ ⎫
3⎪
⎨ Δα 5Δ
3 sin Δ
⎪
⎬
1 μ
λα , 2.26
2 a5 ⎪
⎩ 10Δ
2 6Δ
sin Δ
− 64 sin2 Δ
/2 ⎪⎭
⎧ ⎫
⎪ ⎪
μ3 ⎨ 8Δα sin Δ
/2 ⎬
λh − ⎪, 2.27
a5 ⎪
⎩ 2
10Δ
6Δ
sin Δ
− 64 sin2 Δ
/2 ⎭
λk 0, 2.28
where α a/a, h e cos ω, k e sin ω, where a is the semimajor

axis, e is the eccentricity, ω
is the argument of the pericenter,

0 nt − t0 , and n μ/a3 is the mean motion. The
overbar denotes the reference orbit O about which the linearization is done.
The optimal thrust acceleration Γ∗ during the maneuver is expressed by:
1
Γ∗ λh sin
− λk cos
er 2 λα λh cos
λk sin
es , 2.29
na
where er and es are unit vectors extending along radial and circumferential directions in a
moving reference frame, respectively.
The linear theory is applicable only for orbits which are not separated by large radial
distance. If the reference orbit is chosen in the conventional way, that is, with the semimajor
axis as the radius of the initial orbit, the radial excursion to the final orbit will be maximized
14. A better reference orbit is defined with a semimajor axis given by an intermediate value
between the values of semimajor axes of the terminal orbits. In this study, a is taken as a
a0 af /2 in order to improve the accuracy in the calculations.
In the next section, the results of this linear theory are compared to the ones provided
by the proposed algorithms.
3. Results
The results of a numerical analysis for optimal low-thrust limited power simple transfers
no rendezvous between coplanar circular orbits in an inverse-square force field, obtained
through the analytical and numerical methods described in the preceding sections, are
presented for various radius ratios ρ rf /r0 and for various time of flight tf − t0 presented in
Tables 1–8. All results are presented in canonical units as described in Section 2. A preliminary
analysis of some interplanetary missions considering transfers from Earth to Venus, Mars,
asteroid belt, Jupiter, and Saturn, which correspond to ρ 0.727, 1.523, 2.500, 5.203, and
9.519, respectively, is presented. In this preliminary analysis of interplanetary missions, the
following assumptions are considered:
1 the orbits of the planets are circular;

2 the orbits of the planets lie in the plane of the ecliptic;
3 the flight of the space vehicle takes place in the plane of the ecliptic;
4 only the heliocentric phase is considered, that is, the attraction of planets on the
spacecraft is neglected.
Table 1: Consumption variable Jρ > 1 for transfers with small time of flight.
ρ tf − t0 Janal Jgrad JNeigh1 JNeigh2 drel 1 drel 2 drel 3

2.0 3.5856 × 10−4 3.5855 × 10−4 3.5854 × 10−4 3.5854 × 10−4 0.00 0.00 0.00
3.0 8.4459 × 10−5 8.4462 × 10−5 8.4456 × 10−5 8.4456 × 10−5 0.00 0.01 0.00
1.0250
4.0 3.1226 × 10−5 3.1233 × 10−5 3.1230 × 10−5 3.1230 × 10−5 0.01 0.01 0.00
5.0 1.7138 × 10−5 1.7147 × 10−5 1.7143 × 10−5 1.7143 × 10−5 0.03 0.02 0.00
2.0 1.4463 × 10−3 1.4463 × 10−3 1.4459 × 10−3 1.4459 × 10−3 0.03 0.03 0.00
3.0 3.4169 × 10−4 3.4166 × 10−4 3.4164 × 10−4 3.4164 × 10−4 0.01 0.01 0.00
1.0500
4.0 1.2533 × 10−4 1.2538 × 10−4 1.2537 × 10−4 1.2537 × 10−4 0.03 0.00 0.00
5.0 6.7541 × 10−5 6.7611 × 10−5 6.7598 × 10−5 6.7598 × 10−5 0.08 0.02 0.00
2.0 5.8778 × 10−3 5.8741 × 10−3 5.8716 × 10−3 5.8716 × 10−3 0.11 0.04 0.00
3.0 1.3977 × 10−3 1.3970 × 10−3 1.3969 × 10−3 1.3969 × 10−3 0.06 0.00 0.00
1.1000
4.0 5.0619 × 10−4 5.0666 × 10−4 5.0664 × 10−4 5.0664 × 10−4 0.09 0.00 0.00
5.0 2.6374 × 10−4 2.6453 × 10−4 2.6451 × 10−4 2.6451 × 10−4 0.29 0.01 0.00
2.0 2.4187 × 10−2 2.4097 × 10−2 2.4097 × 10−2 2.4097 × 10−2 0.37 0.00 0.00
3.0 5.8370 × 10−3 5.8200 × 10−3 5.8199 × 10−3 5.8199 × 10−3 0.29 0.00 0.00
1.2000
4.0 2.0813 × 10−3 2.0845 × 10−3 2.0844 × 10−3 2.0844 × 10−3 0.15 0.00 0.00
5.0 1.0260 × 10−3 1.0346 × 10−3 1.0345 × 10−3 1.0345 × 10−3 0.82 0.00 0.00
Table 2: Consumption variable Jρ < 1 for transfers with small time of flight.
ρ tf − t0 Jlinear Jgrad JNeigh1 JNeigh2 drel 1 drel 2 drel 3

2.0 2.0951 × 10−2 2.0842 × 10−2 2.0842 × 10−2 2.0842 × 10−2 0.52 0.00 0.00
3.0 4.9040 × 10−3 4.9173 × 10−3 4.9172 × 10−3 4.9172 × 10−3 0.27 0.00 0.00
0.8000
4.0 2.0703 × 10−3 2.1047 × 10−3 2.1046 × 10−3 2.1046 × 10−3 1.63 0.00 0.00
5.0 1.3838 × 10−3 1.4198 × 10−3 1.4197 × 10−3 1.4197 × 10−3 2.53 0.00 0.00
2.0 5.4740 × 10−3 5.4672 × 10−3 5.4671 × 10−3 5.4671 × 10−3 0.13 0.00 0.00
3.0 1.2771 × 10−3 1.2772 × 10−3 1.2771 × 10−3 1.2771 × 10−3 0.00 0.01 0.00
0.9000
4.0 5.0063 × 10−4 5.0198 × 10−4 5.0198 × 10−4 5.0198 × 10−4 0.27 0.00 0.00
5.0 3.0496 × 10−4 3.0653 × 10−4 3.0652 × 10−4 3.0652 × 10−4 0.51 0.00 0.00
2.0 1.3958 × 10−3 1.3955 × 10−3 1.3955 × 10−3 1.3955 × 10−3 0.02 0.00 0.00
3.0 3.2649 × 10−4 3.2649 × 10−4 3.2647 × 10−4 3.2647 × 10−4 0.01 0.01 0.00
0.9500
4.0 1.2451 × 10−4 1.2459 × 10−4 1.2458 × 10−4 1.2458 × 10−4 0.06 0.01 0.00
5.0 7.2585 × 10−5 7.2671 × 10−5 7.2667 × 10−5 7.2667 × 10−5 0.11 0.01 0.00
2.0 3.5225 × 10−4 3.5231 × 10−4 3.5223 × 10−4 3.5223 × 10−4 0.01 0.02 0.00
3.0 8.2555 × 10−5 8.2560 × 10−5 8.2554 × 10−5 8.2553 × 10−5 0.00 0.01 0.00
0.9750
4.0 3.1120 × 10−5 3.1126 × 10−5 3.1124 × 10−5 3.1124 × 10−5 0.01 0.01 0.00
5.0 1.7765 × 10−5 1.7772 × 10−5 1.7771 × 10−5 1.7771 × 10−5 0.03 0.00 0.00
Tables 1–4 show the values of the consumption variable J for small-amplitude trans-
fers computed through the different approaches and the absolute relative difference in per-
cent between the numerical and analytical results, according to the following definition:
% %
% Jneigh1 − Jlinear %
% %
drel 1 % % × 100%,
% Jneigh1 %
% %
% Jneigh1 − Jgrad %
% %
drel 2 % % × 100%, 3.1
% Jneigh1 %
% %
% Jneigh1 − Jneigh2 %
% %
drel 3 % % × 100%.
% Jneigh1 %
Table 3: Consumption variable Jρ > 1 for transfers with moderate time of flight.

−6 −6 −6 −6
20.0 3.7722 × 10 3.7733 × 10 3.7730 × 10 3.7714 × 10 0.02 0.01 0.04
30.0 2.5221 × 10−6 2.5226 × 10−6 2.5230 × 10−6 2.5209 × 10−6 0.04 0.02 0.08
1.0250
40.0 1.8855 × 10−6 1.8859 × 10−6 1.8860 × 10−6 1.8841 × 10−6 0.03 0.00 0.10
50.0 1.5066 × 10−6 1.5072 × 10−6 1.5070 × 10−6 1.5049 × 10−6 0.03 0.01 0.14
20.0 1.4520 × 10−5 1.4536 × 10−5 1.4533 × 10−5 1.4531 × 10−5 0.09 0.02 0.01
30.0 9.7411 × 10−6 9.7482 × 10−6 9.7480 × 10−6 9.7460 × 10−6 0.07 0.00 0.02
1.0500
40.0 7.2599 × 10−6 7.2659 × 10−6 7.2660 × 10−6 7.2636 × 10−6 0.08 0.00 0.03
50.0 5.8158 × 10−6 5.8199 × 10−6 5.8200 × 10−6 5.8182 × 10−6 0.07 0.00 0.03
20.0 5.4007 × 10−5 5.4168 × 10−5 5.4167 × 10−5 5.4165 × 10−5 0.30 0.00 0.00
30.0 3.6278 × 10−5 3.6390 × 10−5 3.6389 × 10−5 3.6387 × 10−5 0.30 0.00 0.01
1.1000
40.0 2.7003 × 10−5 2.7083 × 10−5 2.7078 × 10−5 2.7077 × 10−5 0.27 0.02 0.00
50.0 2.1653 × 10−5 2.1719 × 10−5 2.1718 × 10−5 2.1716 × 10−5 0.30 0.00 0.01
20.0 1.8980 × 10−4 1.9172 × 10−4 1.9154 × 10−4 1.9154 × 10−4 0.91 0.09 0.00
30.0 1.2543 × 10−4 1.2695 × 10−4 1.2693 × 10−4 1.2693 × 10−4 1.18 0.02 0.00
1.2000
40.0 9.4416 × 10−5 9.5396 × 10−5 9.5391 × 10−5 9.5390 × 10−5 1.02 0.01 0.00
50.0 7.5157 × 10−5 7.5976 × 10−5 7.5928 × 10−5 7.5927 × 10−5 1.02 0.06 0.00
Table 4: Consumption variable Jρ < 1 for transfers with moderate time of flight.

20.0 3.4529 × 10−4 3.5015 × 10−4 3.5011 × 10−4 3.5010 × 10−4 1.37 0.01 0.00
30.0 2.2973 × 10−4 2.3313 × 10−4 2.3316 × 10−4 2.3311 × 10−4 1.47 0.01 0.02
0.8000
40.0 1.7196 × 10−4 1.7467 × 10−4 1.7465 × 10−4 1.7465 × 10−4 1.54 0.01 0.00
50.0 1.3736 × 10−4 1.3978 × 10−4 1.3959 × 10−4 1.3959 × 10−4 1.59 0.13 0.00
20.0 7.3862 × 10−5 7.4146 × 10−5 7.4146 × 10−5 7.4144 × 10−5 0.38 0.00 0.00
30.0 4.8663 × 10−5 4.8851 × 10−5 4.8852 × 10−5 4.8850 × 10−5 0.39 0.00 0.00
0.9000
40.0 3.6467 × 10−5 3.6588 × 10−5 3.6589 × 10−5 3.6587 × 10−5 0.33 0.00 0.00
50.0 2.9218 × 10−5 2.9317 × 10−5 2.9316 × 10−5 2.9314 × 10−5 0.33 0.00 0.01
20.0 1.7023 × 10−5 1.7042 × 10−5 1.7040 × 10−5 1.7038 × 10−5 0.10 0.01 0.01
30.0 1.1240 × 10−5 1.1251 × 10−5 1.1249 × 10−5 1.1247 × 10−5 0.08 0.02 0.02
0.9500
40.0 8.4569 × 10−6 8.4642 × 10−6 8.4640 × 10−6 8.4620 × 10−6 0.08 0.00 0.02
50.0 6.7519 × 10−6 6.7581 × 10−6 6.7580 × 10−6 6.7561 × 10−6 0.09 0.00 0.03
20.0 4.0858 × 10−6 4.0869 × 10−6 4.0870 × 10−6 4.0847 × 10−6 0.03 0.00 0.06
30.0 2.7075 × 10−6 2.7081 × 10−6 2.7080 × 10−6 2.7059 × 10−6 0.02 0.00 0.08
0.9750
40.0 2.0361 × 10−6 2.0366 × 10−6 2.0360 × 10−6 2.0349 × 10−6 0.00 0.03 0.05
50.0 1.6230 × 10−6 1.6234 × 10−6 1.6230 × 10−6 1.6216 × 10−6 0.00 0.02 0.09
Table 5: Consumption variable J for Earth-Venus transfers.

−2 −2 −2 −2
2.0 3.7654 × 10 3.7299 × 10 3.7298 × 10 3.7298 × 10 0.95 0.00 0.00
3.0 8.9269 × 10−3 9.0261 × 10−3 9.0259 × 10−3 9.0259 × 10−3 1.10 0.00 0.00
4.0 4.0482 × 10−3 4.2133 × 10−3 4.2131 × 10−3 4.2131 × 10−3 3.91 0.00 0.00
0.7270 5.0 2.8941 × 10−3 3.0573 × 10−3 3.0572 × 10−3 3.0572 × 10−3 5.33 0.00 0.00
20.0 7.2355 × 10−4 7.4857 × 10−4 7.4856 × 10−4 7.4856 × 10−4 3.34 0.00 0.00
30.0 4.8236 × 10−4 4.9863 × 10−4 4.9862 × 10−4 4.9862 × 10−4 3.26 0.00 0.00
40.0 3.6177 × 10−4 3.7385 × 10−4 3.7384 × 10−4 3.7383 × 10−4 3.22 0.00 0.00
50.0 2.8942 × 10−4 2.9903 × 10−4 2.9901 × 10−4 2.9897 × 10−4 3.20 0.01 0.01
Table 6: Consumption variable J for Earth-Mars transfers.

2.0 1.7743 × 10−1 1.7434 × 10−1 1.7434 × 10−1 1.7434 × 10−1 1.77 0.00 0.00
3.0 4.4947 × 10−2 4.4067 × 10−2 4.4066 × 10−2 4.4066 × 10−2 1.99 0.00 0.00
4.0 1.6051 × 10−2 1.5889 × 10−2 1.5889 × 10−2 1.5889 × 10−2 1.02 0.00 0.00
1.5236 5.0 7.2498 × 10−3 7.3352 × 10−3 7.3351 × 10−3 7.3351 × 10−3 1.16 0.00 0.00
20.0 8.6591 × 10−4 9.3232 × 10−4 9.3151 × 10−4 9.3158 × 10−4 7.02 0.09 0.01
30.0 5.7537 × 10−4 6.1074 × 10−4 6.1071 × 10−4 6.1073 × 10−4 5.79 0.00 0.00
40.0 4.2991 × 10−4 4.5311 × 10−4 4.5296 × 10−4 4.5299 × 10−4 5.09 0.03 0.01
50.0 3.4273 × 10−4 3.6096 × 10−4 3.6093 × 10−4 3.6095 × 10−4 5.04 0.01 0.01
Table 7: Consumption variable J for interplanetary transfers with large-amplitude.
ρ tf − t0 JNeigh1 JNeigh2
20.0 3.7736 × 10−3 3.7733 × 10−3
30.0 2.3900 × 10−3 2.3900 × 10−3
2.500 40.0 1.7541 × 10−3 1.7541 × 10−3
50.0 1.3859 × 10−3 1.3859 × 10−3
60.0 1.1454 × 10−3 1.1453 × 10−3
20.0 1.3746 × 10−2 1.3746 × 10−2
30.0 7.5307 × 10−3 7.5309 × 10−3
5.203 40.0 5.0533 × 10−3 5.0533 × 10−3
50.0 3.7103 × 10−3 3.7100 × 10−3
60.0 2.9897 × 10−3 2.9896 × 10−3
60.0 5.9485 × 10−3 5.9392 × 10−3
70.0 4.6970 × 10−3 4.6980 × 10−3
9.519 80.0 3.9003 × 10−3 3.9009 × 10−3
90.0 3.3295 × 10−3 3.3285 × 10−3
100.0 2.8858 × 10−3 2.8857 × 10−3
The results provided by the neighboring extremals algorithm based on the state transition
matrix denoted by number 1 have been chosen as the exact solution for each maneuver, in
view of the accuracy obtained in fulfillment of the terminal constraints.
Similar results for interplanetary transfers are presented in Tables 5, 6 and 7. Results
for large-amplitude transfers with long time of flight are presented in Table 8. In both cases,
the transfers are only computed through the neighboring extremals algorithms.
From the results presented in Tables 1–8, major comments are as follows:
1 The linear theory provides a very good approximation for the fuel consumption
considering small-amplitude transfers with |ρ − 1| ≤ 0.100, that is, for transfers
between close circular coplanar orbits. For the most of the maneuvers, drel 1 < 0.5%;
2 For transfers with small time of flight tf − t0 2.0, 3.0, 4.0, 5.0 time units, Tables
1, 2, 5 and 6 show that the maximum absolute relative difference drel 1 occur for the
most of the transfers with tf − t0 5. This maximum value of drel 1 is about 2% for
ρ > 1 and 5.5% for ρ < 1;
3 For transfers with moderate time of flight tf − t0 20.0, 30.0, 40.0, 50.0 time units,
Tables 3, 4, 5, and 6 show that the maximum absolute relative difference drel 1 is
about 7% for ρ > 1, and 3.5% for ρ < 1;
4 In all cases described above, the maximum absolute relative differences drel 1 occur
for transfers with large radial excursion;
Table 8: Consumption variable J for large transfers.
ρ tf − t0 JNeigh1 JNeigh2
100.0 6.7827 × 10−4 6.7818 × 10−4
125.0 5.4241 × 10−4 5.4241 × 10−4
2.500 150.0 4.5154 × 10−4 4.5154 × 10−4
175.0 3.8651 × 10−4 3.8688 × 10−4
200.0 3.3812 × 10−4 3.3812 × 10−4
100.0 1.1975 × 10−3 1.1966 × 10−3
125.0 9.5018 × 10−4 9.4987 × 10−4
3.750 150.0 7.8787 × 10−4 7.8784 × 10−4
175.0 6.7324 × 10−4 6.7325 × 10−4
200.0 5.8793 × 10−4 5.8794 × 10−4
100.0 1.6105 × 10−3 1.6106 × 10−3
125.0 1.2678 × 10−3 1.2678 × 10−3
5.000 150.0 1.0441 × 10−3 1.0441 × 10−3
175.0 8.8948 × 10−4 8.8947 × 10−4
200.0 7.7584 × 10−4 7.7584 × 10−4
5 For transfers between close orbits with small time of flight, the fuel consumption
can be greatly reduced if the duration of the transfer increases: for instance, the fuel
consumption for transfers with time of flight tf − t0 2.0 time units is approx-
imately ten times the fuel consumption for transfers with time of flight tf − t0
4.0 time units, and, it is approximately hundred times the fuel consumption for
transfers with time of flight tf − t0 20.0 time units, considering any value of ρ;
6 For transfers with moderate time of flight, the fuel consumption decreases almost
linearly with the time of flight;
7 For transfers with moderate amplitude ρ 0.727 and ρ 1.523, Tables 5 and 6
show that 7.0% > drel 1 > 1.0%;
8 Tables 1–6 show that the results obtained through the numerical algorithms—
gradient and neighboring extremals—are very quite similar, regardless the
amplitude of maneuver and the time of flight;
9 Table 7 shows that an Earth-asteroid belt mission with tf − t0 20.0 time units
approximately, 3.2 years and an Earth-Jupiter mission with tf − t0 50.0 time
units approximately, 8.0 years spend almost the same quantity of fuel. This result
is closely related to the concept of transversals payoff curves in a field of extremals
introduced by Edelbaum 7 in the study of optimal limited-power transfers in
strong gravity field. Low-thrust limited power transfers with different amplitude
ρ and different time of flight tf − t0 can be performed with the same amount of
fuel as shown in Figures 7, 8, and 9. From Figure 7, one finds that an Earth-Venus
mission with tf − t0 30.0 time units approximately, 4.8 years and an Earth-Venus
mission with tf − t0 36.5 time units approximately, 5.8 years also spend almost
the same quantity of fuel, J 4.98 × 10−4 canonical units.
In order to follow the evolution of the optimal thrust acceleration vector during the
maneuver, it is convenient to plot the locus of its tip in the moving frame of reference. Figures
1, 2, 3, and 4 illustrate these plots for ρ 0.727, 0.950, 0.975, 1.025, 1.050, and 1.523, with
tf − t0 2.0, 3.0, 30.0 and 50.0. Note that the agreement between the numerical and analytical
Thrust acceleration-ρ = 0.975 Thrust acceleration-ρ = 1.025

0.01 0.03
Circumferential acceleration
0 0.02
−0.01 0.01
−0.02 0
−0.03 −0.01
−0.04 −0.02 0 0.02 0.04 −0.04 −0.02 0 0.02 0.04
Radial acceleration Radial acceleration

0.02 0.06
0 0.04
−0.02 0.02
−0.04 0
−0.06 −0.02
−0.06 −0.04 −0.02 0 0.02 0.04 0.06 −0.08 −0.04 0 0.04 0.08
Thrust acceleration-ρ = 0.727 0.6 Thrust acceleration-ρ = 1.523

0.1
0 0.4
−0.1
0.2
−0.2
0
−0.3
−0.4 −0.2
−0.3 −0.2 −0.1 0 0.1 0.2 0.3 −0.8 −0.4 0 0.4 0.8
Linear theory Linear theory

Gradient-based algorithm Gradient-based algorithm
Neighboring extremals Neighboring extremals
Figure 1: Thrust acceleration for tf − t0 2.0.


0.004 0.016
0 0.012
−0.004 0.008
−0.008 0.004
−0.012 0
−0.016 −0.004
−0.008 −0.004 0 0.004 0.008 −0.008 −0.004 0 0.004 0.008

0.01 0.03
0 Circumferential acceleration 0.02
−0.01 0.01
−0.02 0
−0.03 −0.01
−0.02 −0.01 0 0.01 0.02 −0.02 −0.01 0 0.01 0.02

0 0.3
−0.04 0.2
−0.08 0.1
−0.12 0
−0.16 −0.1
−0.08 −0.04 0 0.04 0.08 −0.3 −0.2 −0.1 0 0.1 0.2 0.3

Gradient-based algorithm Gradient-based algorithm


−0.0004 0.00046
−0.00041 0.00044
−0.00042 0.00042
−0.00043 0.0004
−0.00044 0.00038
−0.00045 0.00036
−2E−005
−1E−005
1E−005
2E−005
−1.2E−005
−8E−006
−4E−006
4E−006
8E−006
1.2E−005

−0.00085 0.00092
−0.000855 0.00088
−0.00086
0.00084
−0.000865
0.0008
−0.00087
0.00076
−0.000875
−0.00088 0.00072
−6E−005
−4E−005
−2E−005
2E−005
4E−005
6E−005
−8E−006
−4E−006
4E−006
8E−006

−0.0052 0.008
−0.0054 0.0075
0.007
−0.0056
0.0065
−0.0058
0.006
−0.006 0.0055
−0.0062 0.005
−0.0002
−0.0001
0.0001
0.0002
0.0003
−0.0008
−0.0004
0.0004
0.0008



−0.000248 0.000256
0.000252
−0.000252
0.000248
−0.000256
0.000244
−0.00026
0.00024
−0.000264 0.000236
−6E−006
−4E−006
−2E−006
2E−006
4E−006
6E−006
0
0
−4E−006
−2E−006
2E−006
4E−006

−0.00049 0.00052
−0.0005
0.0005
−0.00051
−0.00052 0.00048
−0.00053
0.00046
−0.00054
−0.00055 0.00044
0
0
−1E−005
−5E−006
5E−006
1E−005
−2E−005
−1E−005
1E−005
2E−005
−1.5E−005
1.5E−005

−0.0032 0.0042
−0.0033 0.004
−0.0034
0.0038
−0.0035
0.0036
−0.0036
0.0034
−0.0037
0.0032
0
−8E−005
−4E−005
4E−005
8E−005
0.00012
0
−0.0002
−0.0001
0.0001
0.0002


Thrust acceleration-ρ = 2.500, and tf −t0 = 100.0

0.0042 0.009
0.004 0.008
0.007
0.0038
0.006
0.0036
0.005
0.0034
0.004
0.0032 0.003
−0.0004 −0.0002 0 0.0002 0.0004 −0.002 −0.001 0 0.001 0.002
Figure 5: Thrust acceleration for large-amplitude transfers with tf − t0 100.0.

0.00195 0.0036
0.0019
0.0032
0.00185
0.0028
0.0018
0.00175 0.0024
0.0017 0.002
−8E−005
−4E−005
4E−005
8E−005
0
0.00012
−0.0008
−0.0004
0.0004
0.0008
Figure 6: Thrust acceleration for large-amplitude transfers with tf − t0 200.0.
results is better for small-amplitude transfers. For moderate- and large-amplitude transfers,
this difference increases with the duration of the maneuvers.
Similarly, Figures 5 and 6 show the evolution of the optimal thrust acceleration vector
for large-amplitude transfers with long time of flight. Note that the magnitude of the thrust
acceleration becomes smaller as the time of flight increases.
Figures 3, 4, and 5 show that for transfers with long time of flight the circumferential
thrust acceleration is the main component of the optimal thrust acceleration. As the time of
Consumption for Earth-Venus and

Earth-Mars transfers
0.001
0.0008
Consumption variable J 0.0006
0.0004
0.0002
20 30 40 50
Time of flight
Earth-Venus transfer
Earth-Mars transfer
Figure 7: Consumption for Earth-Venus and Earth-Mars transfers with moderate time of flight.
Consumption for interplanetary transfers

0.016
0.012
Consumption variable J
0.008
0.004
0
20 30 40 50 60
Time of flight
Earth-asteroid belt transfer

Earth-Jupiter transfer
Figure 8: Consumption for Earth-asteroid belt and Earth-Jupiter transfers.

Consumption for large transfers

0.002
0.0016
Consumption variable J
0.0012
0.0008
0.0004
0
100 120 140 160 180 200
Time of flight
ρ = 2.500
ρ = 3.750
ρ = 5.000
Figure 9: Consumption for large transfers.
flight increases the contribution of the radial component of the optimal thrust acceleration
decreases and the optimal thrust acceleration tends to the circumferential acceleration.
Figures 7 and 8 show the consumption variable for interplanetary transfers with
moderate flight of time, and Figure 9 shows the consumption variable for large-amplitude
transfers with long time of flight. Note that transfers with different amplitude ρ and different
time of flight tf − t0 can be performed with the same amount of fuel see Comment 9.
4. Conclusion
In this paper, a gradient-based algorithm and two different algorithms of the neighboring
extremals method are applied to the analysis of optimal low-thrust limited power transfers
between circular coplanar orbits in an inverse-square force field. The numerical results given
by these algorithms have been compared to the analytical results obtained by a linear theory.
The good agreement between these results shows that the linear theory provides a good
approximation for the solution of the transfer problem concerned with small-amplitudes,
that is, for transfers between close circular coplanar orbits. The linear theory can be used in
preliminary mission analysis involving such kind of transfers. The results presented in the
paper also show that the fuel consumption can be reduced if the time of flight of the transfer
increases. For transfers with long time of flight, the circumferential thrust acceleration
becomes the main component of the optimal thrust acceleration. A preliminary analysis of
some interplanetary missions is presented using the neighboring extremals algorithms.
Acknowledgment
This paper has been supported by CNPq under Contract 302949/2009-7.
References
1 V. Coverstone-Carroll and S. N. Williams, “Optimal low thrust trajectories using differential inclusion
concepts,” Journal of the Astronautical Sciences, vol. 42, no. 4, pp. 379–393, 1994.
2 J. A. Kechichian, “Optimal low-thrust rendezvous using equinoctial orbit elements,” Acta Astronau-
tica, vol. 38, no. 1, pp. 1–14, 1996.
3 J. A. Kechichian, “Reformulation of Edelbaum’s low-thrust transfer problem using optimal control
theory,” Journal of Guidance, Control, and Dynamics, vol. 20, no. 5, pp. 988–994, 1997.
4 A. A. Sukhanov and A. F. B. A. Prado, “Constant tangential low-thrust trajectories near on oblate
planet,” Journal of Guidance, Control, and Dynamics, vol. 24, no. 4, pp. 723–731, 2001.
5 M. Vasile, F. B. Zazzera, R. Jehn, and G. Janin, “Optimal interplanetary trajectories using a combi-
nation of low-thrust and gravity assist manoeuvres,” in Proceedings of the 51th IAF Congress, Rio de
Janeiro, Brasil, July 2000, IAF-00-A.5.07.
6 G. D. Racca, “New challenges to trajectory design by the use of electric propulsion and other new
means of wandering in the solar system,” Celestial Mechanics and Dynamical Astronomy, vol. 85, no. 1,
pp. 1–24, 2003.
7 T. N. Edelbaum, “Optimum power-limited orbit transfer in strong gravity fields,” The American
Institute of Aeronautics and Astronautics Journal, vol. 3, no. 5, pp. 921–925, 1965.
8 J. P. Marec and N. X. Vinh, “Optimal low-thrust, limited power transfers between arbitrary elliptical
orbits,” Acta Astronautica, vol. 4, no. 5-6, pp. 511–540, 1977.
9 C. M. Haissig, K. D. Mease, and N. X. Vinh, “Minimum-fuel, power-limited transfers between
coplanar elliptical orbits,” Acta Astronautica, vol. 29, no. 1, pp. 1–15, 1993.
10 S. Geffroy and R. Epenoy, “Optimal low-thrust transfers with constraints—generalization of
averaging techniques,” Acta Astronautica, vol. 41, no. 3, pp. 133–149, 1997.
11 S. da Silva Fernandes and F. das Chagas Carvalho, “A first-order analytical theory for optimal low-
thrust limited-power transfers between arbitrary elliptical coplanar orbits,” Mathematical Problems in
Engineering, Article ID 525930, 30 pages, 2008.
12 J. P. Marec, Optimal Space Trajectories, Elsevier, New York, NY, USA, 1979.
13 T. N. Edelbaum, “Optimum low-thrust rendezvous and station keeping,” The American Institute of
Aeronautics and Astronautics Journal, vol. 2, pp. 1196–1201, 1964.
14 F. W. Gobetz, “A linear theory of optimum low-thrust rendez-vous trajectories,” The Journal of the
Astronautical Sciences, vol. 12, no. 3, pp. 69–76, 1965.
15 S. da Silva Fernandes and W. A. Golfetto, “Numerical and analytical study of optimal low-thrust
limited-power transfers between close circular coplanar orbits,” Mathematical Problems in Engineering,
Article ID 59372, 23 pages, 2007.
16 A. E. Bryson, Jr. and Y. C. Ho, Applied Optimal Control, Hemisphere Publishing, New York, NY, USA,
1975.
17 M. McDermott, Jr. and W. T. Fowler, “Steepest-descent failure analysis,” Journal of Optimization Theory
and Applications, vol. 23, no. 2, pp. 229–243, 1977.
18 J. V. Breakwell, “The optimization of trajectories,” Journal of the Society for Industrial and Applied Math-
ematics, vol. 7, no. 2, pp. 215–247, 1959.
19 J. V. Breakwell, J. L. Speyer, and A. E. Bryson, “Optimization and control of nonlinear systems using
the second variation,” Journal of the Society for Industrial and Applied Mathematics A, vol. 1, no. 2, pp.
193–233, 1963.
20 L. S. Pontryagin, V. G. Boltyanskii, R. V. Gamkrelideze, and E. F. Mishchenko, The Mathematical Theory
of Optimal Control Processes, John Wiley & Sons, New York, NY, USA, 1962.
21 S. da Silva Fernandes, “Optimization of low-thrust limited-power transfers between coplanar circular
orbits in strong gravity field,” in Proceedings of the 20th International Congress of Mechanical Engineering
(COBEM ’09), Rio Grande do Sul, Brazil, November 2009.
22 S. da Silva Fernandes and W. A. Golfetto, “Numerical computation of optimal low-thrust limited-
power trajectories—transfers between coplanar circular orbits,” Journal of Brazilian Society of
Mechanical Sciences, vol. 27, no. 2, pp. 177–184, 2005.
23 W. E. Williamson and B. D. Tapley, “Riccati transformations for control optimization using the second
variation,” IEEE Transactions on Automatic Control, vol. AC-17, no. 3, pp. 319–327, 1972.
24 M. R. Hestenes, “Multiplier and gradient methods,” Journal of Optimization Theory and Applications,
vol. 4, no. 5, pp. 303–320, 1969.
25 R. J. O’Doherty and B. L. Pierson, “A numerical study of augmented penalty function algorithms for
terminally constrained optimal control problems,” Journal of Optimization Theory and Applications, vol.
14, no. 4, pp. 393–403, 1974.
26 D. E. Kirk, Optimal Control Theory—An Introduction, Dover, New York, NY, USA, 1998.
27 A. G. Longmuir and E. V. Bohn, “Second-variation methods in dynamic optimization,” Journal of
Optimization Theory and Applications, vol. 3, no. 3, pp. 164–173, 1969.
doi:10.1155/2012/239357
Research Article
Application of the Hori Method in the Theory of
Nonlinear Oscillations

Departamento de Matemática, Instituto Tecnológico de Aeronáutica, 12228-900,
São José dos Campos, SP, Brazil
Correspondence should be addressed to Sandro da Silva Fernandes, sandro@ita.br
Academic Editor: Antonio F. Bertachini A. Prado
Copyright q 2012 Sandro da Silva Fernandes. This is an open access article distributed under
Some remarks on the application of the Hori method in the theory of nonlinear oscillations are
presented. Two simplified algorithms for determining the generating function and the new system
of differential equations are derived from a general algorithm proposed by Sessin. The vector
functions which define the generating function and the new system of differential equations
are not uniquely determined, since the algorithms involve arbitrary functions of the constants
of integration of the general solution of the new undisturbed system. Different choices of these
arbitrary functions can be made in order to simplify the new system of differential equations
and define appropriate near-identity transformations. These simplified algorithms are applied
in determining second-order asymptotic solutions of two well-known equations in the theory of
nonlinear oscillations: van der Pol equation and Duffing equation.
1. Introduction
In da Silva Fernandes 1, the general algorithm proposed by Sessin 2 for determining the
generating function and the new system of differential equations of the Hori method for
noncanonical systems has been revised considering a new approach for the integration theory
which does not depend on the auxiliary parameter t∗ introduced by Hori 3, 4.
In this paper, this new approach is applied to the theory of nonlinear oscillations for
a second-order differential equation and two simplified versions of the general algorithm are
derived. The first algorithm is applied to systems of two first-order differential equations
corresponding to the second-order differential equation, and the second algorithm is applied
to the equations of variation of parameters associated with the original equation. According
m
to these simplified algorithms, the determination of the unknown functions Tj and
∗m
Zj , defined in the mth-order equation of the algorithm of the Hori method, is not
unique, since these algorithms involve at each order arbitrary functions of the constants
of integration of the general solution of the new undisturbed system. Different choices of
the arbitrary functions can be made in order to simplify the new system of differential
equations and define appropriate near-identity transformations. The problem of determining
second-order asymptotic solutions of two well-known equations in the theory of nonlinear
oscillations—van der Pol and Duffing equations—is taken as example of application of the
simplified algorithms. For van der Pol equation, two generating functions are determined:
one of these generating functions is the same function obtained by Hori 4, and, the
other function provides the well-known averaged equations of variation of parameters in
the theory of nonlinear oscillations. For Duffing equation, only one generating function
is determined, and the second-order asymptotic solution is the same solution obtained
through Krylov-Bogoliubov method 5, through the canonical version of Hori method 6
or through a different integration theory for the noncanonical version of Hori method 7.
For completeness, brief descriptions of the noncanonical version of the Hori method 4 and
the general algorithm proposed by Sessin 2 are presented in the next two sections.
2. Hori Method for Noncanonical Systems

The noncanonical version of the Hori method 4 can be briefly described as follows.
Consider the differential equations:
dzj
Zj z, ε, j 1, . . . , n, 2.1
dt
where Zj z, ε, j 1, . . . , n, are the elements of the vector function Zz, ε. It is assumed that
Zz, ε is expressed in power series of a small parameter ε:
&
Zz, ε Z0 z εm Zm z. 2.2
m1
The system of differential equations described by Z0 z is supposed to be solvable.

Let the transformation of variables z1 , . . . , zn → ζ1 , . . . , ζn be generated by the
vector function T ζ, ε. This transformation of variables is such that the new system:
dζj
Zj∗ ζ, ε, j 1, . . . , n, 2.3
dt
is easier to solve or captures essential features of the system. Zj∗ ζ, ε, j 1, . . . , n, are the
elements of the vector function Z∗ ζ, ε, also expressed in power series of ε:
&
Z∗ ζ, ε Z∗0 ζ εm Z∗m ζ. 2.4
m1
It is assumed that the vector function T ζ, ε, that defines a near-identity transformation, is
also expressed in powers series of ε:
&
T ζ, ε εm T m ζ. 2.5
m1
Following Hori 4, the transformation of variables z1 , . . . , zn → ζ1 , . . . , ζn gener-

ated by T ζ, ε is given by
& 1
zj ζj D k ζj e D T ζj , j 1, . . . , n. 2.6
k1
k! T
For an arbitrary function fz, the expansion formula is given by
& 1
fz fζ DTk fζ eDT fζ. 2.7
k1
k!
The operator DT is defined by
&
n
∂f
DT fζ Tj ,
j1
∂ζj
⎛ ⎞ 2.8
∂f &
n
DTn fζ DTn−1 ⎝ Tj ⎠.
j1
∂ζj
According to the algorithm of the perturbation method proposed by Hori 4, the
vector functions Z and T are obtained, at each order in the small parameter ε, from the
following equations:
0 ∗0
order 0: Zj Zj , 2.9
+ ,
1 ∗1
order 1: Z0 , T 1 Zj Zj , 2.10
j
+ , 1+ ,
2 ∗2
order 2: Z0 , T 2 Z1 Z∗1 , T 1 Zj Zj ,
j 2 j
2.11
..
.
j 1, . . . , n, where j stands for the generalized Poisson brackets

&
n ∂Zj ∂Tj
Z, T j Tk − Zk . 2.12
k1
∂ζk ∂ζk
Z∗0 , Zm , Z∗m , and T m are written in terms of the new variables ζ1 , . . . , ζn .
The mth-order equation of the algorithm can be put in the general form:
+ ,
m ∗m
Z∗0 , T m Ψj Zj , j 1, . . . , n, 2.13
j
m
where the functions Ψj are obtained from the preceding orders.
3. The General Algorithm

∗m m
The determination of the functions Zj and Tj from 2.13 is based on the following
proposition presented in da Silva Fernandes 1.
Proposition 3.1. Let F be a n × 1 vector function of the variables ζ1 , . . . , ζn , which satisfy the system
of differential equations:
dζj ∗0
Zj ζ R∗j ζ; ε, j 1, . . . , n, 3.1
dt
where Z∗0 describes an integrable system of differential equations:
dζj ∗0
Zj ζ, j 1, . . . , n, 3.2
dt
a general solution of which is given by
ζj ζj c1 , . . . , cn , t, j 1, . . . , n, 3.3
being c1 , . . . , cn arbitrary constants of integration; then
∗0
+ , ∂Fj & n ∂Z
j
∗0
F, Z − Fk , j 1, . . . , n. 3.4
j ∂t k1
∂ζk
A corollary of this proposition can be stated.
Corollary 3.2. Consider the same conditions of Proposition 3.1 with the general solution of 3.2
given by
ζj ζj c1 , . . . , cn−1 , M, j 1, . . . , n, 3.5
being c1 , . . . , cn−1 arbitrary constants of integration and M t τ, where τ is an additive constant;

then
∗0
+ , ∂Fj &n ∂Z
j
∗0
F, Z − Fk , j 1, . . . , n. 3.6
j ∂M k1 ∂ζk
Now, consider 2.13. According to Proposition 3.1, this equation can be put in the
form:
m ∗0
∂Tj &
n ∂Z
j m m ∗m
− Tk Ψj − Zj , j 1, . . . , n, 3.7
∂t k1
∂ζk
m
with Ψj written in terms of the general solution 3.3 of the undisturbed system 3.2,
∗m m
involving n arbitrary constants of integration—c1 , . . . , cn . Zj and Tj are unknown
functions.
Equation 3.7 is very similar to the one presented by Hori 4, which is written in
terms of an auxiliary parameter t∗ through an ordinary differential equation, that is,
m ∗0
dTj &
n ∂Z
j m m ∗m
− Tk Ψj − Zj , j 1, . . . , n. 3.8
dt∗ k1
∂ζk
∗m m
To determine Zj and Tj , j 1, . . . , n, Hori 4 extends the averaging principle
∗m m
applied in the canonical version: Zj
are determined so that the Tj are free from secular
or mixed secular terms. However, this procedure is not sufficient to determine Z∗ such that
the new system of differential equations 2.3 becomes more tractable, and, a tractability
condition is imposed
+ ,
Z0 , Z∗ 0, j 1, . . . , n. 3.9
j
This condition is analogous to the condition {F 0 , F ∗ } 0 in the canonical case, which
provides the first integral F0 ξ, η const, where ξ, η denotes the new set of canonical
variables, and, F 0 and F ∗ are the undisturbed Hamiltonian and the new Hamiltonian,
respectively, and {} stands for Poisson brackets 3, 4.
∗m m
In the next paragraphs, the general algorithm for determining Zj and Tj , j
1, . . . , n, proposed by Sessin 2 and revised in da Silva Fernandes 1 is briefly presented.
Introducing the n × 1 matrices:

m m ∗m
T m Tj , Ψm Ψj , Z∗m Zj , j 1, . . . , n, 3.10
and the n × n Jacobian matrix

⎛ ∗0
⎞
∂Zj
Jt ⎝ ⎠, j, k 1, . . . , n, 3.11
∂ζk
the system of partial differential equations 3.7 can be put in the following matrix form:
∂T m
− JtT m Ψm − Z∗m . 3.12
∂t
The vector functions Z∗m and T m are determined from the following equations:

∗m ∂
Z Δζ Δζ Δζ D Δζ Δ−1
−1 m
ζ Ψ m
dt , 3.13
∂t s

m
T Δζ D Δζ Δ−1
m
ζ Ψ m
dt , 3.14
p
where Δζ ∂ζj c1 , . . . , cn , t/∂ck is the Jacobian matrix associated to the general solution
3.3 of the undisturbed system 3.2, s denotes the secular or mixed secular terms, and p
m
denotes the remaining part. Dm is the n × 1 vector, Dm Dj , which depends only on
the arbitrary constants of integration c1 , . . . , cn of the general solution 3.3. The choice of Dm
is arbitrary. Recall that in the integration process, the arbitrary constants of integration of the
general solution 3.3 are taken as parameters.
Equations 3.13 and 3.14 assure that T m is free from secular or mixed secular terms.
Moreover, these equations provide the tractability condition 3.9 as it will be shown in the
case of nonlinear oscillation problems presented in the next section.
Finally, it should be noted that Dm can be chosen at each order to simplify the
generating function T and the new system of differential equations 2.3. This aspect is
discussed thoroughly in the examples of Section 5.
4. Simplified Algorithms in the Theory of Nonlinear Oscillations

In this section two simplified algorithms will be derived from the general algorithm in the
case of nonlinear oscillations described by a second-order differential equation of the general
form:
ẍ x εfx, ẋ. 4.1
The first algorithm is applied to the system of first-order differential equations with
x and ẋ as elements of the vector z, and, the second algorithm is applied to the system of
equations of variation of parameters associated to the differential equation with c and θ as
elements of the vector z; c and θ are defined in 4.26.
4.1. Simplified Algorithm I

For completeness, we present now the first simplified algorithm 1. Additional remarks are
included at the end of section.
Introducing the variables z1 x and z2 ẋ, 4.1 can be put in the form:
ż1 z2 , ż2 −z1 εfz1 , z2 . 4.2

According to the notation introduced in 2.1 and 2.2:

z2 0
Z0 , Z1 . 4.3
−z1 fz1 , z2
Following the algorithm of the Hori method for noncanonical systems, one finds from
zero-th-order equation, 2.9, that

∗0 ζ2
Z . 4.4
−ζ1
Applying Proposition 3.1, it follows that the undisturbed system 3.2 is given by
ζ̇1 ζ2 , ζ̇2 −ζ1 , 4.5
general solution of which can be written in terms of the exponential matrix as 2:

ζ1 c
eEt 1 , 4.6
ζ2 c2
where

cos t sin t
e Et
, 4.7
− sin t cos t
and E is the symplectic matrix:

0 1
E , 4.8
−1 0
and ci , i 1, 2, are constants of integration. The Jacobian matrix Δζ associated with the
solution 4.6 is then given by
Δζ eEt , 4.9
with inverse Δ−1

ζ
e−Et ΔTζ , since Δζ is an orthogonal matrix.
m
In view of 4.6, the functions Ψj defined at each order of the algorithm see 2.13
or 3.7 are expressed by Fourier series with multiples of t as arguments such that the vector
m
function ΨI can be written as
⎡ m m ⎤
&∞ ak,I cos kt bk,I sin kt
m ⎢ ⎥
ΨI ⎣ m m
⎦, 4.10
k0 ck,I cos kt dk,I sin kt
m m m m
where the coefficients ak,I , bk,I , ck,I , and dk,I are functions of the constants ci , i 1, 2. The
Fourier series can also be put in matrix form:
∞
&
m m m
ΨI eEkt Ak,I e−Ekt Bk,I , 4.11
k0
with
⎡ m m ⎤ ⎡ m m ⎤
ak,I − dk,I ak,I dk,I
m 1⎢ ⎥ m 1⎢ ⎥
Ak,I ⎣ m ⎦, Bk,I ⎣ m ⎦. 4.12
2 b cm 2 c − bm
k,I k,I k,I k,I
The subscript I is introduced to denote the first simplified algorithm.

Substituting 4.9 and 4.11 into 3.13, one finds
∗m ∂ −Et + Et m ,

m
ZI eEt e te A1,I periodic terms eEt A1,I . 4.13
∂t s
On the other hand,

- .
m m
e−Et ΨI A1,I , 4.14
where
stands for the mean value of the function.
Therefore, from 4.9, 4.13, and 4.14, it follows that
- . - .
∗m m m
ZI eEt e−Et ΨI Δζ Δ−1
ζ ΨI . 4.15
The second equation of the general algorithm, 3.14 can be simplified as described
bellow.
From 3.14, 4.14, and 4.15, one finds
- . / 0
m m m m
te Et
A1,I Δζ Δ−1
ζ ΨI dt Δζ DI Δζ Δ−1
ζ ΨI dt . 4.16
s
On the other hand,

/ 0
m m m m m
Δζ DI Δζ Δ−1
ζ ΨI dt Δζ DI Δζ Δ−1
ζ ΨI dt teEt A1,I . 4.17
p
Thus, introducing 4.16 and 4.17 into 3.14, one finds

/ 0 1 - .2
m m m m −1 m −1 m
TI Δζ DI Δζ Δ−1
ζ Ψ I dt Δζ D I Δζ Δζ Ψ I − Δζ Ψ I dt. 4.18
p
Equations 4.15 and 4.18 define the first simplified form of the general algorithm
applicable to the nonlinear oscillations problems described by 4.1 with x and ẋ as elements
of vector z.
Finally, we note that 4.15 satisfies the tractability condition 3.9 up to order m. In
m
order to show this equivalence, one proceeds as follows. Since, from 4.14, Δ−1 ζ
ΨI
does
not depend explicitly on the time t, it follows that
∂Z∗m ∂Δζ - −1 m . -

m
.
Δζ ΨI JΔζ Δ−1
ζ ΨI JZ∗m . 4.19
∂t ∂t
∗0
Using Proposition 3.1 and taking into account that J ∂Zj /∂ζk , this equation can be put in
the following form:
∗m ∗0
+ , ∂Zj &n ∂Z
∗m 0 j ∗m
Z ,Z − Zk 0, j 1, 2, 4.20
j ∂t k1
∂ζ k
which is the tractability condition 3.9 up to order m.
Remark 4.1. It should be noted that 4.15 and 4.18 for determining the vector functions
∗m m
ZI and TI , respectively, are invariant with respect to the form of the general solution
of the undisturbed system described by Z∗0 . This means that if the general solution of the
undisturbed system is written in terms of a second set of constants of integration, for instance,
if this solution is given by
ζ1 c cost θ,
4.21
ζ2 −c sint θ,
∗m m
where c and θ denote new constants of integration, then ZI and TI are determined
through 4.15 and 4.18, with the Jacobian matrix Δζ given by

cost θ −c sint θ
Δζ . 4.22
− sint θ −c cost θ
This result can be proved as follows. The two sets of constants of integration c1 , c2 and c, θ
are related through the following transformation:
c2 c12 c22 ,
c2 4.23
tan θ − .
c1
In view of this transformation, the Jacobian matrix Δ1ζ can be written in terms of the Jacobian
matrix Δ2ζ as
Δ1ζ Δ2ζ ΔC , 4.24

where the superscripts 1 and 2 are introduced to denote the form of the general solution of
the undisturbed system described by Z∗0 with respect to the set of constants of integration
c1 , c2 and c, θ, respectively. ΔC is the Jacobian matrix of the transformation. Since ΔC does
not depend on the time t, it follows from 4.15 and 4.18 that
3 −1 34 −1 m 4
∗m m
ZI Δ1ζ Δ1ζ Δ2ζ ΔC Δ−1
Ψ1I C Δζ
2
Ψ2I
3 −1 4 3 −1 4
−1 2 m 2 m
Δζ ΔC ΔC
2
Δζ
2
ΨI Δζ Δζ
2 2
ΨI ,
3 −1 m 4
m m −1 m
TI Δ1ζ DI1 Δ1ζ Δ1ζ Ψ1I − Δ1ζ Ψ1I dt
−1 m 3 −1 m 4 4.25
m
Δ2ζ ΔC DI1 Δ2ζ ΔC Δ−1 C Δ 2
ζ Ψ 2
I − Δ−1
C Δζ
2
Ψ2I dt
3 −1 4
m −1
2 m 2 m
Δ2ζ ΔC DI1 Δ2ζ ΔC Δ−1 C Δ 2
ζ Ψ I − Δ 2
ζ Ψ I dt
3 −1 m 4
m −1 m
Δ2ζ DI2 Δ2ζ Δ2ζ Ψ2I − Δ2ζ Ψ2I dt.
Finally, we note that the general solution given by 4.21 is more suitable in practical
applications than the general solution given by 4.6, that, in turn, is more suitable for
theoretical purposes.
4.2. Simplified Algorithm II

In this section, a second simplified algorithm is derived from the general one. Introducing the
transformation of variables x, ẋ → c , θ defined by the following equations

x c cos t θ ,
4.26
ẋ −c sin t θ ,
equation 4.1 is transformed into
dc
−εf c cos t θ , −c sin t θ sin t θ ,
dt 4.27
dθ 1
−ε f c cos t θ , −c sin t θ cos t θ .
dt c
These differential equations are the well-known variation of parameters equations associated
to the second-order differential equation 4.1. Equation 4.27 define a nonautonomous
system of differential equations.
The sets c, θ and c , θ , defined, respectively in 4.21 and 4.26, have different
meanings in the theory: in 4.21, c and θ are constants of integration of the general solution
of the new undisturbed system described by Z∗0 ζ1 , ζ2 ; in 4.26, c and θ are new
variables which represent the constants of integration of the general solution of the original
undisturbed system described by Z0 z1 , z2 in the variation of parameter method. These
sets, c, θ and c , θ , are connected through a near identity transformation.
Remark 4.2. It should be noted that a second transformation of variables involving a fast
phase, x, ẋ → c , φ , can also be defined. This second transformation is given by
x c cos φ ,
4.28
ẋ −c’sin φ .
In this case, 4.1 is transformed into
dc
−εf c’cos φ , −c’sin φ sin φ ,
dt 4.29
dφ 1
1 − ε f c’cos φ , −c’sin φ cos φ .
dt c
These equations define an autonomous system of differential equations. In what follows, the
first set of variation of parameters equations, 4.27, will be considered.
Now, introducing the variables z1 c and z2 θ , one gets from 4.27 that

0
Z0 ,
0
⎡ ⎤
−fz1 cost z2 , −z1 sint z2 sint z2 4.30
⎢ ⎥
Z 1
⎢ ⎥
⎣− 1 fz1 cost z2 , −z1 sint z2 cost z2 ⎦.
z1
Applying Proposition 3.1, it follows that the undisturbed system 3.2 is given by
ζ̇1 0, ζ̇2 0, 4.31
and its general solution is very simple,
ζi c i , i 1, 2, 4.32
where ci , i 1, 2, are constants of integration. The Jacobian matrix Δζ associated with this
general solution is also very simple, and it is given by
Δζ I, 4.33
where I is the identity matrix.

Substituting 4.33 into 3.13 and 3.14, it follows that

∗m ∂ m m
ZII DII ΨII dt ,
∂t
s 4.34
m m m
TII DII ΨII dt .
p
The subscript II is introduced to denote the second simplified algorithm.

Equation 4.34 can be put in a more suitable form as follows. In view of 4.30, the
m
functions Ψj defined at each order of the algorithm see 2.13 or 3.7 are expressed by
m
Fourier series with multiples of t z2 as arguments such that the vector function ΨII can be
written as
⎡ m m ⎤
&∞ ak,II cos kt c2 bk,II sin kt c2
m ⎢ ⎥
ΨII ⎣ m m
⎦, 4.35
k0 ck,II cos kt c2 dk,II sin kt c2
m m m m

where the coefficients ak,II , bk,II , ck,II , and dk,II are functions of the constant c1 . The vector
m
function ΨII can also be put in matrix form:
∞
&
m m m
ΨII eEkt c2 Ak,II e−Ekt c2 Bk,II , 4.36
k0
with
⎡ m m ⎤ ⎡ m m ⎤

a − dk,II a dk,II
m 1 ⎢ k,II ⎥ m 1 ⎢ k,II ⎥
Ak,II ⎣ m m
⎦, Bk,II ⎣ m m
⎦. 4.37
2 b c 2 c − b
k,II k,II k,II k,II
m m
Note that ΨII is very similar to ΨI , defined by 4.11. They represent different forms of
Fourier series of Ψm , but they are not the same, since they involve different sets of arbitrary
constants of integration.
Thus, it follows from 4.36 that

m m m m m
DII ΨII dt DII A0,II B0,II t periodic terms, 4.38
with the periodic terms given by
∞
&
m m
Ek−1 eEkt c2 Ak,II −Ek−1 e−Ekt c2 Bk,II . 4.39
k1
Therefore,

m m m m m
DII ΨII dt DII A0,II B0,II t,
s
∞
& 4.40
m m m m
DII ΨII dt Ek−1 eEkt c2 Ak,II −Ek−1 e−Ekt c2 Bk,II .
p k1
Substituting 4.40 into 4.34, one finds

- .
∗m m m m
ZII A0,II B0,II ΨII , 4.41
- .
m m m m
TII DII ΨII − ΨII dt. 4.42
m
Note that DII depends only on c1 ζ1 .
It should be noted that 4.41 and 4.42 can be straightforwardly obtained from 3.12
m
by applying the averaging principle if DII is assumed to be zero, since in this second
approach:
⎛ ∗0
⎞
∂Zj
Jt ⎝ ⎠ O, 4.43
∂ζk
where O denotes the null matrix. Thus, the general algorithm defined by 3.13 and 3.14 is
equivalent to the averaging principle usually applied in the theory of nonlinear oscillations
5, 7.
Remark 4.3. Equations 4.41 and 4.42 are also obtained, if the second set of variation of
parameters equations is considered see Remark 4.2. In this case, the undisturbed system
3.2 is given by
ζ̇1 0, ζ̇2 1, 4.44
with general solution defined by
ζ1 c 1 , ζ1 t c 2 , 4.45
and Jacobian matrix Δζ I.

m
Finally, we note that 4.41 is the tractability condition 3.9 up to order m. Since ΨII
does not depend explicitly on the time t, it follows that
∗m
∂Zj + ,
Z∗m , Z0 0, j 1, 2, 4.46
∂t j
which is the tractability condition 3.9 up to order m.
5. Application to Nonlinear Oscillations Problems

In order to illustrate the application of the simplified algorithms, two examples are presented.
The noncanonical version of the Hori method will be applied in determining second-order
asymptotic solutions for van der Pol and Duffing equations. For the van der Pol equation,
two different choices of the vector Dm will be made, and two generating functions T m
will be determined, one of these generating functions is the same function obtained by Hori
4 through a different approach, and, the other function gives the well-known averaged
variation of parameters equations in the theory of nonlinear oscillations obtained through
Krylov-Bogoliubov method 5. It should be noted that the solution presented by Hori defines
a new system of differential equations with a different frequency for the phase in comparison
with the solution obtained by Ahmed and Tapley 7 and by Nayfeh 5, using different
perturbation methods. For the Duffing equation, only one generating function is determined,
and the second simplified algorithm gives the same generating function obtained through
Krylov-Bogoliubov method.
The section is organized in two subsections: in the first subsection, the asymptotic solu-
tions are determined through the first simplified algorithm, and, in the second subsection,
they are determined through the second simplified algorithm.
5.1. Determination of Asymptotic Solutions through Simplified Algorithm I

5.1.1. Van der Pol Equation
Consider the well-known van der Pol equation:

ẍ ε x2 − 1 ẋ x 0. 5.1
Introducing the variables z1 x and z2 ẋ, this equation can be written in the form:
dz1 dz1
z2 , −z1 − ε z21 − 1 z2 . 5.2
dt dt
Thus

z2
Z0 , 5.3
−z1

0
Z1 . 5.4
− z21 − 1 z2
As described in preceding paragraphs, two different choices of Dm will be made, and
two generating functions T m will be determined. Firstly, we present the solution obtained
by Hori 4.
(1) First Asymptotic Solution: Hori’s [4] Solution

Following the simplified algorithm I defined by 4.15 and 4.18, the first-order terms Z∗1
and T 1 are calculated as follows.
Introducing the general solution given by 4.21 of the undisturbed system described
by Z∗0 ζ1 , ζ2 into 5.4, with ζ replacing z, one gets
⎡ ⎤
0
Z1 ⎣ 1 1 ⎦. 5.5
−c c3 sint θ c3 sin 3t θ
4 4
Computing Δ−1
ζ
Z1 ,
⎡1 1

1 1 ⎤
c 1 − c2 − c cos 2t θ c3 cos 4t θ
⎢2 4 2 8 ⎥
Δ−1
ζ Z
1
⎢
⎣ ⎥,
⎦ 5.6
1 1 2 1 2
1 − c sin 2t θ − c sin 4t θ
2 2 8
and taking its secular part, one finds
⎡ ⎤
- . 1 1 2
c 1 − c
Δ−1
ζ Z
1
⎣2 4 ⎦. 5.7
0
From 4.15 and 4.22, it follows that Z∗1 is given by
⎡1 1
⎤
c 1 − c2 cost θ
⎢2 4 ⎥
Z∗1 ⎢
⎣ 1 ⎥.
⎦ 5.8
1 2
− c 1 − c sint θ
2 4
In view of 4.21, Z∗1 can be written explicitly in terms of the new variables ζ1 and ζ2 as
follows:
⎡1 1 2

⎤
ζ 1− ζ1 ζ22
⎢2 1 4 ⎥
Z∗1 ⎢
⎣ ⎥ . 5.9
1 1 2 ⎦
ζ2 1 − ζ1 ζ22
2 4
5
To determine T 1 , the indefinite integral Δ−1
ζ
Z1 − Δ−1
ζ
Z1
dt is calculated:
⎡ 1 1 3 ⎤
+ - ., − c sin 2t θ c sin 4t θ
⎢ 4 32 ⎥
Δ−1
ζ Z
1
− Δ−1
ζ Z
1
dt ⎢
⎣ 1 1
⎥.
⎦ 5.10
1 2
− − c cos 2t θ c cos 4t θ
2
4 8 32
Thus, from 4.18, it follows that T 1 is given by
⎡ 1 1

1 ⎤
− c 1 − c2 sint θ − c3 sin 3t θ
⎢ 4 4 32 ⎥
T 1 ⎢
⎣1 ⎥ Δζ D1 .
⎦ 5.11
1 2 3 3
c 1 − c cost θ − c cos 3t θ
4 4 32
In view of 4.21, T 1 can be written explicitly in terms of the new variables ζ1 and ζ2 as
follows:
⎡1 1 2

1 3⎤
ζ 1 ζ ζ22
− ζ
⎢4 2 8 1 8 2⎥
T 1 ⎢
⎣1 ⎥ Δζ D1 , 5.12
7 2 3 3⎦
ζ1 1 ζ ζ22 − ζ
4 8 1 8 1
with Δζ D1 put in the form:
⎡ ⎤
1 1
d 1 ζ 1 d2 ζ 2
⎢ ⎥
Δζ D1 ⎣ 1 1 ⎦, 5.13
d1 ζ 2 − d2 ζ 1
1 1 1 1 1 1

being di di c, i 1, 2, D1 cd1 , and D2 d2 . The auxiliary vector d1 is introduced
in order to simplify the calculations, and, it is calculated in the second-order approximation
as described below.
Following the algorithm of the Hori method described in Section 2, the second-order
equation, 2.11, involves the term Ψ2 given by
1 + 1 ,
Ψ2 Z Z∗1 , T 1 . 5.14
2
The determination of Ψ2 is very arduous. The generalized Poisson brackets must be
calculated in terms of ζ1 and ζ2 through 2.12, and, the general solution of the undisturbed
1
system, defined by 3.1, must be introduced. It should be noted that di , i 1, 2, in 5.12
2
are functions of the new variables ζ1 and ζ2 through c2 ζ1 ζ22 . So, their partial derivatives
must be considered in the calculation of the generalized Poisson brackets. After lengthy
calculations performed using MAPLE software, one finds

7 3 3 5
Δ−1
ζ Ψ
2
c − c sin 2t θ
1 64 128
1 3 1 5
− c sin 4t θ − c sin 6t θ
32 128
1 1 1
d1 − c3 c3 cos 4t θ
4 8

1 1 1
d2 c sin 2t θ − c3 sin 4t θ
2 4
1
dd1 1 2 1 4
− c − c cost θ
dζ1 4 4
1
dd1 3 2 1 4 1
c − c sint θ − c4 sin 3t θ ,
dζ2 4 4 8
5.15
1 3 11 1 1
Δ−1
ζ Ψ 2
− c 2
− c 4
− c 2
c 4
cos 2t θ
2 8 16 256 32 64

1 1 4 1 4
− c2 c cos 4t θ − c cos 6t θ
32 128 128

1 1 2 1 2
d1 − c sin 2t θ − c sin 4t θ
4 8

1 1 1 2 1
d2 − c cos 2t θ − c2 cos 4t θ
2 4 4
1
dd2 1 1
− c − c3 cost θ
dζ1 4 4
1
dd2 3 1 3 1 3
c − c sint θ − c sin 3t θ .
dζ2 4 4 8
In order to obtain the same result presented by Hori 4 for the new system of
differential equations and the near-identity transformation, the following choice is made for
the auxiliary vector d1 . Taking
⎡ ⎤
0
d 1
⎣ 1 1 ⎦, 5.16
− c2
4 16
it follows from 5.15 that
⎡ ⎤
- . 0
Δ−1
ζ Ψ
2
⎣ 1 1 2 7 4 ⎦. 5.17
− c − c
8 8 256
From 4.15, 4.21, and 5.17, one finds

⎡ 1 2 7 2

⎤
2 2
− ζ 1 − ζ1 ζ2 2
ζ ζ2
⎢ 8 2 32 1 ⎥
Z∗2 ⎢
⎣ 1 ⎥ . 5.18
2 7 2 ⎦
2 2
ζ1 1 − ζ1 ζ2
2
ζ ζ2
8 32 1
In view of the choice the auxiliary vector d1 , 5.12 can be simplified, and T 1 is then
given by
⎡ ⎤
1 2
⎢ 32 2 1ζ 3ζ − ζ 2
2 ⎥
T 1 ⎢
⎣1 ⎥.
⎦ 5.19
ζ1 16 − 7ζ12 5ζ22
32
5
Computing the indefinite integral Δ−1 ζ
Ψ2 − Δ−1ζ
Ψ2
dt and substituting the
general solution of the new undisturbed system, it follows that T 2 is given by
⎡ ⎤
1 5 3 13 5 1 3 2 11
⎢ ζ 1 − ζ ζ ζ ζ ζ 1 ζ 4
⎥
16 64 1 768 1 96 1 2 768 2
T 2 ⎢
⎣ 1
⎥ Δζ D2 , 5.20
3 2 1 3 29 5 7 5⎦
− ζ2 ζ1 ζ2 ζ2 − ζ2 ζ −
4
ζ ζ −
2 3
ζ
16 64 16 768 1 192 1 2 768 2
with Δζ D2 put in the form:

⎡ ⎤
2 2
d 1 ζ 1 d2 ζ 2
⎢ ⎥
Δζ D2 ⎣ 2 2 ⎦, 5.21
d1 ζ 2 − d2 ζ 1
2 2 2 2 2 2

being di di c, i 1, 2, D1 cd1 , and D2 d2 . D2 is obtained from the third-order
approximation.
In order to get the same result presented by Hori 4, one finds, repeating the
procedure described in the preceding paragraphs, that the auxiliary vector d2 must be taken
as follows:
⎡ ⎤
1 15 2 7 4
⎢− c − c
d2 ⎣ 16 256 512 ⎥⎦. 5.22
0
Accordingly, T 2 is given by

⎡ ⎤
5 5 13 3 2 1 5 3 15
⎢ ζ − ζ ζ ζ 1 ζ 4
− ζ ζ 1 ζ 2
⎥
1536 1 768 1 2 1536 2 256 1 256 2
T 2 ⎢
⎣ 35
⎥. 5.23
41 2 3 79 31 3 27 2 1 ⎦
− ζ −
5
ζ ζ − ζ2 ζ
4
ζ ζ ζ2 − ζ2
1536 2 768 1 2 1536 1 256 2 256 1 8
The new system of differential equations and the generating function are given, up to
the second-order of the small parameter, by
2
dζ1 1 1 2 21
7
ζ2 ε ζ 1 1 − ζ ζ2
2
− ε ζ2 1 − ζ1 ζ2 2 2
ζ ζ2
2 2
,
dt 2 4 1 8 32 1
5.24
dζ2 1 1 2 1 7 2
−ζ1 ε ζ2 1 − ζ1 ζ22 ε2 ζ1 1 − ζ12 ζ22 ζ12 ζ22 ,
dt 2 4 8 32

1 5 5 13 3 2 1 5 3 15
T1 ε ζ2 3ζ12 − ζ22 ε2 ζ1 − ζ1 ζ2 ζ1 ζ24 − ζ1 ζ1 ζ22 ,
32 1536 768 1536 256 256

1 35 41 79 31 27 2 1
T2 ε ζ1 16 − 7ζ12 5ζ22 ε2 − ζ25 − ζ12 ζ23 − ζ2 ζ14 ζ23 ζ1 ζ2 − ζ2 .
32 1536 768 1536 256 256 8
5.25
These results are in agreement with the ones obtained by Hori 4 using a different approach.
Following da Silva Fernandes 1, the Lagrange variational equations—equations of
variation of parameters—for the noncanonical version of the Hori method are given by
dC
Δ−1 ∗
ζ R , 5.26
dt
6
where R∗ m ∗m
m1 ε Z , and C is the n × 1 vector of constants of integration of the
general solution of the new undisturbed system 3.3. In view of 4.15, Lagrange variational
equations can be put in the form:
dC & m - −1 m .
ε Δζ ΨI . 5.27
dt m1
Accordingly, the Lagrange variational equations for the new system of differential equations,
5.24, are given by

dc εc 1
1 − c2 , 5.28a
dt 2 4

dθ 1 1 7 4
ε2 − c2 − c . 5.28b
dt 8 8 256
The solution of the new system of differential equations can be obtained by

introducing the solution of the above variational equations into 4.21.
The originalvariables x and ẋ are calculated through 2.6, and the second-order
asymptotic solution is

1 2 43 5 29 3 2 59 1 3 9
x ζ1 ε ζ2 3ζ1 − ζ2 ε −
2 2
ζ ζ ζ − ζ1 ζ 4
ζ 2
ζ1 ζ ,
32 6144 1 3072 1 2 6144 2 256 1 256 2

1 155 5 35 3 2 715 29 3 53 1
ẋ ζ2 ε ζ1 16−7ζ12 5ζ22 ε2 − ζ2 − ζ2 ζ1 − ζ2 ζ14 ζ2 ζ2 ζ12 − ζ2 .
32 6144 3072 6144 256 256 8
5.29
Equations 5.29 define exactly the same solution presented by Hori.

(2) Second Asymptotic Solution

Now, let us to consider a different choice of the auxiliary vector d1 . Taking d1 as a null
vector, it follows straightforwardly from 5.15 that
⎡ ⎤
- . 0
Δ−1
ζ Ψ
2
⎣ 1 3 11 4 ⎦. 5.30
− c2 − c
8 16 256
Thus, from 4.15, 4.21, 5.28a, and 5.28b, one finds
⎡ 1 3 11 2

2 ⎤
− ζ2 1 − ζ12 ζ22 ζ1 ζ22
⎢ 8 2 32 ⎥
Z∗2 ⎢
⎣ 1 ⎥⎦. 5.31
3 2 11 2
ζ1 1 − ζ1 ζ22 ζ12 ζ22
8 2 32
Since d1 is a null vector, 5.12 simplifies, and T 1 is given by
⎡1 1 2

1 3⎤
ζ 1 ζ1 ζ22 − ζ
⎢4 2 8 8 2⎥
T 1 ⎢
⎣1 ⎥. 5.32
7 2 3 3⎦
ζ1 1 ζ1 ζ22 − ζ
4 8 8 1
Now, repeating5 the procedure described in the previous section, that is, computing
the indefinite integral Δ−1
ζ
Ψ2 − Δ−1
ζ
Ψ2
dt, substituting the general solution of the new
undisturbed system defined by 4.21, and taking d2 is a null vector, it follows that T 2 is
given by
⎡ ⎤
3 3 1 5 5 1 3 2 7
− ζ ζ ζ 2
ζ − ζ ζ
⎢ 64 1 16 1 2 384 1 192 1 2 384 1 2 ⎥ ζ ζ 4
T 2 ⎢
⎣ 7
⎥. 5.33
1 3 1 1 2 3 5 5⎦
− ζ1 ζ2 ζ2 −
2
ζ2 ζ − ζ ζ −
4
ζ
64 16 384 1 96 1 2 384 2
So, the new system of differential equations and the generating function are given, up
to the second-order of the small parameter ε, by
2
dζ1 1 1 2 21 3 2 11
ζ2 ε ζ 1 1 − ζ ζ22
− ε ζ2 1 − ζ ζ2
2
ζ ζ2
2 2
,
dt 2 4 1 8 2 1 32 1
5.34
dζ2 1 1 2 1 3 2 11 2
−ζ1 ε ζ2 1 − ζ1 ζ22 ε 2 ζ1 1 − ζ1 ζ22 ζ12 ζ22 ,
dt 2 4 8 2 32
1
1 1 2
T1 ε ζ2 1 ζ1 ζ22 − ζ23
4 8 8

3 1 5 1 3 2 7
ε2 − ζ13 ζ1 ζ22 ζ15 − ζ1 ζ2 ζ1 ζ24 ,
64 16 384 192 384
3 5.35
1 7 2
T2 ε ζ1 1 ζ ζ22 − ζ13
4 8 1 8

7 1 1 1 5 5
ε2 − ζ12 ζ2 ζ23 − ζ2 ζ14 − ζ12 ζ23 − ζ2 .
64 16 384 96 384
The Lagrange variational equations for the new system of differential equations,
defined by 5.34, are given by

dc εc 1
1 − c2 , 5.36a
dt 2 4

dθ 1 3 11 4
ε2 − c2 − c . 5.36b
dt 8 16 256
These differential equations are the well-known averaged equations obtained through
Krylov-Bogoliubov method 5. Note that 5.28b and 5.36b define the phase θ with slightly
different frequencies.
As described in the preceding subsection, the solution of the new system of differential
equations, defined by 5.34, can be obtained by introducing the solution of the above
variational equations into 4.21.
The original variables x and ẋ are calculated through 2.6, which provides the
following second-order asymptotic solution,

1 1 2
x ζ 1 ε ζ2 1 ζ − 3ζ2 2
4 8 1

65 5 65 3 2 95 1 1 1
ε2 ζ1 ζ1 ζ2 − ζ1 ζ24 − ζ13 ζ1 ζ22 ζ1 ,
6144 3072 6144 16 16 32
5.37
1 1 2
ẋ ζ2 ε ζ1 1 − 5ζ1 − 7ζ22
4 8

143 5 193 3 2 271 5 3 7 1
ε −
2
ζ ζ ζ − ζ2 ζ ζ − ζ2 ζ ζ2 .
4 2
6144 2 3072 2 1 6144 1 64 2 64 1 32
Finally, note that 5.29 and 5.37 give different second-order asymptotic solution for
van der Pol equation.
5.1.2. Duffing Equation

Consider the well-known Duffing equation:
ẍ εx3 x 0. 5.38
Introducing the variables z1 x and z2 ẋ, this equation can be written in the form:
dz1 dz2
z2 , −z1 − εz31 . 5.39
dt dt
Thus,

0
Z1 . 5.40
−z31
Following the simplified algorithm I and repeating the procedure described in

Section 5.1.1, the first-order terms Z∗1 and T 1 are obtained as follows. Introducing 4.21
into 5.40, and computing the secular part, one gets
⎡ ⎤
- . 0
Δ−1
ζ Z
1
⎣ 3 2 ⎦. 5.41
c
8
Thus, from 4.15 and 5.41, it follows that Z∗1 is given by
⎡ ⎤
3 3
−
⎢ 8 c sint θ ⎥
Z∗1 ⎢
⎣ 3
⎥.
⎦ 5.42
− c3 cost θ
8
In view of 4.21, Z∗1 can be written explicitly in terms of the new variables ζ1 and ζ2 :
⎡ ⎤
3 2
⎢ 8 ζ 2 ζ 1 ζ 2
2 ⎥
Z∗1 ⎢
⎣ 3
⎥.
⎦ 5.43

− ζ1 ζ12 ζ22
8
5
Calculating the indefinite integral Δ−1
ζ
Z1 − Δ−1
ζ
Z1
dt, one finds
⎡ ⎤
1 3
+ - ., − c 4 cos 2t θ cos 4t θ
⎢ 32 ⎥
Δ−1
ζ Z
1
− Δ−1
ζ Z
1
dt ⎢
⎣ 1
⎥.
⎦ 5.44
c 8 sin 2t θ sin 4t θ
2
32
Multiplying this result by Δζ , it follows, according to 4.18, that T 1 is given by
⎡ ⎤
3 3 1 3
⎢− 16 c cost θ 32
c cos 3t θ⎥
T 1 ⎢
⎣ 3
⎥ Δζ D1 .
⎦ 5.45
3 3
− c3 sint θ − c sin 3t θ
16 32
Taking D1 as a null vector and using 4.21, T 1 can be written explicitly in terms of the new
variables ζ1 and ζ2 as follows:
⎡ ⎤
5 3 9
− ζ −
⎢ 32 1 32 2 ⎥ ζ 1 ζ 2
T 1 ⎢
⎣ 15
⎥.
⎦ 5.46
3
ζ12 ζ2 ζ23
32 32
In the second-order approximation, one finds after lengthy calculations using MAPLE
software:
⎡ ⎤
69 4 27 2 3 27 5
− ζ ζ ζ ζ
⎢ 256 1 2 128 1 2 256 2 ⎥ζ
Ψ2 ⎢
⎣ 165
⎥. 5.47
69 3 2 27 4 ⎦
ζ
5
ζ ζ − ζ ζ1
256 1 128 1 2 256 2
Repeating the procedure described in the above paragraphs, one finds
⎡ ⎤
- . 0
Δ−1ζ Ψ
2
⎣ 51 4 ⎦,
− c
256
⎡ ⎤
51 2 2 2 5.48
−
⎢ 256 ζ 2 ζ 1 ζ 2 ⎥
Z∗2 ⎢
⎣ 51
⎥.
⎦
2
ζ1 ζ12 ζ22
256
Taking D2 as a null vector, it follows that
⎡ ⎤
19 5 13 3 2 65
ζ ζ ζ ζ ζ
⎢ 256 1 32 1 2 256 1 2 ⎥
4
T 2 ⎢
⎣ 95
⎥. 5.49
13 2 3 13 5 ⎦
− ζ2 ζ − ζ ζ −
4
ζ
256 1 32 1 2 256 2
The new system of differential equations and the generating function are given, up to
the second-order of the small parameter ε, by
dζ1 3 51 2 2
ζ2 ε ζ2 ζ12 ζ22 − ε2 ζ2 ζ1 ζ22 ,
dt 8 256 5.50
dζ2 3 2
2 51
2
−ζ1 − ε ζ1 ζ1 ζ2 ε
2
ζ1 ζ12 ζ22 ,
dt 8 256

5 3 9 19 5 13 3 2 65
T1 −ε ζ ζ1 ζ ε
2 2
ζ ζ ζ 4
ζ1 ζ ,
32 1 32 2 256 1 32 1 2 256 2
5.51
15 2 3 95 13 13 5
T2 ε ζ1 ζ2 ζ23 ε2 − ζ2 ζ14 − ζ12 ζ23 − ζ2 .
32 32 256 32 256
The Lagrange variational equations for the new system of differential equations,
defined by 5.50, are given by
dc
0,
dt 5.52
dθ 3 51 4
ε c2 − ε2 c .
dt 8 256
These differential equations are the well-known equations obtained through Krylov-
Bogoliubov method 5.
As described in Section 5.1.1, the solution of the new system of differential equations,
defined by 5.50, can be obtained by introducing the solution of the above variational
equations into 4.21.
The original variables x and ẋ are calculated through 2.6, which provides the
following second-order asymptotic solution:
1 2 1
x ζ1 − ε ζ1 5ζ1 9ζ22 ε2 227ζ15 742ζ13 ζ22 547ζ1 ζ24 ,
32 2048 5.53
3 2
2 1

ẋ ζ2 ε ζ2 5ζ1 ζ2 − ε
2
77ζ25 922ζ23 ζ12 685ζ2 ζ14 .
32 2048
These equations are in agreement with the solution obtained through the canonical version
of the Hori method 6.
5.2. Determination of Asymptotic Solutions through Simplified Algorithm II

5.2.1. Van der Pol Equation
For the van der Pol equation, the function fx, ẋ is written in terms of the variables z1 c
and z2 θ by

fx, ẋ fz1 cost z2 , −z1 sint z2 − z21 co s2 t z2 − 1 z1 sint z2 . 5.54
⎡1 ⎤
1 1
z1 1 − z21 − cos 2t z2 z21 cos 4t z2
⎢2 4 4 ⎥
Z1 ⎢
⎣ 1 ⎥.
⎦ 5.55
1 2 1 2
1 − z1 sin 2t z2 − z1 sin 4t z2
2 2 8
As mentioned before, two different choices of Dm will be made, and two generating
functions will be determined.
(1) First Asymptotic Solution

Following the simplified algorithm II defined by 4.41 and 4.42, the first-order terms Z∗1
and T 1 are calculated as follows.
Taking the secular part of Z1 , with ζ replacing z, one finds
⎡ ⎤
1 1 2
ζ 1 − ζ
Z ∗1
⎣2 1
4 1 ⎦, 5.56
0
and, integrating the remaining part,
⎡ 1 1 ⎤
− ζ1 sin 2t ζ2 ζ13 sin 4t ζ2
⎢ 4 32 ⎥
T 1 ⎢
⎣ 1 ⎥ D1 ,
⎦ 5.57
1 2 1 2
− 1 − ζ1 cos 2t ζ2 ζ1 cos 4t ζ2
4 2 32
1 1
with Di Di ζ1 , i 1, 2.
Following the algorithm of the Hori method described in Section 2, the second-order
equation, 2.11, involves the term Ψ2 given by
1 + 1 ,
Ψ2 Z Z∗1 , T 1 . 5.58
2
After tedious lengthy calculations using MAPLE software, one finds

2 7 3 3 5
Ψ1 ζ − ζ sin 2t ζ2
64 1 128 1
1 1 5
− ζ13 sin 4t ζ2 − ζ sin 6t ζ2
32 128 1
1 1 3 2 1 3 2
D1 − ζ − cos 2t ζ2 ζ1 cos 4t ζ2
2 8 1 4 16

1 1 1
D2 ζ1 sin 2t ζ2 − ζ13 sin 4t ζ2
2 4
1
dD1 1 1 1 1
− ζ1 ζ13 ζ1 cos 2t ζ2 − ζ13 cos 4t ζ2 ,
dζ1 2 8 4 16
5.59
2 1 3 2 11 4 1 2 1 4
Ψ2 − ζ1 − ζ − ζ ζ cos 2t ζ2
8 16 256 1 32 1 64 1

1 1 4 1 4
− ζ12 ζ cos 4t ζ2 − ζ cos 6t ζ2
32 128 1 128 1

1 1 1
D1 − ζ1 sin 2t ζ2 − ζ1 sin 4t ζ2
4 8

1 1 1 2 1 2
D2 − ζ cos 2t ζ2 − ζ1 cos 4t ζ2
2 4 1 4
1
dD2 1 1 3 1 1 3
− ζ1 ζ1 ζ1 cos 2t ζ2 − ζ1 cos 4t ζ2 .
dζ1 2 8 4 16
In order to obtain the same averaged Lagrange variational equations given by 5.28a
and 5.28b, D1 must be taken as
⎡ ⎤
0
⎢ ⎥
D1 ⎣ 1 1 ⎦. 5.60
− ζ12
4 16
Thus, it follows that
⎡ ⎤
0
⎢ ⎥
Z∗2 ⎣ 1 1 2 7 4 ⎦. 5.61
− ζ1 − ζ1
8 8 256
In view of the choice of D1 , T 1 is then given by
⎡ 1 1 ⎤
⎢ 4 32 ⎥
T 1 ⎢
⎣ 1 ⎥.
⎦ 5.62
1 2 1 1 2 1 2
− ζ1 − 1 − ζ1 cos 2t ζ2 ζ1 cos 4t ζ2
4 16 4 2 32
Repeating the procedure for the third-order approximation, and, taking
⎡ ⎤
1 15 3 7 5
− ζ ζ − ζ
D2 ⎣ 16 1 256 1 512 1 ⎦, 5.63
0
one finds
1 1 9 3 9 5

2
T1 −96ζ1 90ζ13 − 21ζ15 ζ1 − ζ1 ζ1 cos 2t ζ2
1536 16 128 768

1 3 1 5 1 5
− ζ ζ cos 4t ζ2 ζ cos 6t ζ2 ,
128 1 256 1 768 1
5.64
2 1 3 2 1 4 1 2 1 4
T2 − ζ1 − ζ1 sin 2t ζ2 ζ − ζ sin 4t ζ2
16 64 64 128 1 256 1
1 4
− ζ sin 6t ζ2 .
768 1
The new system of differential equations is given, up to the second-order of the small
parameter ε, by

dζ1 1 1
ε ζ1 1 − ζ12 , 5.65a
dt 2 4

dζ2 1 1 7 4
ε2 − ζ12 − ζ1 . 5.65b
dt 8 8 256
These differential equations are exactly the same equations given by 5.28a and 5.28b.
The generating function is obtained from 5.62, 5.64, and it is given, up to the
second-order of the small parameter ε, by

1 1
T1 ε − ζ1 sin 2t ζ2 ζ13 sin 4t ζ2
4 32
1
1 9 3 9 5
ε2 −96ζ1 90ζ13 − 21ζ15 ζ1 − ζ1 ζ1 cos 2t ζ2
1536 16 128 768

1 3 1 5 1 5
− ζ ζ cos 4t ζ2 ζ cos 6t ζ2 ,
128 1 256 1 768 1
5.66
1 2 1 1 1
T2 ε ζ1 − 1 − ζ12 cos 2t ζ2 ζ12 cos 4t ζ2
16 4 2 32

1 3 1 1 2 1 4
ε2 − ζ12 − ζ14 sin 2t ζ2 ζ1 − ζ1
16 64 64 128 256

1 4
× sin 4t ζ2 − ζ sin 6t ζ2 .
768 1
The original variables x and ẋ are calculated through 2.7, which provides, up to the
second-order of the small parameter, the following solution:
1
x ζ1 cost ζ2 − ε ζ13 sin 3t ζ2
32
3 3 9 5 1 3 1 5
ε 2
ζ − ζ cost ζ2 − ζ ζ cos 3t ζ2
256 1 2048 1 128 1 1024 1

5 5
− ζ cos 5t ζ2 ,
3072 1
5.67
1 1 3
ẋ −ζ1 sint ζ2 ε ζ1 − ζ13 cost ζ2 − ζ13 cos 3t ζ2
2 8 32

1 35 65 3 3 15 5
ε 2
ζ1 − ζ
3
ζ sint ζ2 −
5
ζ − ζ
8 256 1 2048 1 128 1 1024 1

25 5
× sin 3t ζ2 ζ sin 5t ζ2 ,
3072 1
with ζ1 and ζ2 given by the solution of 5.65a and 5.65b.

Note that 5.29 and 5.67 give the same second-order asymptotic solution for the van
der Pol equation. Recall that ζ1 and ζ2 have different meaning in these equations, but they are
related through an equation similar to 4.21.
(2) Second Asymptotic Solution

Now, let us to take D1 and D2 as null vectors. Equations 5.59 simplifies, and Z∗2 is given
by
⎡ ⎤
0
Z∗2 ⎣ 1 3 11 4 ⎦. 5.68
− ζ12 − ζ
8 16 256 1
The functions T 1 and T 2 are then given by
⎡ 1 1 ⎤
⎢ 4 32 ⎥
T 1 ⎢
⎣ 1 ⎥,
⎦
1 2 1 2
− 1 − ζ1 cos 2t ζ2 ζ1 cos 4t ζ2
4 2 32

2 7 3 3 5 1 3
T1 − ζ − ζ cos 2t ζ2 ζ cos 4t ζ2
128 1 256 1 128 1 5.69
1 5
ζ1 cos 6t ζ2 ,
768
2 1 2 1 4 1 2 1 4
T2 − ζ ζ sin 2t ζ2 − ζ ζ sin 4t ζ2
64 1 128 1 128 1 512 1
1 4
− ζ sin 6t ζ2 .
768 1
The new system of differential equations is obtained from 5.56 and 5.68, and it is
given, up to the second-order of the small parameter, by

dζ1 1 1
ε ζ1 1 − ζ12 ,
dt 2 4
5.70
dζ2 1 3 11 4
ε2 − ζ12 − ζ1 .
dt 8 16 256
These differential equations are exactly the same equations given by 5.36a and 5.36b.
The generating function is obtained from 5.69, and it is given, up to the second-order
of the small parameter ε, by

1 1 3
T1 ε − ζ1 sin 2t ζ2 ζ1 sin 4t ζ2
4 32

7 3 3 5 1 3 1 5
ε2 − ζ1 − ζ1 cos 2t ζ2 ζ1 cos 4t ζ2 ζ1 cos 6t ζ2 ,
128 256 128 768

1 1 2 1 2
T2 ε − 1 − ζ1 cos 2t ζ2 ζ1 cos 4t ζ2 5.71
4 2 32

1 2 1 4 1 2 1 4
ε2 − ζ1 ζ1 sin 2t ζ2 − ζ1 ζ sin 4t ζ2
64 128 128 512 1

1 4
− ζ sin 6t ζ2 .
768 1
Equations 5.71 are in agreement with the solution obtained by Ahmed and Tapley 7
through a different integration theory for the Hori method.
A second-order asymptotic solution for the original variables x and ẋ is calculated
through 2.7, and it is given by

1 1 1
x ζ1 cost ζ2 ε − ζ1 ζ13 sint ζ2 − ζ13 sin 3t ζ2
4 16 32

1 1 15 1 5 5
ε2 ζ1 − ζ13 ζ15 cost ζ2 − ζ13 ζ1
32 32 2048 32 1024

5 5
× cos 3t ζ2 − ζ cos 5t ζ2 ,
3072 1
5.72
1 1 3
ẋ −ζ1 sint ζ2 ε ζ1 − ζ13 cost ζ2 − ζ13 cos 3t ζ2
4 16 32

1 1 25 3 3 3 5
ε2 − ζ1 − ζ13 ζ15 sint ζ2 ζ1 − ζ1
32 32 2048 64 1024

25 5
× sin 3t ζ2 ζ sin 5t ζ2 ,
3072 1
with ζ1 and ζ2 given by the solution of 5.70.

As before, note that 5.37 and 5.72 give the same second-order asymptotic solution
for the van der Pol equation. Recall that ζ1 and ζ2 have different meaning in these equations,
but they are related through an equation similar to 4.21.
5.2.2. Duffing Equation

For the Duffing equation, the function fx, ẋ is written in terms of the variables z1 c and
z2 θ , by
fx, ẋ fz1 cost z2 , −z1 sint z2 −z31 co s3 t z2 . 5.73
⎡ ⎤
1 3 1 3
⎢ z sin 2t z2 z sin 4t z2 ⎥
4 1 8 1
Z1 ⎢
⎣3
⎥.
⎦ 5.74
1 2 1 2
z1 z1 cos 2t z2 z1 cos 4t z2
2
8 2 8
Following the simplified algorithm II and repeating the procedure described in

Section 5.2.1, the first-order terms Z∗1 and T 1 are obtained as follows. Taking the secular
part of Z∗1 , with ζ replacing z, and, integrating the remaining part, one finds
⎡ ⎤
0
⎢ ⎥
Z∗1 ⎣ 3 2 ⎦, 5.75
ζ1
8
⎡ ⎤
1 3
− ζ
⎢ 32 1 4 cos 2t ζ 2 cos 4t ζ
2 ⎥
T 1 ⎢
⎣ 1
⎥.
⎦ 5.76
ζ1 8 sin 2t ζ2 sin 4t ζ2
2
32
In the second-order approximation, one finds
⎡ ⎤
1 5
⎢ ζ −33 sin 2t ζ2 − 12 sin 4t ζ2 3 sin 6t ζ2 ⎥
256 1
Ψ2 ⎢
⎣ 1
⎥.
⎦ 5.77
ζ14 −51 − 99 cos 2t ζ2 − 18 cos 4t ζ2 3 cos 6t ζ2
256
Taking the secular part of Ψ2 , and, integrating the remaining part, one finds
⎡ ⎤
0
Z∗2 ⎣ 51 4 ⎦, 5.78
− ζ
256 1
⎡ ⎤
1 5
ζ
⎢ 512 1 33 cos 2t ζ 2 6 cos 4t ζ 2 − cos 6t ζ
2 ⎥
T 2 ⎢
⎣ 1
⎥.
⎦ 5.79
ζ1 −99 sin 2t ζ2 − 9 sin 4t ζ2 sin 6t ζ2
4
512
The new system of differential equations is given, up to the second-order of the small
parameter ε, by
dζ1
0,
dt 5.80
dζ2 3 51 4
ε ζ12 − ε2 ζ .
dt 8 256 1
These differential equations are exactly the same equations given by 5.52.
The generating function is obtained from 5.76 and 5.79, and it is given, up to the
second-order of the small parameter ε, by
1 3
T1 −ε ζ 4 cos 2t ζ2 cos 4t ζ2
32 1
1 5
ε2 ζ 33 cos 2t ζ2 6 cos 4t ζ2 − cos 6t ζ2 ,
512 1 5.81
1
T2 ε ζ12 8 sin 2t ζ2 sin 4t ζ2
32
1 4
ε2 ζ −99 sin 2t ζ2 − 9 sin 4t ζ2 sin 6t ζ2 .
512 1
A second-order asymptotic solution for the original variables x and ẋ is calculated

through 2.7, and it is given by
1 3
x ζ1 cost ζ2 ε ζ −6 cost ζ2 cos 3t ζ2
32 1
1 5
ε2 ζ 303 cost ζ2 − 78 cos 3t ζ2 2 cos 5t ζ2 ,
2048 1 5.82
1
ẋ −ζ1 sint ζ2 − ε ζ13 6 sint ζ2 3 sin 3t ζ2
32
1 5
ε2 ζ 249 sint ζ2 162 sin 3t ζ2 − 10 sin 5t ζ2 .
2048 1
These equations are in agreement with the solution obtained through the canonical version
of the Hori method 6. Note that 5.53 and 5.82 give the same second-order asymptotic
solution for the Duffing equation. Recall that ζ1 and ζ2 have different meaning in these
equations, but they are related through an equation similar to 4.21.
6. Conclusions
In this paper, the Hori method for noncanonical systems is applied to theory of nonlinear
oscillations. Two different simplified algorithms are derived from the general algorithm
m ∗m
proposed by Sessin. It has been shown that the mth-order terms Tj and Zj that define the
near-identity transformation and the new system of differential equations, respectively, are
not uniquely determined, since the algorithms involve at each order arbitrary functions of the
constants of integration of the general solution of the undisturbed system. This arbitrariness
is an intrinsic characteristic of perturbation methods, since some kind of averaging principle
must be applied to determine these functions. The simplified algorithms are then applied in
determining second-order asymptotic solutions of two well-known equations in the theory
of nonlinear oscillations: van der Pol and Duffing equations. For van der Pol equation, the
appropriate use of the arbitrary functions allows the determination of the solution presented
by Hori. This solution defines a new system of differential equations with a different
frequency for the phase in comparison with the solution obtained by Ahmed and Tapley,
who used a different approach for determining the near-identity transformation and the new
system of differential equations for the Hori method, and, with the solution obtained by
Nayfeh through the method of averaging. For the Duffing equation, only one generating
function is determined, and the second simplified algorithm gives the same generating
function obtained through Krylov-Bogoliubov method.
References
1 S. da Silva Fernandes, “Notes on Hori method for non-canonical systems,” Celestial Mechanics and
Dynamical Astronomy, vol. 87, no. 3, pp. 307–315, 2003.
n ∗n
2 W. Sessin, “A general algorithm for the determination of Tj and Zj in Hori’s method for non-
canonical systems,” Celestial Mechanics, vol. 31, pp. 109–113, 1983.
3 G. Hori, “Theory of general perturbations with unspecified canonical variables,” Publications of the
Astronomical Society of Japan, vol. 18, pp. 287–295, 1966.
4 G. Hori, “Theory of general perturbations for non-canonical systems,” Publications of the Astronomical
Society of Japan, vol. 23, pp. 567–587, 1971.
5 A. H. Nayfeh, Perturbation Methods, Wiley-VCH, New York, NY, USA, 2004.
6 S. da Silva Fernandes, “Notes on Hori method for canonical systems,” Celestial Mechanics and Dynamical
Astronomy, vol. 85, no. 1, pp. 67–77, 2003.
7 A. H. Ahmed and B. D. Tapley, “Equivalence of the generalized Lie-Hori method and the method of
averaging,” Celestial Mechanics, vol. 33, no. 1, pp. 1–20, 1984.
doi:10.1155/2012/740752
Research Article
IMU Fault Detection Based on χ2 -CUSUM
Élcio Jeronimo de Oliveira,1 Hélio Koiti Kuga,2

and Ijar Milagre da Fonseca2
1
Space Systems Division, Institute of Aeronautics and Space (IAE),
12.228-904 São José dos Campos, SP, Brazil
2
Mechanic and Control Division, National Institute for Space Research (INPE),
12.227-010 São José dos Campos, SP, Brazil
Correspondence should be addressed to Élcio Jeronimo de Oliveira, elcioejo@bol.com.br
Received 16 December 2011; Accepted 1 February 2012
Academic Editor: Tadashi Yokoyama
Copyright q 2012 Élcio Jeronimo de Oliveira et al. This is an open access article distributed under
The problem of fault detection and isolation FDI on inertial measurement units IMUs has
received great attention in the last years, mainly with growing use of IMU strapdown platforms
using fiber optic gyros FOG or micro electro mechanical systems MEMSs. A way to solve
this problem makes use of sensor redundancy and parity vector PV analysis. However, the
actual sensor outputs can include some anomalies, as impulsive noise which can be associated
with the sensors itself or data acquisition process, committing the elementary threshold criteria as
commonly used. Therefore, to overcome this problem, in this work, it is proposed an algorithm
based on median filter MF for prefiltering and chi-square cumulative sum χ2 -CUSUM only for
fault detection FD applied to an IMU composed by four FOGs.
1. Introduction
The design of a FDI algorithm for applications on IMUs can, in general, be divided in
two types. The first one, named analytical redundancy, takes into account a mathematical
model of the system in which the IMU is used. This method generates a residue vector as a
result of the state observer 1–4 in order to indicate the IMU operational status. The second
one, named sensor redundancy, is based on specific geometrical configurations using extra
sensors and solving the FDI problem with the aid of the parity vector PV analysis 5, 6.
However, the actual sensor outputs can include some anomalies, as impulsive noise which
can be associated with the sensors itself or with the data acquisition process 7, violating
the elementary threshold criteria as commonly used. This work applies the second type of
algorithm to the FD problem. The use of four sensors minimal redundancy constrains the
algorithm to detect the fault not allowing the identification isolation of the faulty sensor 8.
g4
g3
g2
g1
α α
Y Z
a b
Figure 1: Tetrahedral base.
To overcome the anomalies addressed before, in this work is proposed an algorithm based
on MF that performs the prefiltering of FOG outputs and the use of the χ2 -CUSUM for FD
problem.
2. Background
2.1. Geometry
The geometrical arrangement used in this work considers four gyros mounted on the faces of
a tetrahedral structure tetrad, and three accelerometers in a triad configuration internally
fixed in the tetrahedral. The analysis performed here takes into account the gyros only,
and the extension for accelerometers is straightforward. The tetrad configuration and the
reference frame are shown in Figure 1. The mathematical representation of the arrangement
of the gyros is given in terms of direct cosine matrix DCM, and is obtained from angular
relationship between sensor axes and the analytical triorthogonal axes analytical triad.
Therefore, considering the schema shown in Figure 1, where α 54.736◦ , the DCM is given
by
⎛ √ ⎞
1 6
⎜√ 0 ⎟
⎜ 3 √ √ ⎟
3
⎜ 1 6⎟
⎜ 2 ⎟
⎜√ − ⎟
H⎜ 3 √2 √6 ⎟ . 2.1
⎜ ⎟
⎜ 1 2 6 ⎟
⎜√ − − ⎟
⎝ 3 2 6 ⎠
1 0 0
The matrix H relates the sensor measurements gi with the angular rate in the main axes in
the form of
go Hω. 2.2
The estimate of the angular rate components in the main axes can be obtained from 2.2 in
the following manner:
9 H∗ go ,
ω 2.3
−1
H∗ HT H HT , 2.4
where go g1 g2 g3 g4 T is the vector of the gyro outputs, ω ωx ωy ωz T is the angular
9 is the angular rate estimate vector, and H∗ is the generalized
rate vector about main axes, ω
inverse of H.
Equation 2.4 provides the best state estimation in the least squares sense.
2.2. Parity Vector

The sensor equation considered in 2.2 can be rewritten by adding the faults, biases, and
noises components as follows:
: o Hω δgo f ηs ,
g 2.5
where δgo is the bias vector whose magnitude is constant, f is the fault vector, and ηs is a
random term vector Gaussian noise. Applying the singular value decomposition SVD
on H, it can be obtained the range and null spaces from this matrix 9. In addition, also is
computed the biases influence on the arrangement. Decomposing H as follows:

Σ
U HV Λ
T
, 2.6
0
H UΛVT , 2.7
where U, Λ, and VT are matrices obtained from SVD of H. The matrix Σ is a diagonal
matrix whose elements are the eigenvalues of H. The superscript T indicates the transpose.
Applying 2.7 into 2.5 and multiplying both sides by UT , it can be obtained the following
relationship:

: o ΛVT ω UT δgo f ηs .
UT g 2.8
Partitioning U as follows:

..
U U1 . U2 , 2.9
where U1 ∈ R4 × 3 and U2 ∈ R4 × 1 , and applying it into 2.8, the resulting equations are

: o ΣVω UT1 δgo f ηs ,
UT1 g 2.10

UT2 g: o UT2 δgo f ηs p, 2.11
p C:
go . 2.12
Equation 2.10 leads to least squares estimate of ω, what is equivalent to 2.3, and can be
expressed by
9 LS ΣV−1 UT1 g
ω :o. 2.13
The meaning of 2.11 is that, if sensors faults and biases are zero, the resulting product of
parity vector with sensor measurements is a white noise with zero mean. Otherwise, if the
biases and/or faults values differ from zero, 2.11 is a “weighted” sensor errors summation.
Then, UT2 or C is null space of H and p is the parity vector.
2.3. Median Filter

In the image processing field, it is very common to employ filters that preserve edges or
abrupt transitions between distinct parts of the image or remove salt and pepper noise. These
filters generally compare the pixel under observation with its neighbors at certain window
and take a decision based on a statistical or threshold criteria. The simplest filter that meets
these requirements is the median filter MF. Being the MF not an optimal filter, it preserves
discontinuities as jumps 10 and is also a robust estimator 11, besides being an easy and
fast way to remove outliers from signals 7.
2.3.1. Recursive Median Filter

In the time sequence processing by MF, it is possible to process the samples in the filter
window in two ways. a Process the samples in the window to obtain the output at
discrete time k, shift the window to process the time k 1 over the original data without
replacement; this is the nonrecursive processing. b Process the samples in the window to
obtain the output at discrete time k, replace the original sample xk by the filter output
yk, shift the window and process the new samples with replacement; this is the recursive
median filter. The nonrecursive MF is shown by 2.14 and the recursive MF is shown by 2.15,
yk median{xk − N, . . . , xk, . . . , xk N}, 2.14

; <
yk median yk − N, . . . , yk − 1, xk, . . . , xk N , 2.15
where x represents the time series and the filter window has 2N 1 samples.
The recursive form of the MF presents an expressive noise attenuation capacity
comparing with the nonrecursive form, this property leads the signal to a fast convergence
12.
The appropriate size of the MF can be defined comparing the variance of the signal
filtered with the variance of the same signal filtered by an average filter. In this work it was
chosen the exponentially weighted moving average EWMA filter 13, setting the β factor
properly. The EWMA filter is defined as follows:

μn βμn−1 1 − β xn , 2.16
where μ is the estimated mean of the random variable x at time n and β is the EWMA factor.
The absolute value of the difference between variances is defined as
+ ,
Δvar j abs var MFj − var FExpβ , 2.17
where MFj is the size j MF output and FExpβ is the EWMA filter output related to β factor.
2.4. χ2 -CUSUM Algorithm

The cumulative sum CUSUM algorithm is widely used to detect changes in the mean value
of an independent Gaussian sequence. This algorithm is based on log-likelihood ratio and
defined as follows 14:

pθ1 y
s y ln , 2.18
pθ0 y
where sy is the sufficient statistics, pθ y is probability density function with conditional
parameter θ.
Before a change, the parameter θ is constant and equals to θ0 ; after the variation it
assumes the value θ1 .
Considering 2.18 and a sampling set of size N, the following decision function can
be established:
&
k
Skj si ,
ij
2.19
pθ1 yi
si ln ,
pθ0 yi
where the decision is stated as,
=
0 if SN
1 < λ, H0 is chosen
d 2.20
1 ≥ λ, H1 is chosen
1 if SN
and the hypothesis test is defined as,
H0 : θ θ0 ,
2.21
H1 : θ θ1 ,
where λ is a suitable threshold.

The alarm time ta after a change is defined by the following stopping rule:
ta N · min{K : dK 1}. 2.22
In this form, the CUSUM algorithm requires prior knowledge about the parameter θ1 and
cannot be processed recursively. So, for online applications and considering the parameter
θ1 as unknown, it is required a modified version of the CUSUM algorithm. It is presented in
this work a modified version of the CUSUM based upon the weighted likelihood ratio and
the sequential probability ratio test SPRT to suit the unknown θ1 and allow for the online
processing as in 14:
∞
pθ1 yj , . . . , yk
:k
Λ dFθ1 , 2.23
j
−∞ pθ0 yj , . . . , yk
where the stopping time or alarm time ta is defined as

:k ≥ λ
ta min k : max ln Λ 2.24
j
1≤j≤k
and dFθ1 is a weighting function.

Considering the prior knowledge about σ 2 and μ0 of a Gaussian sequence with
distribution Fθ1 Fμ concentrated on two points μ0 −ν, and μ0 ν the weighted likelihood
ratio becomes
∞
b2
:k
Λj exp bS:kj − k − j 1 dFν, 2.25
−∞ 2
where b ν/σ is the signal-to-noise ratio SNR, ν μ1 − μ0 , and
1& k

S:kj yi − μ0 . 2.26
σ ij
Solving 2.25,

: k cosh bS:k e−b2 /2k−j 1
Λ j j
+ , 2.27
cosh b k − j 1 χkj e−b /2k−j 1 ,
2
where
% %
1 % :k %
χkj %S %. 2.28
k−j 1 j
Equation 2.27 is the probability ratio to noncentral parameter test of a χ2 distribution with
one degree-of-freedom between values 0 and k − j 1b2 . So, by using this approach, we
reach χ2 -CUSUM algorithm.
As a consequence of the previous derivation, ta is now defined as
; <
ta min k : gk ≥ λ , 2.29
where

b2
gk max ln cosh bS:kj − k−j 1 . 2.30
1≤j≤k 2
By making a slight modification and considering the CUSUM algorithm as a repeated SPRT
14 with lower and upper thresholds fixed in 0 and λ, respectively, the recursive form of the
χ2 -CUSUM algorithm takes the following form:

gk S̆kk−Nk 1 ,
1
S̆kk−Nk 1 − Nk b2 ln cosh bS:kk−Nk 1 ,
2 2.31
S:kk−Nk 1 Sk ,
yk − μ0
Sk Sk−1 1{gk−1 >0} ,
σ
where Nk Nk−1 1{gk−1 >0} 1.
3. Fault Detection Algorithm

3.1. FD Algorithm Structure
The detection of faults in a IMU with redundant sensors can be performed by PV analysis.
Under normal conditions, that is, bias and faults with null values, the PV should present a
normal distribution N0, σ 2 . However, in actual systems, impulsive noise may occur. This
situation is described in 7. As stated above, the MF is used in order to remove those
impulses and to improve the detection performed by χ2 -CUSUM algorithm. In Figure 2, it
is shown the FD processing structure. Again, in this form, the algorithm only alarms the
fault occurrence because it is not possible to define which sensor is under fault condition.
After processing the signal from gyros by a data acquisition system, the data are filtered by
a recursive MF 2.15 of size 3 or 5 in order to remove the spikes or outliers generated in
this phase. In the sequence, the information in mV /deg/s is converted by a scale factor, SF
in deg/s. The SF block makes a polynomial conversion, whose degree is 7. The PV block
Gyro number 1
Gyro number 2 Sensor x

MF (1) SF Filtering matrix Y
GN&C
(H∗ ) z
Gyro number 3
Gyro number 4
PV
Fault detection block

MF (2)
χ2 -CUSUM
N Y
Fault ? Alarm
Figure 2: Gyro processing and fault detection block diagram. MF: median filter MF1: size 3//MF2:
size 11; SF: scale factor; GN&C: guidance, navigation, and control; PV: parity vector.
×10−4
abs (var (MF)-var (FExp))
0
3 5 7 9 11 13 15
MF window size (samples)
MF: 3–15 samples

FExp: β = 0.95
Figure 3: Absolute value of the difference between variances of the FMj and FExp0.95 .
performs the multiplication of the null space of the actual H : matrix 3.1 with the gyro output
vector in deg/s, forming the parity vector. This PV is filtered, again, by another recursive MF
of size 11. This size was obtained using 2.16 and 2.17, setting β 0.95. The results obtained
by processing 2.17 are shown in Figure 3. It can be seen that difference from the results
obtained for sizes 9 to 15 is negligible. So, after some tests, it was concluded that size 11 is the
best choice. The actual parameters used in the algorithm were obtained from the calibration
of the IMU and are presented in 7. The values and operations associated to blocks in the
Figure 2 are given as follows.
0.8
0.7
σ = 0.022 ◦ /s
total (number of samples)]

0.6
Ratio [total (gk > λ)/

μ0 = 0.002 ◦ /s
0.5
λ = 30
0.4
0.3
0.2
0.1
0%
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
SNR (b)
Figure 4: Ratios for b varying from 0.1 to 5.
: and
For the actual sensor matrix this matrix is obtained from calibration process H
:
respective null space matrix C and generalized inverse H ,∗
⎛ ⎞
0.57868624 −0.00432436 0.81553879
⎜ ⎟
: ⎜0.57733766 0.70734792 −0.40784817⎟,
H 3.1
⎝0.57695344 −0.70679030 −0.40935582⎠
0.99999805 0.00100995 0.00169158
> ?
C: −0.40955 −0.40903 −0.40585 0.70729 ,
:H
C : 0, 3.2
−1
H∗ H: TH
: : T.
H
For the scale factor block, the polynomials of degree 7 are,

◦
go1 0.0657gv7 − 3.64 × 10−4 gv6 − 7.799 × 10−2 gv5 9.911 × 10−4 gv4
s
1.200gv3 3.085 × 10−4 gv2 37.432gv − 3.724 × 10−3 ,

◦
go2 0.09452gv7 1.083 × 10−5 gv6 − 0.1965gv5 − 2.041 × 10−3 gv4
s
1.5219gv3 6.910 × 10−3 gv2 38.305gv − 4.615 × 10−3 ,
3.3
◦
go3 0.0761gv7 − 2.898 × 10−4 gv6 − 0.13277gv5 − 8.253 × 10−4 gv4
s
1.3711gv3 2.326 × 10−3 gv2 38.234gv − 1.1713 × 10−2 ,

◦
go4 0.0814gv7 − 1.0544 × 10−4 gv6 − 0.18366gv5 − 1.3751 × 10−3 gv4
s
1.4068gv3 3.9596 × 10−3 gv2 37.282gv − 4.1806 × 10−3 ,
0.1 1000
500
0.05
0
−500
−0.05 −1000
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Samples ×104 Samples ×104
a b
100 100
80
λ = 30
50 gk
60
40
0
20
−50 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Samples ×104 Samples ×104
c d
Figure 5: χ2 -CUSUM results for 20000 samples of a movement sequence without faults with SNR b 0.1.
a PV; b Sk ; c S̆k ; d gk .
and the PV block processes the following operation:
: go ,
p C: 3.4
where p is filtered by MF2 and applied to χ2 -CUSUM algorithm 2.31 in the form of yk .
The filtering block in the Figure 2 is an optional filter to comply with the GN&C block
requirements.
3.2. Calibration of the FD Algorithm

The calibration of the FD algorithm was performed offline by using a time series extracted
from the IMU movement on a 2DOF rotating table. The first parameter to be determined is
0.06 30
0.04
20
0.02
0 10
−0.02
0
−0.04
0 0.5 1 1.5 2 0 0.5 1 1.5 2
×104 ×104
Samples Samples
a b
30
30
20
20
10 λ = 30
0 10 gk
−10
0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
×104 ×104
Samples Samples
c d
Figure 6: χ2 -CUSUM results for 20000 samples of a movement sequence without faults with SNR b 3.5.
a PV; b Sk ; c S̆k ; d gk .
the threshold λ, obtained by use of the Kullback information 14 as follows:
λ
l≈ ,
Kν
3.5
ν2 1
Kν 2
b2 ,
2σ 2
1 2
λ≈ b l, 3.6
2
where l is the average number of samples until the alarm time ta .
Of course, the parameters b and l in 3.6 are unknown. So, to overcome this condition,
it is assumed the SNR b equals to 1 and l with properly size. In this work, it was considered
l equals to 60. Thus, the threshold is fixed as
1 2
λ 1 × 60 30. 3.7
2
Parity vector ltered by MF (blue)and nonltered (red)

0.4
0.3
0.2
(◦ /s)
0.1
0
−0.1
−0.2
−0.3
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
×104
Samples
a
Parity vector filtered by MF (blue) and FExp (red)

0.15
0.1
(◦ /s)
0.05
0.06
0.04
0 0.02
0
−0.02
0.99 1 1.01
×104
−0.05
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Samples ×104
b
Figure 7: PV with step fault for sample 10000. a PV nonfiltered red and filtered by MF of size 11 blue;
b PV filtered by MF of size 11 blue and by FExp with β 0.95 red; the zoom in b indicates the
difference between MF and FExp during the transition time.
Once the threshold value is set and considered the actual values associated to the PV σ
and μ0 as given in Table 1, it was performed the following computation the χ2 -CUSUM
algorithm:
Total of gk > λ
% 100 × . 3.8
Total number of samples
The ratio in 3.8 is used to determine the actual value of b that reach 0% of samples crossing
the threshold in normal condition without fault. In other words, 0% of false alarms. It is
shown in Figure 4 the ratios for b varying from 0.1 to 5, in whose range from 3.5 for SNR
meets the objective. The efficiency of the calibration can be seen in Figures 5 and 6, it is shown
the algorithm outputs for SNR b 0.1 and b 3.5, respectively. The main information is
presented in Figures 5 and 6d, where it is illustrated the behavior of the decision function
with respect to the threshold λ 30.
×104
0.15 3
0.1 2
0.05 1
0 0
−0.05 −1
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Samples ×104 ×104
Samples
a b
50
30
40 ta = 10054
25
λ = 30
30 20
20 15 gk
10 10 t0 = 10000
0 5
−10 0
1 1.002 1.004 1.006 1 1.002 1.004 1.006
×104 ×104
Samples Samples
c d
Figure 8: χ2 -CUSUM results for 20000 samples of a movement sequence with a step fault at sample 10000
and magnitude 0.15◦ /s; SNR b 3.5. a PV; b Sk ; c S̆k ; d gk .
4. Results
After the calibration phase, it was injected a step bias fault of magnitude 0.15 deg/s into one
of sensors for the sample 10000. This fault generated a step variation in the PV according to
Figures 7 and 8a and slopes as a result of the χ2 -CUSUM algorithm processing in the Figures
8b–8d. In the Figure 7a, it is shown the efficiency of the median filter in impulsive
noise reduction process, and in letter b is shown a comparison between MF and FExp
in terms of filtering delay. In the decision function curve Figure 8d, it is illustrated the
difference between actual time occurrence of the fault t0 10000 and the actual time alarm
ta 10054, whose delay is 54 samples. In addition, there are another delays that should
be computed which are associated to median filters MF1 and MF2. The total processing
delay of the filters and algorithm is summarized in Table 2. In this table, it is presented the
processing of the step faults with values 0.10, 0.15, 0.20, and 0.30 deg/s in order to compare
the effect of the fault level on the detection delay. Considering the case of this work, where
the FOGs are sampled at 100 Hz, the delay is less than 0.15 seconds for the fault level higher
than 0.20 deg/s.
Table 1: Tuning parameters for χ2 -CUSUM algorithm.
σ◦/s μ0 ◦/s b λ
0.022 0.002 0.1 to 5 30
Table 2: Fault detection delay as a function of fault magnitude given in number of samples with respect to
the parameters b 3.5 and λ 30.
Step fault 1st MF delay 2nd MF delay Detection Total delay

◦/s MF3 MF11 delay ta − t0
0.10 1 5 430 436
0.15 1 5 54 60
0.20 1 5 9 15
0.30 1 5 4 10
5. Conclusions
In this paper, a method based on χ2 -CUSUM algorithm combined with median filter was
applied to detect faults in inertial measurement units with minimal redundancy of fiber
optics gyros. The effectiveness of algorithm in case of low level step faults was demonstrated
achieving the requirements of short delay to alarm with high reliability. In addition, the
calibration technique presented here also proved to be efficient.
References
1 E. Y. Chow and A. S. Willsky, “Analytical redundancy and the design of robust failure detection
systems,” IEEE Transactions on Automatic Control, vol. AC-29, no. 7, pp. 603–614, 1984.
2 J. C. Deckert, M. N. Desai, J. J. Deyst, and A. S. Willsky, “F-8 dfbw sensor failure identification using
analytic redundancy,” IEEE Transactions on Automatic Control, vol. AC-22, no. 5, pp. 795–804, 1977.
3 X. C. Lou, A. S. Willsky, and G. C. Verghese, “Failure detection with uncertain models,” in Proceedings
of the IEEE American Control Conference, pp. 956–959, 1983.
4 Z. Han, W. Li, and S. L. Shah, “Fault detection and isolation in the presence of process uncertainties,”
Control Engineering Practice, vol. 13, no. 5, pp. 587–599, 2005.
5 M. A. Sturza, “Skewed axis inertial sensor geometry for optimal performance,” in Proceedings of the
Digital Avionics Systems Conference (AIAA ’88), pp. 128–135, San Jose, Calif, USA, 1988.
6 S. Kim, Y. Kim, and C. Park, “Failure diagnosis of skew-configured aircraft inertial sensors using
wavelet decomposition,” IET Control Theory and Applications, vol. 1, no. 5, pp. 1390–1397, 2007.
7 E. J. de Oliveira, Fault detection and isolation in inertial measurement units with minimal redundancy of
fiber optic gyros, Ph.D. thesis, National Institute for Space Research INPE, Sao Jose dos Campos, SP,
Brazil, 2011.
8 P. G. Savage, Introduction to Strapdown Inertial Navigation Systems, Strapdown Associates, Maple Plain,
Minn, USA, 11st edition, 2005.
9 D. S. Shim and C. K. Yang, “Geometric FDI based on SVD for redundant inertial sensor systems,” in
Proceedings of the 5th Asian Control Conference, pp. 1094–1100, July 2004.
10 S. V. Vaseghi, Advanced Digital Signal Processing and Noise Reduction, John Wiley & Sons, New York,
NY, USA, 2nd edition, 2000.
11 P. J. Rousseeuw, “Robust estimation and identifying outliers,” in Handbook of Statistical Methods for
Engineers and Scientists, H. M. Wadsworth, Ed., pp. 16.1–16.24, MacGraw-Hill, New York, NY, USA,
1st edition, 1990.
12 G. Qiu, “An improved recursive median filtering scheme for image processing,” IEEE Transactions on
Image Processing, vol. 5, no. 4, pp. 646–648, 1996.
13 V. A. Siris and F. Papagalou, “Application of anomaly detection algorithms for detecting SYN flooding
attacks,” Computer Communications, vol. 29, no. 9, pp. 1433–1442, 2006.
14 M. Basseville and I. V. Nikiforov, Detection of Abrupt Changes, Prentice-Hall, Upper Saddle River, NJ,
USA, 1 edition, 1993.
doi:10.1155/2012/905209
Research Article
Low-Thrust Orbital Transfers in the
Two-Body Problem
A. A. Sukhanov1 and A. F. B. A. Prado2

1
Space Research Institute (IKI) RAS, Moscow, Russia
2
Division of Space Mechanics and Control, National Institute for Space Research (INPE),
Av dos Astronautas 1758, Jd. da Granja, São José dos Campos, SP 12227-010, Brazil
Correspondence should be addressed to A. F. B. A. Prado, prado@dem.inpe.br
Received 6 December 2011; Accepted 6 February 2012
Copyright q 2012 A. A. Sukhanov and A. F. B. A. Prado. This is an open access article distributed
Low-thrust transfers between given orbits within the two-body problem are considered; the thrust
is assumed power limited. A simple method for obtaining the transfer trajectories based on the
linearization of the motion near reference orbits is suggested. Required calculation accuracy can
be reached by means of use of a proper number of the reference orbits. The method may be used
in the case of a large number of the orbits around the attracting center; no averaging is necessary
in this case. The suggested method also is applicable to the cases of partly given final orbit and if
there are constraints on the thrust direction. The method gives an optimal solution to the linearized
problem which is not optimal for the original nonlinear problem; the difference between the
optimal solutions to the original and linearized problems is estimated using a numerical example.
Also examples illustrating the method capacities are given.
1. Introduction
In this paper spiral transfers with power-limited low thrust between two given orbits are
considered. The transfer trajectory may include a large number of orbits around the attracting
center if the transfer is performed in a strong gravity field; this situation takes place, for
example, near Earth. This makes optimization of the transfer more difficult.
There are various mathematical methods for optimization of multirevolution orbital
transfers 1–7, most of them are based on averaging of motion. However, the methods have
some defects: they are complicated 3, 5 or have a limited application, that is, are applicable
only to the circular or neighboring orbits 1, 6, 7 or only to the planar or coaxial orbits 2, 4,
6, 7.
This paper suggests a simple mathematical method for calculation of the orbital
transfers within the two-body problem. A paper describing a generalization of the method
to any force field will be published in the Cosmic Research journal. The method is based on a
linearization of motion near some reference orbits; in this respect the method is similar to the
Method of Transporting Trajectory MTT for solving the two-point boundary value problem
TPBVP suggested in 8 and developed in 9–12. The method suggested here makes it
possible obtaining transfer trajectories of the following types:
i transfer between a given position in the initial orbit and an optimally determined
position in the final orbit this case takes place, for example, if the spacecraft
launches from a given orbit at a given time and phasing of the final orbit is not
necessary;
ii transfer between the initial orbit and a given position in the final orbit with
obtaining an optimal position of the launch this situation takes place, for instance,
in the transfer to a given point of geostationary orbit if the time of launch from the
initial orbit may be selected arbitrarily;
iii transfer between two given orbits with obtaining optimal launch and arrival
positions this is a classical case of the orbital transfer.
The method suggested in this paper does not need any averaging of motion. Required
calculation accuracy is reached by means of use of a proper number of the reference orbits:
the bigger is the number, the shorter are intervals of linearization and the higher is accuracy.
The method does not put any limits on the number of orbits around the attracting center and
on the number of the reference orbits. The suggested method is applicable also to the case
of a partly given final orbit e.g., only semimajor axis and eccentricity or only orbital energy
may be given and to the case of constraints imposed on the thrust direction.
However, the method gives an optimal solution to the linearized problem and this
solution is not optimal for the initial nonlinear problem. Note that this nonoptimality is
not caused by the linearization errors, but is a principal sequence of the substitution of the
original problem for the linearized one. A comparison of the optimal solutions to the initial
and linearized problems made for a numerical example is given in the paper. Other numerical
examples illustrate the method capacities.
2. Statement of the Problem

The spacecraft motion is described by the following:
ẋ f x, t g, 2.1
where x {r, v} is the spacecraft state vector, f fx, t {v, fv }, fv fv x, t is the
acceleration caused by the external forces,
g {0, α}. 2.2
In the considered case of the two-body problem,
μ
fv − r. 2.3
r3
Final orbit
qf
x(t) Transfer trajectory yfT = xT

ξ(t) ξT
qi Initial orbit
yiT t=T
t=0 yi0 = x0
dddd
Figure 1: Transfer between given orbits.
Let us assume for simplicity that the electric power of the propulsion system is constant. This
case takes place, for example, if a nuclear power source is used or if the transfer is performed
near a planet with nearly circular orbit. Then the performance index for the power-limited
thrust is
T
1
J α2 dt, 2.4
2 0
where α |α|. Minimal value of the functional 2.4 gives minimal propellant consumption
13.
Let qi , qf be 5-dimensional vectors of orbital elements defining the initial and final
orbits. The problem is to find transfer trajectory between the initial and final orbits in given
time T in which minimal value of the performance index 2.4 is reached.
3. Transfer between Given State and Given Orbit

The boundary values for the problem formulated in Section 2 may be written as
x0 yi0 , xT yfT , 3.1
where yi , yf are state vectors of the initial and final orbits. The case will be considered in this
section when vector yi0 is given and vector yfT is not given.
3.1. Transfer between Neighboring Orbits

First let us assume that the initial and final orbits are close to each other. Then the equation
of motion 2.1 can be linearized near the initial orbit as follows Figure 1:
ξ̇ Fξ g, 3.2
where
ξ ξt xt − yi t, 3.3

is state vector of the linearized motion and
∂f
F . 3.4
∂yi
Vector f and matrix F in 3.2 and 3.4 are calculated in the initial orbit, this is why matrix F
is a function of time even if vector f does not depend on time explicitly. The Hamiltonian for
3.2 and performance index 2.4 is 14
α2
H− pt Fξ Ptv α pt , 3.5
2
where p {pr , pv } is a vector of the costate variables corresponding to 3.2, pv is Lawden’s

primer vector, pt is costate variable corresponding to the additional equation ṫ 1 making
the system autonomous. Vector p satisfies the following costate variational equation:
t
∂H
ṗt − −pt F, 3.6
∂ξ
variable pt satisfies the following equation:
∂H
ṗt − −pò Ḟξ. 3.7
∂t
Let us consider sixth-order matrix Ψ Ψt which is a general solution to 3.6 with initial
value Ψ0 I. Matrix Ψ is costate transition matrix given by
Ψ ∂yi0 /∂yi . 3.8
Dividing matrix Ψ into two 6 × 3-dimensional submatrices as follows:
Ψ Ψr Ψv , 3.9
the costate variables may be represented as
p Ψt β, pv Ψtv β, 3.10
where β is a constant 6-dimensional vector. The Hamiltonian 3.5 reaches its maximum if
α pv Ψtv β. 3.11
Solution to 3.2 is given by the Cauchy formula

t
ξ ξt ξ 0 Φt, τgdτ, 3.12
0
where
∂yi t1
Φt1 , t2 3.13
∂yi t2
is the state transition matrix. Due to 3.1 and 3.3 ξ0 0 in 3.12. Using 2.2, 3.9, and
3.11 and relations
Φt, τ Φt, 0Φ0, τ Φt, 0Φ−1 τ, 0, Φ Φt, 0 Ψ−1 , 3.14
the solution 3.12 can be represented as
ξ ΦSβ, 3.15
where
t
S St Ψv Ψtv dt 3.16
0
is matrix of sixth order also see 9. As follows from 3.3, 3.11, and 3.15, in order to find
optimal thrust vector and the state vector of the transfer trajectory, it is sufficient to obtain
vector β. Nonspecified state vector yfT may be found using transversality condition which in
the considered case is
t
∂qf
pT pT σ, 3.17
∂yfT
where σ is an arbitrary constant 5-dimensional vector. Due to the closeness of the initial and
final orbits, the approximate equality ∂qf /∂yf ≈ U is fulfilled, where
∂qi
U Ut , 3.18
∂yi t
is a 5 × 6-dimensional matrix. Then the condition 3.17 takes the form
pT UtT σ. 3.19
Equations 3.10 and 3.19 and the last relation of 3.14 give
β UT ΦT t σ. 3.20
The closeness of the initial and final orbits makes it possible to use the following approximate
linear relation:
Δq qf − qi UT ξ T . 3.21
Equations 3.15, 3.20, and 3.21 give
Δq Wσ, 3.22
where
W UT ΦT ST UT ΦT t 3.23
is 5-dimensional matrix. Using 3.8, 3.9, 3.13, 3.14, 3.16, and 3.18, we obtain
∂qi ∂yiT ∂qi

UT ΦT U0 , 3.24
∂yiT ∂yi0 ∂yi0
T t t
∂qi ∂yi0 ∂yi0 ∂qi
UT ΦT ST UT ΦT t dt , 3.25
∂yi0 0 ∂vi ∂vi ∂yi0
where vi vi t is velocity vector in the initial orbit. 3.23, 3.25 give
T
W U0 ST Ut0 QQt dt, 3.26
0
where
∂qi
Q Qt − 3.27
∂vi
is a 5 × 3-dimensional matrix; as is seen from 3.18 and 3.27, matrix Q is a part of matrix U.
Using 3.20, 3.22, and 3.24, vector β can be found as follows:
β Ut0 W−1 Δq. 3.28
Now due to 3.3, 3.11, 3.15, and 3.28, the optimal thrust vector and the spacecraft state
vector are
α Ψtv Ut0 W−1 Δq, 3.29
xt yi t ΦSUt0 W−1 Δq. 3.30
Equations 3.1 and 3.30 give the state vector of entry into the final orbit
yfT yiT ΦT ST Ut0 Δq. 3.31

Final orbit
Transfer trajectory tj
qn = qf
qj
t2 q2
Reference orbits qn−1
tn = T
q1
t1
Initial orbit
q0 = qi
t0 = 0
Figure 2: Transfer between arbitrary orbits.
Substituting 3.29 into 2.4 and taking into account definitions 3.16, 3.26, the minimal
value of the performance index can be found as follows:
1
J Δqt W−1 Δq. 3.32
2
Equations 3.29, 3.30, and 3.32 give the solution to the problem.
Note that matrices U, W, Φ, Ψ are calculated in 9, 15, 16 in an explicit form which
makes the suggested method analytical.
3.2. Transfer between Arbitrary Orbits

Let us consider transfer between arbitrary orbits and divide the time interval T into n
subintervals defined by instants t0 t1 , . . . , tn−1 , tn T . Also we assume that n−1 intermediate
reference orbits between the initial and final orbits are specified somehow and q1 , . . . , qn−1 are
5-dimentional vectors of elements of the reference orbits see Figure 2. These vectors may be
given, for instance, in the following way:
j
qj qj qf − qi , j 1, . . . , n − 1 . 3.33
n
Let us define vectors

Δqj qj − qj−1 , j 1, . . . , n , 3.34
where q0 qi , qn qf here subscript “0” is the number of the reference orbit. Equations
3.21 and 3.34 give the following equality:
&
n
Δqj Δq. 3.35
j1
Let us divide the transfer trajectory into n arcs corresponding to the time subintervals and
assume that the jth arc begins in the j–1st reference orbit and ends in the jth one. Assuming
number n big enough to make the j–1st and jth reference orbits close to each other for all
j 1, . . . , n, the results of the Section 3.1 may be applied to each pair of the reference orbits.
Now the problem is to find the reference orbits providing optimality of the transfer. Due to
3.32 the performance index for the jth transfer arc is
1
Jj Δqtj W−1
j Δqj , j 1, . . . , n , 3.36
2
where similarly to 3.26, 3.27 the following definitions are used:
tj
∂qj−1
Wj Qj Qtj dt, Qj j 1, . . . , n , 3.37
tj−1 ∂vj−1
where 41 vj is velocity vector of the jth reference orbit. Performance index of the whole
problem is
&
n
J Jj . 3.38
j1
In order to find the transfer trajectory that gives minimum value for J, it is sufficient to find
intermediate reference orbits providing a minimum for 3.38. Thus, function 3.38 should
be minimized with respect to vectors Δqj , j 1, . . . , n taking into account 3.35. Let us
introduce the helping function
⎛ ⎞
&
n
L J − λt ⎝ Δqj − Δq⎠, 3.39
j1
where λ is a Lagrange multiplier. Necessary conditions of a minimum for the functional 3.38
are
t
∂L
W−1
j Δqj − λ 0 j 1, . . . , n . 3.40
∂Δqj
Thus,

Δqj Wj λ j 1, . . . , n , 3.41
and 3.35 and 3.41 give
λ W−1 Δq, 3.42

where
&
n
W Wj . 3.43
j1
Now new values of vectors Δqj can be found from 3.41 and 3.42 as follows:

Δqj Wj W−1 Δq j 1, . . . , n . 3.44
New reference orbits are defined by the new values of the orbital elements given by

qj 1 qj Δqj j 0, . . . , n − 1 . 3.45
Assuming optimal locations of the reference orbits found and applying 3.29 and 3.30 to
the jth arc of the transfer trajectory, we obtain optimal thrust vector and the trajectory state
vector in the time interval tj−1 ≤ t ≤ tj j 1, . . . , n as follows:
αt Ψtjv Utj W−1 Δq, 3.46
xt yj−1 t Φj Sj Utj0 W−1 Δq, 3.47
where
t
∂qj−1
Sj Sj t Ψjv Ψtjv dt, Uj , 3.48
tj−1 ∂yj−1
matrix W is given by 3.43, matrices Φ Φj t, 0, Ψjv Ψjv t are calculated in the j − 1st
reference orbit.
Multiplying 3.40 by Δqj and taking into account 3.35, 3.36, and 3.38, we obtain
1 t
J λ Δq. 3.49
2
Due to 3.42 and 3.49 minimal value of the performance index in the considered case is
given by 3.32 with matrix W defined by 3.43.
3.3. Calculation Procedure

Let y0j , y1j be the state vectors of the jth reference orbit at the beginning and at the end of the
jth time subinterval i.e., at times tj−1 , tj resp.. Then, the solution to the problem considered
here may be obtained by means of the following iterative calculation procedure.
1 n−1 intermediate reference orbits are specified somehow, for example, using 3.33.
The launch position in the initial orbit is specified in the case considered here i.e.,
state y00 yi 0 is given and the respective initial state vector of the transfer
trajectory is x0 y00 .
2 Vector y0j is calculated for j 1 using the following equation, that is, similar to
3.30:
y0j y1j−1 Φ1j S1j Utj0 W−1

j Δqj , 3.50
where S1j Sj tj , Uj0 Uj 0 matrix Ψjv Ψv tj is calculated in the j − 1st reference
orbit.
3 Step 2 is repeated for j 2, . . . , n − 1 and for j n 3.50 gives state vector yfT in the
final orbit.
4 New vectors Δqj are calculated using 3.44 and new reference orbits with elements
given by 3.45 are found.
5 Performance index is calculated using 3.32 and steps 2–4 are repeated until
decrement ΔJ of the performance index gets smaller than a given parameter ε > 0.
As soon as |ΔJ| < ε, the thrust vector α and the state vector x of the transfer
trajectory may be calculated at each time subinterval using 3.46 and 3.47.
The suggested method is approximate, although any desired accuracy can be reached by
means of selecting an appropriate amount n of subintervals. It can be shown that if n −→ ∞
then the solution converges to a limit which is an accurate solution to the linearized problem.
4. Other Types of the Transfer

4.1. Transfer between Given Orbit and Given State Vector
Now let us consider the case when the launch position can be selected in the initial orbit
arbitrarily and the position of the entry into the final orbit is given i.e., vector yi0 is not given
and vector yfT is given in 3.1. In this case the method described in Section 3 should be
applied in the backward direction with decreasing time, that is, vector y0n−1 of the start from
n − 1st reference orbit can be found for a given state vector yfT of the arrival to the final
orbit, and so forth, until vector yi0 is found. The equations necessary to solve the problem
considered here can be easily derived from the equations given in Section 3.
4.2. Transfer between Two Given Orbits

Let us assume that neither launch position in the initial orbit nor arrival position in the final
orbit are given and should be determined in an optimal way during solving the transfer
problem. This case may be solved using the suggested method in the following way.
A first guess for the launch position should be given somehow. This position defines
the vector yi0 and in the first iteration of the calculation procedure described in Section 3.3 the
final state vector yfT may be found. In the second iteration of the calculation procedure for
this state vector a new value of the vector yi0 may be found as described in Section 4.1, and
so forth, that is, odd iterations of the calculation procedure use the case described in Section 3
and even iterations use the case described in Section 4.1.
5. Partly Given Final Orbit

The suggested method can be also used in the case of partly given elements of the final orbit,
that is, if vector qf has dimension m < 5. For instance, only energy of the final orbit m 1
or semimajor axis and eccentricity m 2 may be given. In this case the method described
above is applied with m-dimensional vectors qi , qf , qj j 1, . . . , n − 1, m × 6-dimensional
matrix Uj , and m-order matrices Wj , Sj . Nongiven orbital elements are determined using
transversality condition 3.17 and the respective conditions for vectors qj .
6. Constraint Imposed on the Thrust Direction

Here the case when a constraint is imposed on the thrust direction is considered. This
constraint may be caused by specific features of the spacecraft design or of its attitude control
system. Let us assume the constraint given by
Bα 0, 6.1
where B Bx, t is a matrix of dimension 1 × 3 i.e., B is a row or 2 × 3 in this case the thrust
direction is given and the problem is to find optimal thrust value. In this case, as is shown
in 17, 18, the suggested optimization method is also applicable with matrices Wj , Sj and
vector αj from 41, 3.48, and 3.46 replaced by
tj t
Wj Qj PQtj dt, Sj Sj t Ψjv PΨtjv dt, αj PΨtv Utj W−1
j Δqj , 6.2
tj−1 tj−1
where third-order matrix
−1
P I − Bt BBt B, 6.3
is a projector onto the constraining set given by 6.1. Note that rank of matrix P is less
than 3. This is why matrices Wj given by 6.2 may be singular, while in the case of no
constraint on the thrust direction matrices Wj given by 41 are nonsingular in any interval
of integration 9. As is shown in 18 nonsingularity of all matrices Wj given by 6.2 is a
sufficient condition of feasibility of the transfer with constraint 6.1.
7. On Optimality of the Method

One of the necessary conditions of optimality of the thrust vector is constancy of the
Hamiltonian in the whole time interval of the transfer 14. It can be shown that the
Hamiltonian of the linearized problem given by 3.5 is constant in the solution given by the
suggested method, although the Hamiltonian of the original nonlinear problem calculated in
Table 1: Comparison of the solutions to the original and linearized problems.
Transfer duration Solution to the Solution to the

in periods of original problem linearized problem Difference, %
the initial orbit No. of No. of
J J
complete orbits complete orbits
20 9 1, 012 × 10−3 8 1, 053 × 10−3 4,1
40 18 4, 997 × 10−4 16 5, 193 × 10−4 3,9
100 46 1, 991 × 10−4 42 2, 070 × 10−4 4.0
200 93 0, 995 × 10−4 84 1, 034 × 10−4 3.9
the solution of the linear problem is not constant. Thus, the method described in the paper
gives the solution which is not optimal in the original problem.
A comparison of the solutions to the problem of the orbital transfer in the original
formulation and linearized one was performed. Transfer between coplanar circular orbits of
radii ai 1 and af 4 with gravitational parameter of the primary body μ 1 was considered.
Solution to the original problem was provided by Petukhov. He used the mathematical
developed by his method for solving two-point boundary value problem 19; optimal
solution to the orbital transfer problem was obtained by Petukhov by means of variation
of the arrival point in the final orbit. Results of the comparison are given in Table 1.
As is seen in Table 1, the difference between the values of the performance index for
the optimal solutions to the original and linearized problems is small and practically does not
depend on the transfer duration. Although in more complicated transfers the difference may
be bigger.
8. Illustrative Examples
This section demonstrates potentialities of the suggested method by means of examples of
transfers in the Earth’s sphere of influence. Orbits are given by the orbital elements
q {rπ , rα , i, Ω, ω}, 8.1
where rπ , rα are radii of perigee and apogee in thousands of kilometers Mm, i is the
orbital inclination, Ω is the longitude of the ascending node, ω is the argument of the
perigee. Angular elements are given in degrees. Direction of the thrust vector is given in the
examples by angles between the projection of the thrust vector onto the instantaneous orbital
plane angle ϕ and between the thrust vector and orbital plane angle Ψ. In all examples
given below the optimal positions of departure from the initial orbit and arrival to the final
orbit were obtained i.e., the classical case of orbital transfer described in Section 4.2 was
considered. The changes ΔV of the velocity by means of the thrust also are given in the
examples. These changes were calculated using numerical integration of the magnitude of
the thrust vector given by 3.46 or 6.2. The number of the time subintervals was taken
n 5000 in all examples.
y
80
x
−80 80
−80
z
80
x
−80 80
−80
Figure 3: Transfer between two elliptic orbits with high mutual inclination.
8.1. Transfer between Elliptical Orbits with High Mutual Inclination

Transfer between two orbits given by qi {7, 30, 50, 80, −60}, qf {40, 80, 80, −80, 70}
with T 400 hour is considered here. The transfer trajectory is shown in Figure 3 in
projections onto the equator plane xy and the polar plane xz. Dashed lines show node lines
of the initial and final orbits. The jet acceleration value divided by g 9.8066 m/s2 is shown
in Figure 4; Figure 5 shows angles ϕ, Ψ. Performance index and total ΔV for the transfer are
J 44.42 m2 s−3 , ΔV 10.05 km/s.
0.0015
0.001
α (g)
0.0005
0
0 100 200 300 400
Time of flight (hour)
Figure 4: Jet acceleration for transfer between two elliptic orbits with a high mutual inclination.
30
ϕ (deg)
−30
90
ψ (deg)
−90
0 100 200 300 400
Figure 5: Angles of the thrust direction transfer between two elliptic orbits with a high mutual inclination.
8.2. Transfer to an Orbit with Given Perigee and Apogee Radii

Transfer in time T 400 hrs to a partly given orbit, namely, to an orbit with given only perigee
and apogee radii, is considered here. Optimal transfer is planar in this case. Only perigee and
apogee radii of the initial orbit are specified, because the other initial orbital elements are not
important and may be taken equal to zero. Elements of the initial and final orbits are taken as
follows: qi {7, 20, 0, 0, 0}, qf {40, 80, 0, 0, 0}. Transfer orbit is shown in Figure 6; as
is seen in this figure optimal is coaxial collocation of the final orbit with respect to the initial
orbit. Figure 7 shows respective propulsion acceleration value divided by g, and angle ϕ is
y
60
x
−100 60
−60
Figure 6: Transfer trajectory to an orbit with given perigee and apogee radii.
0.0005
0.0004
0.0003
α (g)
0.0002
0.0001
0
0 100 200 300 400
Figure 7: Jet acceleration for transfer to orbit with given perigee and apogee distances.
shown in Figure 8 Ψ 0 in the planar transfer. Performance index and total ΔV are J 3.01
m2 s−3 , ΔV 2.82 km/s.
8.3. Transfer to an Orbit with Given Energy

A transfer to an orbit with a given energy, namely, to a hyperbolic orbit given only by C3
1 km2 /s2 , is considered here. Initial orbit with elements qi {7, 20, 0, 0, 0} is taken, the
transfer duration is T 1000 hrs. The transfer trajectory which is obviously planar in this
case and the jet acceleration are shown in Figures 7 and 8. The dot at the end of the spiral
trajectory in Figure 7 marks the entry into the hyperbolic orbit. Corresponding performance
index and total ΔV are J 4.85 m2 s−3 , ΔV 4.93 km/s. The thrust direction is tangential in
this case; this confirms nonoptimality of the solution to the linearized problem of the orbital
transfer, because in the original nonlinear problem the optimal thrust is not tangential see,
e.g., 20.
10
ϕ (deg)
0
−10
0 100 200 300 400
Figure 8: Angle ϕ for transfer to orbit with given perigee and apogee distances.
y
500
x
−1500 1500
−500
Figure 9: Transfer trajectory to the hyperbolic orbit with given energy.
8.4. Constrained Thrust Direction

A transfer between the orbits given by the elements qi {7, 20, 0, 0, 0}, qf
{50, 100, 60, 60, 60}, in a time T 400 hours, is considered here. The transfer trajectory is
shown in Figure 9. The performance index and total ΔV are J 10.83 m2 s−3 , ΔV 5.39 km/s.
Now let us consider the following constraint on the thrust direction: the thrust is
always orthogonal to the spacecraft position vector, that is, B rt in 6.1. The projector 6.3
in this case is P I−rrt /r 2 . The transfer trajectory for the constrained thrust direction visually
does not differ from the one for the unconstrained direction shown in Figure 9. Performance
index and total ΔV in the case of the constrained thrust direction are J 11.44 m2 s −3 , ΔV
5.52 km/s.
Acceleration value versus time for the unconstrained and the constrained thrust
direction is shown in Figure 10, the transfer trajectories are shown in Figure 11, the Jet
acceleration for the transfers are shown in Figure 12 and the angles of the thrust direction
for the transfers are shown in Figure 13.
9. Conclusion
The mathematical method for calculation of low-thrust orbital transfers presented in this
paper has two essential disadvantages, as follows.
1 The method is applicable to the power-limited thrust, while the existing thrusters
are close to ones with constant or given exhaust velocity.
2 The original nonlinear problem is replaced in the method by a linearized problem
solution to which it is not optimal for the original problem.
0.0006
0.0005
0.0004
α (g)
0.0003
0.0002
0.0001
0
0 200 400 600 800 1000
Figure 10: Jet acceleration for transfer to hyperbolic orbit with given energy.
y y
80 80
x x
−80 80 −80 80
z z
80 80
x x
−80 80 −80 80
−80 −80
a b
Figure 11: Transfer trajectories without a and with b constraint on the thrust direction.
0.0006
0.0005
0.0004
α (g)
0.0003
0.0002
0.0001
0
0 100 200 300 400
a
0.0006
0.0005
0.0004
α (g)
0.0003
0.0002
0.0001
0
0 100 200 300 400
b
Figure 12: Jet acceleration for transfer without a and with b constraint on the thrust direction.
These disadvantages are compensated by simplicity of the method and its analytical form at
each iteration, and also by the wide applicability of the method: it works well in the case of a
big difference between the initial and final orbits, for a very high number of orbits around the
attracting center; also it is applicable for different transfer types such as point-to-orbit, orbit-
to-point, and orbit-to-orbit transfers, in the cases of partly given final orbit and of a constraint
imposed on the thrust direction. Despite the fact that the method is based on linearization of
motion, any necessary accuracy of calculations may be reached by means of augmentation of
the number n of the reference orbits.
The suggested method may be used at early phases of the mission design when a high
optimization accuracy is not needed and at the same time massive calculations are necessary
for selection of a best mission scheme.
270 270
180 180
ϕ (deg)
ϕ (deg)
90 90
0 0
−90 −90
90 90
ψ (deg)
ψ (deg)
0 0
−90 −90
0 100 200 300 400 0 100 200 300 400
Time of flight (hour) Time of flight (hour)
a b
Figure 13: Angles of the thrust direction for transfer without a and with b constraint on the thrust
direction.
Nomenclature
r, v: Position and velocity vectors
t: Current time
t 0, t T : Initial and final instants of the transfer
I: Unit matrix
α: Jet acceleration vector thrust vector
α: |α|
μ: Gravitational parameter of the attracting center
ϕ: Angle between the projection of the thrust vector onto the orbital
plane and the velocity vector
ψ: Angle between the thrust vector and the orbital plane
Subscripts “0” and “T ”: Values of the parameters at the time instants 0 and T if another
meaning of the “0” subscript is not stipulated, superscript “t”
denotes transposition.
Acknowledgment
The authors are grateful to the Brazilian São Paulo Research Foundation FAPESP for fi-
nancial support of this study.
References
1 F. W. Gobetz, “Optimal variable-thrust transfer of a power-limited rocket between neighboring
circular orbits,” AIAA Journal, vol. 2, no. 2, pp. 339–343, 1964.
2 T. N. Edelbaum, “Optimum power-limited orbit transfer in strong gravity fields,” AIAA Journal, vol.
3, pp. 921–925, 1965.
3 J. P. Marec and N. X. Vinh, “Optimal low-thrust, limited power transfers between arbitrary elliptical
orbits,” Acta Astronautica, vol. 4, no. 5-6, pp. 511–540, 1977.
4 C. M. Haissig, K. D. Mease, and N. X. Vinh, “Minimum-fuel, power-limited transfers between
coplanar elliptical orbits,” Acta Astronautica, vol. 29, no. 1, pp. 1–15, 1993.
5 B. N. Kiforenko, “Optimal low-thrust orbital transfers in a central gravity field,” International Applied
Mechanics, vol. 41, no. 11, pp. 1211–1238, 2005.
6 S. Da Silva Fernandes and W. A. Golfetto, “Numerical computation of optimal low-thrust limited-
power trajectories—transfers between coplanar circular orbits,” Journal of the Brazilian Society of
Mechanical Sciences and Engineering, vol. 27, no. 2, pp. 178–185, 2005.
7 S. Da Silva Fernandes and W. A. Golfetto, “Numerical and analytical study of optimal low-thrust
limited-power transfers between close circular coplanar orbits,” Mathematical Problems in Engineering,
vol. 2007, Article ID 59372, 23 pages, 2007.
8 V. V. Beletsky and V. A. Egorov, “Interplanetary flights with constant output engines,” Cosmic Research,
vol. 2, no. 3, pp. 303–330, 1964.
9 A. A. Sukhanov, “Optimization of flights with low thrust,” Cosmic Research, vol. 37, no. 2, p. 191, 1999.
10 A. A. Sukhanov, “Optimization of low-thrust interplanetary transfers,” Cosmic Research, vol. 38, no. 6,
pp. 584–587, 2000.
11 A. A. Sukhanov and A. F. B. D. A. Prado, “A modification of the method of transporting trajectory,”
Cosmic Research, vol. 42, no. 1, pp. 103–108, 2004.
12 A. A. Sukhanov and A. F. B. D. A. Prado, “Optimization of low-thrust transfers in the three body
problem,” Cosmic Research, vol. 46, no. 5, pp. 413–424, 2008.
13 J. H. Irving, “Low thrust flight; variable exhaust velocity in gravitational fields,” in Space Technology,
H. S. Seifert, Ed., chapter 10, John Wiley and Sons, New York, NY, USA, 1959.
14 L. S. Pontryagin and R. V. Gamkrelidze, The Mathematical Theory of Optimal Processes, Gordon & Breach
Science Publishers, 1986.
15 B. Bakhshiyan and A. A. Sukhanov, “First and second isochronous derivatives in the two-body
problem,” Cosmic Research, vol. 16, no. 4, p. 391, 1978.
16 A. A. Sukhanov, “Isochronous derivatives in the two-body problem,” Cosmic Research, vol. 28, no. 2,
pp. 264–266, 1990.
17 A. A. Sukhanov and A. F. B. De A Prado, “Optimization of transfers under constraints on the thrust
direction: I,” Cosmic Research, vol. 45, no. 5, pp. 417–423, 2007.
18 A. A. Sukhanov and A. F. B. D. A. Prado, “Optimization of transfers under constraints on the thrust
direction: II,” Cosmic Research, vol. 46, no. 1, pp. 49–59, 2008.
19 V. G. Petukhov, “Optimization of interplanetary trajectories for spacecraft with ideally regulated
engines using the continuation method,” Cosmic Research, vol. 46, no. 3, pp. 219–232, 2008.
20 R. A. Jacobson and W. F. Powers, “Asymptotic solution to the problem of optimal low-thrust energy
increase,” AIAA Journal, vol. 10, no. 12, pp. 1679–1680, 1972.
doi:10.1155/2012/493507
Research Article
Four-Impulsive Rendezvous Maneuvers
for Spacecrafts in Circular Orbits Using
Genetic Algorithms
Denilson Paulo Souza dos Santos,1 Antônio Fernando Bertachini

de Almeida Prado,1 and Guido Colasurdo2
1
Division of Space Mechanics and Control INPE, CP 515, 12227-310 São José dos Campos SP, Brazil
2
Department of Aerospace and Mechanical Engineering, Università degli Studi di Roma “La Sapienza”,
Piazzale Aldo Moro 5, 00185 Roma, Italy
Correspondence should be addressed to

Denilson Paulo Souza dos Santos, denilson2011b@hotmail.com
Received 1 December 2011; Accepted 27 January 2012
Academic Editor: Maria Zanardi
Copyright q 2012 Denilson Paulo Souza dos Santos et al. This is an open access article distributed
Spacecraft maneuvers is a very important topic in aerospace engineering activities today. In a more
generic way, a spacecraft maneuver has the objective of transferring a spacecraft from one orbit to
another, taking into account some restrictions. In the present paper, the problem of rendezvous
is considered. In this type of problem, it is necessary to transfer a spacecraft from one orbit to
another, but with the extra constraint of meeting another spacecraft when reaching the final orbit.
In particular, the present paper aims to analyze rendezvous maneuvers between two coplanar
circular orbits, seeking to perform this transfer with lowest possible fuel consumption, assuming
that this problem is time-free and using four burns during the process. The assumption of four
burns is used to represent a constraint posed by a real mission. Then, a genetic algorithm is used
to solve the problem. After that, a study is made for a maneuver that will make a spacecraft
to encounter a planet, in order to make a close approach that will change its energy. Several
simulations are presented.
1. Introduction
This paper aims to analyze optimal rendezvous maneuvers between two spacecrafts that are
initially in circular coplanar orbits around the Earth. The main goal is to perform this transfer
having the fuel consumption as a penalty function, so the minimization of this quantity is
searched during the process of finding the solutions. The approach used here is to assume
that the problem is time-free, which means that the time of the transfer is not important. The
control assumed to perform this task is an engine that can deliver four burns. This assumption
is used to represent a common constraint posed by real missions. In the present paper, we are
considering a generic problem, not a specific mission, but this type of constraint appears very
often in space activities.
Then, a genetic algorithm is used in order to solve the problem. This type of
approach represents a new alternative to solve this problem and can be used for comparisons
with results obtained by standard procedures available in the literature, as shown in 1–
28. Preliminary studies showed that, in some situations, this algorithm can be faster in
convergence and more accurate, while in some others, it is slower and presents less accuracy.
A detailed comparison still has to be made to evaluate under which circumstances this
algorithm can be more efficient. In any case, several kinds of missions can use the benefits
of the techniques based on the genetic algorithm showed in this work. The main types are
transference with free time to change the orbit of the space vehicle without restrictions in the
time required by the execution of the maneuver, “rendezvous” when one desires that the
space vehicle stands alongside another spacecraft, “flyby” a mission to intercept another
body, however without the objective to remain next to it, “swing-by” a close approach
to a celestial body to gain or lose energy, and so forth. But, in the present paper, only the
rendezvous maneuver is considered.
2. Description of the Problem

The problem of orbital maneuvers has been studied in several published papers. Some of
them are shown in 1–28. The different approaches to solve this problem can be appreciated
in those references. Some authors assumed that a low magnitude force is applied to the
spacecraft during a finite time. This is the so-called continuous thrust approach. References
7, 10, 17, 28 have some details on this topic. As an alternative approach, the idea of an
impulsive maneuver is also studied. In this situation, a high magnitude force is applied
during a time that can be considered negligible. References 3, 5, 8, 27 used this important
approach. More recently, two more ideas appeared in the literature to perform orbital
maneuvers. The first one is the use of a close approach with a celestial body to change the
orbit of a spacecraft. It is the swing-by maneuver. References that used this approach are
2, 13. The second recent approach is the gravitational capture, where the force generated
by the perturbation of a third body 14 can be used to decrease the fuel expenditure of a
space maneuver. References 11, 12 have some details of this idea. Some publications cover
all those topics in more details, like 6, 9, 15. Studies more related to the research shown in
the present paper are the ones considering the Lambert’s problem 1, 16, the rendezvous
maneuver 20–26, and genetic algorithms itself 18, 19.
In the present research, in order to solve the transfer proposed here, the Lambert’s
problem is used, in the way described below. The Lambert’s problem can be formulated
as follows: “Find an unperturbed orbit, under the mathematical model given by a law that
works with the inverse square of the distance Newtonian formulation, that connects two
given points P1 and P2 , with the transfer time Δt specified.” In the literature, several
researchers have solved this problem by using distinct formulations. Reference 1 shows
several of them. In this way, the parameters of the transfer orbit can be defined by
1 ν1 is the true anomaly of the departure point P1 on the initial orbit. ν1 ∈ 0, 2π,
2 Δν is the angular length of the transfer. Δν ∈ 0, 2π,

P2 P1
c
ν
F1
F2
Figure 1: Instantaneous scenario of the problem.
3 at is the semimajor axis of the transfer orbit. Note that, for each pair of departure
and arrival points, a minimum value amin exists for at . Two transfer orbits can be
found for the same value of at , depending on the sense of the transfer.
The parameter at is usually replaced by a different parameter y. The advantage of this

substitution is that the new variable has values between 0 and 1. The relationship is shown
below 19
amin
at . 2.1
4y 1 − y
The parameters ν1 and Δν determine the position of the points P1 and P2 that can be
related to the radius vectors r1 and r2 . Any permitted value of the parameter y determines
univocally one transfer orbit. These parameters are, from the point of view of the genetic
optimizer, the genes of the members of the population.
The genetic algorithm searches for the best solution among a number of possible
solutions, represented by vectors in the solution space. To find a solution is to look for some
extreme value minimum or maximum in the solution space. The fitness of each individual
is represented by the total velocity impulse ΔV required to perform the orbital transfer. The
total impulse is given by the sum of the single impulses ΔVi provided in each thrust point in
order to pass from an orbital arc to the following one. It corresponds to the velocity difference
at the relevant thrust point.
The positions of the thrust points and the parameters of the transfer orbit are obtained
using as input the three genes, that is, the parameters previously chosen. The velocities at
the thrust points, before and after firing the engine, are easily computed, and it provides the
total velocity impulse, which is the measurement of the individual fitness. The evolutionary
process will select individuals with the genes corresponding to the optimal maneuver.
Figure 1 shows an instantaneous scenario of the problem.
Note that ν1 is the true anomaly of the point P1 on the initial orbit; ν2 is the true
anomaly of the point P2 on the final orbit; Δν is the angular length of the transfer; the
orientation of the transfer orbit is defined by the angle ω between its axis and the axis of
the initial orbit; c is the distance between P1 and P2 2.4; Fi are the focus of the ellipse.
Random population
of chromosomes
Fitness of each
individual
Display
results
Select next
generation
Crossover
Mutation
Epidemic
Population
Figure 2: The genetic algorithm.
2.1. The Genetic Algorithm

The procedure starts with a random population of up to 800 individuals. The initial
population is generated randomly, and consider its characteristics distance and angles
according to the constrains of each variable. The vectors x are assembled according to the
allowed boundary condition. Then, the fitness of each individual is verified, following the
criteria of the objective function, which is to minimize the fuel consumption measured by
the ΔV found by solving the Lambert’s problem. So, the best individuals are selected to go
to the next generation, parents, and children. The procedure of crossover is then applied, as
well as a mutation to insert diversity in the population Figure 2.
The random variables used for the implementation of the algorithm are x
Δθ 1 , Δθ2 , R1 , R2 , y1 , y3 . Those symbols have the meaning that θi νi − ω is the true
anomaly of the P i points that determine the transfer orbit, as shown in Figure 3; Ri determines
the radius vector position in each thrust; the y 2.7 are the angles between F1 ,
P1 , P2 see
Figure 3 again.
Eventually, there are epidemics, with the goal of inserting diversity and reducing the
elitism. After that, a new population is created, and the procedure is started again, finishing
after n attempts. The block diagram of the genetic algorithm Figure 2 shows the procedures
followed to solve the problem. More details of the genetic algorithm can be obtained in 18,
19.
2.2. Selecting the Next Generation and Performing

the Crossover and the Epidemic Process
The selection of the new generation is made after the analysis of each individual by
measuring its objective function Fitness. The ones with better values for this measurement
are selected to undergo a process of crossing or reproduction crossover, where parents are
selected, and the children of this intersection are raised Figure 2. When the population is too
uniform, measured by the values of their objective functions, part of the population suffers
an epidemic process, where many individuals are killed and replaced by others using again a
random process, to insert diversity in the population and to prevent premature convergence
to local optimal values. The crossover starts by separating the chromosomes of the parents
in two parts. After this separation, the first part of the parent 1 is combined with the second
part of the parent 2, and the first part of the parent 2 is combined with the second part of the
parent 1. In this way, a second generation is created. See 18, 19 for more details.
2.3. Chromosome
The chromosome representation is vital for a genetic algorithm GA, because it is the
way that we translate the information from the problem to a format that can be handled
by the computer. This representation is completely arbitrary, so it varies according to the
choice made by each developer, without any kind of obligation to adopt any representation
available in the literature. This is a very important point to emphasize. The vast majority
of researchers use the binary representation for this problem because it is the simplest
one. In fact, many people, when they imagine a GA, quickly make an association with
binary chromosomes used to facilitate the crossing. However, other formulations using real
chromosomes, modifying the way of performing crossover, get satisfactory results 18. In
this paper, each gene is chosen to be a real number between 0 and 1, being generated in a
binary form and then converted in a real number. The value of the corresponding parameter
is Xi Ximin ui Ximax − Ximin , where Ximin and Ximax are the minimum and maximum values
of those variables, which means that they are the boundary conditions. The main reason to
use the binary approach is to validate this usual approach, in GA problems, in the particular
type of problem considered here. References 18, 19, that studied this same problem using
GA, used different approaches, so the validation of the binary approach was considered
important.
2.4. Objective Function

Most of the selection techniques used in this procedure require comparisons of the fitness to
decide which solutions should be propagated to the next generation. Normally, the fitness
has a direct relation with the value of the objective function, according to the rule that
better values of the objective function generate higher values of the fitness parameter. When
the genetic algorithm calls the objective function, it transfers an array of parameters that
specify the selected solution. This selection parameter must not be changed in any way
by the objective function. Genetic algorithms are based on biological evolution, and they
are able to identify and explore environmental factors to converge to optimal solutions, or
approximately optimal global levels. Then, the fitness of each individual can be computed by
using the five data that define the problem a1 , e1 , a2 , e2 , Δω, the first one being unit because
of the normalization of the variables and the three genes ν1 , Δν, y that characterize the
individual. Then, we can obtain several important parameters 15. The true anomaly of the
arrival point is given by
νi νi−1 Δν. 2.2

P1
P2
c y1
r2 y2
ν r1
c2
c1
F1
F2
Figure 3: Geometry of the problem and the angles involved in the problem.
The radii of the departure and arrival point are given by

a1 1 − e12
r1 ,
1 e1 cos ν1
2.3
a2 1 − e22
r2 .
1 e2 cos ν2
The distance between P1 and P2 is

c r12 r22 − 2r1 r2 cos Δν. 2.4
The semimajor axis of the transfer orbit is
r1 r2 c
amin . 2.5
4
The distances c1 and c2 of P1 and P2 from the vacant focus F2 can be specified by the equations
ci 2a − ri . 2.6
Figure 3 shows a description of several important variables.

The angles can be calculated by y y1 y2 ,

r12 − r22 c2
y arcos ,
2r1 c
2 2.7
c1 − c22 c2
y1 arcos .
2c1 c
The eccentricity of the transfer orbit is given by

c12 r12 − 2c1 r1 cos γ2

et . 2.8
2at
The true anomaly θ1 of the P1 on the transfer orbit is

at 1 − et2
r1 ,
1 e cos θ1
t 2.9
at 1 − et2 − r1
θ1 arcos .
r1 et
The argument of perigee for the transfer orbit is
ω ν1 − θ1 , 2.10
which is the angle between the perigees of the transfer and the initial orbits.
Now that the geometry of the maneuver has been shown, it is possible to calculate
the radial and the tangential components of the spacecraft velocity before and after both
impulses, what permits the computation of the total ΔV , which has been assumed to be the
measurement of the individual fitness.
2.5. Normalization
Nondimensional variables are used in the procedure. They are shown below
r:
r ,
a:1
v: 2.11
v
.
μ/:
a1
The distance and velocity units for the normalized variables are the semimajor axis of
the initial orbit and the velocity

on a circular orbit with the same energy as the initial one. So,
the reference time is Δt a:31 /μ.
3. Numerical Solutions
Several maneuvers were simulated with the procedure developed here, using the genetic
algorithm. Then, the equivalent Hohmann maneuvers were calculated to provide a level of
comparison. The idea is not to find a transfer that has a smaller total ΔV , when compared
to the Hohmann transfers, but to try to minimize the difference in costs, assuming that the
engine of the spacecraft has a limitation that does not allow two impulsive maneuvers to
be performed. In theory, for the cases simulated here, the two impulses maneuvers always
have a lower consumption. So, the idea is to find the best maneuver that has four impulses,
in order to compare with other works 18, 19. The number of impulses is a parameter that
can be modified in the input data of the algorithm to be useful for other applications. The
results shown in the present paper always consider a rendezvous maneuver between two
spacecrafts, where the radius of the orbit of the first spacecraft is ro 1, and several values
were used for the radius of the spacecraft that is in the final orbit see Table 1. The genetic
Table 1: Rendezvous between coplanar circular orbits showing the values of the ΔV for each burn and the
total expenditure ΔVT . The results for the Hohmann are included for comparison.
Simulation Cost using the genetic algorithm Hohmann transfer

n◦ 64 62 ΔV T − ΔV HT
ro 1 ΔV1 ΔV2 ΔV3 ΔV4 ΔV T i1 ΔV i ΔV HT i1 ΔV i
1 rf 1.2 0.104455 0.118925 0.159234 0.154917 0.537531 0.087000 0.450500
2 rf 1.5 0.044987 0.118996 0.165893 0.189594 0.519469 0.181600 0.337900
3 rf 1.6 0.028621 0.120939 0.163349 0.204023 0.516932 0.206600 0.310300
4 rf 1.8 0.000000 0.131481 0.169234 0.225721 0.526435 0.249300 0.277100
5 rf 1.9 0.000000 0.127047 0.178111 0.233180 0.538338 0.267700 0.270600
6 rf 2 0.008136 0.117267 0.185660 0.242748 0.553811 0.284500 0.269300
7 rf 2.5 0.134621 0.000000 0.222416 0.298130 0.655167 0.349600 0.205300
8 rf 3 0.171286 0.000000 0.242839 0.369209 0.783334 0.393800 0.271100
9 rf 5 0.172864 0.310322 0.000000 0.423150 0.906336 0.480000 0.426300
10 rf 10 0.004673 0.000000 0.372418 0.311566 0.688657 0.499300 0.189400
Comparison GA versus Hohmann

1.12
1
0.8
0.6
V
0.4
0.2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Radius of target orbit (rf)
VGA
Hohmann
Figure 4: Comparison between the ΔVGA and ΔVHohmann .
algorithm provided satisfactory solutions, when compared with the solutions of the literature
18, as shown in Table 1. The population is composed by 800 individuals, and up to 400
generations of individuals were used.
The results indicated that the maneuvers using the GA with 4 impulses do not provide
savings over the Hohmann transfer for all cases simulated see Figure 4 and Table 1, as
expected and explained before, but it minimizes the difference in costs for the assumed four
impulsive maneuvers. Figure 4 shows all the details for this comparison.
Figures 5, 6, 7, 8, and 9, as well as Table 1, show a series of maneuvers. In general, an
impulse is applied in the initial orbit ΔV1 , generating the first elliptical transfer arc, and
then, according to the procedure, the second impulse is applied ΔV2 , leading to another
elliptical transfer orbit. The third point of burn will happen ΔV3 to put the spacecraft in
the last transfer arc, and, finally, the last impulse ΔV4 is applied to locate the vehicle in
the desired orbit. The total consumption is the sum of all the intermediate impulses, and
it is named ΔVT Table 1. This total consumption serves as an index of measurement and
comparison between the methods. In other words, the information of the extra cost is due to
Four-burn orbit transfers—genetic algorithm

2
1.5 V4
1
V2
0.5
R(θ)(u.n.) 0 V1
−0.5 V3
−1
−1.5
−2
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
R(θ)(u.n.)
V1 = 0.008094 V4 = 0.24276
V2 = 0.117318 VT = 0.553811
V3 = 0.185638
Figure 5: Four-burn orbit transfers for simulation 6.
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 25 50 75 100 125 150 175 200 225 250 275
Number of objective function evaluations
θ1 Y1
θ2 Y2
R1 Best fitness
R2
Figure 6: The variables of the problem using the method of genetic algorithm and the best fitness for
simulation 6.
the fact that a two-impulse maneuver is not possible and a detailed vision of the best four-
impulse strategy is generated by the GA.
Δθ1 , Δθ2 , R1 , R2 , y1 , y3 see Figures 5 and 7. In
The variables of the problem are x
each new generation of the population, the individuals are approaching the values suggested
by the algorithm, converging to a solution of the problem. The best fitness values of the
parameters show the convergence to the optimal value. Table 2 shows a detailed view of
the maneuver, explaining all the intermediate Keplerian orbits obtained.
Table 2: The Keplerian elements of the intermediate orbits for the case where rf 2.
Orbit a E w Vc
0 1 0 0 1
1 1.067591 0.247661 4.720904 0.967826586
2 1.553574 0.460257 5.401124 0.802294893
3 1.553574 0.460257 5.401124 0.802294893
4 2 0 0.707106781

3
1 V2
R(θ)(u.n.)
0 V3 V1
V4
−1
−2
−3
−3 −2 −1 0 1 2 3
R(θ)(u.n.)
V 1 = 0.171286 V 4 = 0.369209
V2 = 0 V T = 0.783334
V 3 = 0.242839
Simulation 8 and Figure 7 show some new results that confirm that the use of
the procedure with four impulses provides results with higher consumption than the bi-
impulsive maneuver Table 1, but that minimizes the four-impulsive burn technique.
This study can also be applied to find orbital maneuvers that search for the minimum
fuel consumption for a spacecraft that leaves one celestial body and goes back to this same
body Figures 9 and 10. This question is of great importance for missions whose objective
is to shift the position of the satellite in a given orbit, without changing the other orbital
elements. Prado and Broucke 1 also studied this problem using the Lambert method, under
different circumstances.
3.1. The Swing-By Maneuvers

The next step is to use the algorithm developed here to study a maneuver that will make a
spacecraft to encounter a planet, in order to make a close approach that will change its energy.
This problem can be seen as a rendezvous problem, where the second spacecraft, the one to be
reached, is a planet and not a space vehicle. Using this approach, a transfer maneuver using
Best individual
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 25 50 75 100 125 150 175 200 225 250
Number of objective function evaluations
θ1 Y1
θ2 Y2
R1 Best fitness
R2
Figure 8: The variables of the problem using the method of genetic algorithm and the best fitness for
simulation 8.
1
0.8
0.6
V3
0.4
R(θ)(u.n.)
0.2
V2
0 V1
−0.2
−0.4
−0.6 V4
−0.8
−1
−1.5 −1 −0.5 0 0.5 1 1.5

R(θ)(u.n.)
V 1 = 0.104455 V 4 = 0.154917
V 2 = 0.118925 V T = 0.537531
V 3 = 0.159234
an impulsive engine with four burns is followed by a gravity-assisted maneuver to send the
spacecraft further in the solar system. This technique will reduce the cost of an interplanetary
mission. This is a standard procedure in orbital maneuvers, and a more detailed description
is available in references 2, 9. In this case, the system consists of three bodies:
1 the body M1 , with finite mass, situated in the center of mass of the Cartesian system
of reference;

1
V4
0.8
V2
0.6
0.4
R(θ)(u.n.)
0.2
0 V3 V1
−0.2
−0.4
−0.6
−0.8
−1
−1 −0.5 0 0.5 1
R(θ)(u.n.)
V 1 = 0.130534 V 4 = 0.130537
V 2 = 0.143611 V T = 0.54829
V 3 = 0.143608
Figure 10: Transfer maneuver from one orbit back to the same orbit.
2 M2 , a smaller body, that can be a planet or a satellite of M1 , in a Keplerian orbit

around M1 ;
3 a body M3 , a space vehicle with infinitesimal mass, traveling in a Keplerian orbit
around M1 , when it passes close to M2 .
This close approach changes the orbit of M3 and, by the hypothesis assumed for
the problem, it is considered that the orbits of M1 and M2 do not change. Using the
“patched conics” approximation, the equations that quantify those changes are available in
the literature 9.
The standard maneuver can be identified by the following three parameters
Figure 11:
i |V∞ |, the magnitude of the velocity of the spacecraft with respect to M2 when
approaching the celestial body;
ii rp , the distance between the spacecraft and the celestial body during the closest
approach;
iii ΨA , the angle the approach.
Having those variables, it is possible to obtain δ, half of the total deflection angle, by
using the equation 2

1
δ arsin . 3.1
1 rp V∞2 /μ2
Note that V2 is the velocity of the celestial body with respect to the main body and
VP is the velocity of the smaller mass when passing by the periapsis. A complete description
→
−+
V∞
δ
δ
VP δ
δ
rP
→
−
V2
ψA
M1 M2
x →
−−
V∞
Figure 11: The parameters of the swing-by maneuver.
of this maneuver and the derivation of the equations can be found in Prado 9. The final
equations are reproduced below
ΔE −2V2 V∞ sinδ sinΨA ,

−2V2 V∞ sinδ sinΨA 3.2
ΔC ,
ω2
where ω2 is the angular velocity of the motion of the primaries, ΔE is the variation of energy,
ΔC is the variation of the angular momentum, and ΔV is the variation of the magnitude of
the velocity due to the swing-by. For the ΔV , we have the equation
% % % %
% % % %
ΔV %ΔV % 2%V∞ % sinδ 2V∞ sinδ. 3.3
The gravity-assisted maneuver swing-by can provide a considerable change of the

velocity and energy of the spacecraft, reducing the costs of the mission. During this approach,
the spacecraft will be transferred to another orbit of interest of the mission. The dynamics
used to solve this problem is the traditional model given by the “Patched Conics,” so it
is assumed that all three bodies involved are points of mass and do not suffer external
disturbances. The variations given by the swing-by, in terms of velocity variation ΔV and
energy variation ΔE, can now be obtained.
Figure 12 shows the maneuver obtained by the genetic algorithm. The spacecraft
comes from an initial orbit with radius ro 1 u.a., which represents the position of the Earth’s
heliocentric system, in astronomical units. It means that the spacecraft starts from the Earth.
Then, it performs a maneuver with 4 impulses, using three elliptic intermediate transfer
orbits, and finally it arrives in an orbit with rf 5.202803 u.a. Jupiter. At this moment,
it realizes a maneuver of Swing-by with the planet Jupiter. Note that the gain in velocity
was ΔVSB 1.104368 and the gain in energy was ΔE 2.017347. During this approach, the
space vehicle place itself in another orbit of the interest of the mission. In this mission, the
participation of the GA is to find the best procedure to make the spacecrafts reach the planet
Four-burn orbit transfers—Earth-Jupiter-genetic algorithm.

5
4
3
2
V3
V2
R(θ)(u.n.)
1
0 V1
−1
Swing-by Jupiter
−2
VSB = 1.104368
−3 ψ = 259.4993
E = 2.017347
−4 V4
−5
−6 −4 −2 0 2 4 6
R(θ)(u.n.)
V 1 = 0.256974 V 4 = 0.394731
V 2 = 0.281067 V T = 0.932772
V3 = 0
Figure 12: Simulation of a maneuvers leaving the Earth ro 1 and reaching the orbit of Jupiter rf
5.202803, performing a swing-by maneuver on the planet to gain velocity and energy.
Jupiter. From this point, standard procedures of interplanetary trajectories can complete the
mission.
4. Conclusion
Based on the analysis of the results obtained, the genetic algorithm implemented here
shows that this technique brings good results for the proposed four impulsive rendezvous
maneuvers, when compared with the ones obtained by the traditional impulsive methods. It
means that it can be used in real cases, specially when a bi-impulsive transfer is not possible
due to the limitations of the engine of the spacecfraft. The procedure is also effective in
maneuvering the spacecraft from one body back to the same body, that is, making it leaving
and returning to the same orbit.
The results indicate that the maneuver using the genetics algorithm with four impulses
does not provide better fuel consumption in any case simulated, since the bi-impulsive
maneuver is better in this situation, but the method proves to be efficient in minimizing
the four impulsive maneuvers. It is necessary to take into account that, in many cases, the
limitations of the propellers of the spacecraft require that the maneuver has to be performed
using several impulses, passing through intermediate orbits to reach the target.
Then, we studied a maneuver where the goal is to send a spacecraft to encounter the
planet Jupiter to make a swing-by maneuver. The algorithm worked well in finding a good
solution for this problem.
In general, the proposed technique can be used when a rendezvous maneuver is
required between two given circular orbits for a spacecraft that has an engine that requires
the application of four impulses.
In the future, it is possible to apply this technique in three dimensions, in maneuvers

that requires more impulses, and also in maneuvers to avoid collisions between a spacecraft
and asteroids.
Acknowledgments
This work was accomplished with the support of São Paulo State Science Foundation
FAPESP under Contract 2009/16517-7 and National Institute for Space Research, INPE,
Brazil.
References
1 A. F. B. A. Prado and R. A. Broucke, “Study of hénon’s orbit transfer problem using the lambert
algorithm,” AIAA Journal of Guidance, Control, and Dynamics, vol. 17, no. 5, pp. 1075–1081, 1993.
2 D. P. S. Santos, A. F. B. A. P. Prado, and E. M. Rocco, “The use of consecutive collision orbits to obtain
swing-by maneuvers,” in Proceedings of the 56th International Astronautical Congress, Fukuoka, Japan,
October 2005.
3 F. W. Gobetz and J. R. Doll, “A survey of impulsive trajectories,” AIAA Journal, vol. 7, pp. 801–834,
1969.
4 R. H. Goddard, “A method of reaching extreme altitudes,” Smithsonian Institute Public Miscelanea
Collect, vol. 71, no. 2, pp. 809–811, 1920.
5 W. Hohmann, Die Erreichbarkeit Der Himmelskorper, Oldenbourg, Munique, 1925.
6 J. P. Marec, Optimal Space Trajectories, Elsevier, New York, NY, USA, 1979.
7 D. P. S. Santos, L. Casalino, G. Colasurdo, and A. F. B. A. P. Prado, “Optimal trajectories using gravity
assisted maneuver and solar electric propulsion SEP towards near-earth-objects,” in Proceedings of
the 4th WSEAS International Conference on Applied and Theoretical Mechanics (Mechanics ’08), pp. 62–68,
Cairo, Egypt, December 2008.
8 E. M. Rocco, A. F. B. A. Prado, and M. L. O. Souza, “Bi-impulsive orbital transfers between non-
coplanar orbits with time limit,” in Applied Mechanics in the Americas, D. Pamplona, C. Steele, H. I.
Weber, P. B. Gonçalves, I. Jasiuk, and L. Bevilacqua, Eds., vol. 6, pp. 259–262, 1999.
9 A. F. B. A. Prado, Optimal transfer and swing-by orbits in the two- and three- body problems, Ph.D. thesis,
Department of Aerospace Engineering and Engineering Mechanics—University of Texas, 1993.
11 A. F. B. A. Prado, “Numerical and analytical study of the gravitational capture in the bicircular
problem,” Advances in Space Research, vol. 36, no. 3, pp. 578–584, 2005.
12 A. F. B. A. Prado, “Numerical study and analytic estimation of forces acting in ballistic gravitational
capture,” Journal of Guidance, Control, and Dynamics, vol. 25, no. 2, pp. 368–375, 2002.
13 A. F. B. A. Prado and R. Broucke, “Transfer orbits in restricted problem,” Journal of Guidance Control
and Dynamics, vol. 18, no. 3, pp. 593–598, 1995.
14 A. F. B. A. Prado, “Third-body perturbation in orbits around natural satellites,” Journal of Guidance,
Control, and Dynamics, vol. 26, no. 1, pp. 33–40, 2003.
15 V. A. Chobotov, Orbital Motion, American Institute of Aeronautics and Astronautics, 2nd edition, 1996.
16 J. E. Prussing, “Geometrical interpretation of the angles α and β in lambert’s problem,” Journal of
Guidance, Control, and Dynamics, vol. 2, no. 5, pp. 442–443, 1979.
17 J. E. Prussing, “Equation for optimal power-limited spacecraft trajectories,” Journal of Guidance,
18 M. Rosa Sentinella and L. Casalino, “Genetic algorithm and indirect method coupling for low-thrust.
Trajectory optimization,” AIAA 064468, 2006.
19 F. Cacciatore and C. Toglia, “Optimization of orbital trajectories using genetic algorithms,” Journal of
Aerospace Engineering, Sciences and Applications, vol. 1, no. 1, pp. 58–69, 2008.
20 J. E. Prussing and J. H. Chiu, “Optimal multiple-impulse time-fixed rendezvous between circular
orbits,” Journal of Guidance, Control, and Dynamics, vol. 9, no. 1, pp. 17–22, 1986.
21 B. H. Billik and H. L. Roth, “Studies relative to rendezvous between circular orbits,” Astronautica Acta,
vol. 13, 1967.
22 J. E. Prussing, “Optimal four-impulse fixed-time rendezvous in the vicinity of a circular orbit,” AIAA
Journal, vol. 7, no. 5, pp. 928–935, 1969.
23 J. E. Prussing, “Optimal two-and three-impulse fixed-time rendezvous in the vicinity of a circular
orbit,” AIAA Journal, vol. 8, no. 7, pp. 1221–1228, 1970.
24 H. Shen and P. Tsiotras, “Optimal two-impulse rendezvous using multiple-revolution lambert
solutions,” Journal of Guidance, Control, and Dynamics, vol. 26, no. 1, pp. 50–61, 2003.
25 H. Shen and P. Tsiotras, “Optimal two-impulse rendezvous between two circular orbits using multiple
revolution Lambert’s solutions,” Proceedings of the AIAA Guidance, Navigation, and Control Conference,
1999.
26 Y. Z. Luo, G. J. Tang, and H. Y. Li, “Optimization of multiple-impulseminimum-time rendezvous with
impulse constraints using a hybrid genetic algorithm,” Aerospace Science and Technology, vol. 10, no. 6,
pp. 534–540, 2006.
27 E. M. Rocco, A. F. B. A. Prado, M. L. O. Souza, and J. E. Baldo, “Optimal bi-impulsive non-coplanar
maneuvers using hyperbolic orbital transfer with time constraint,” Journal of Aerospace Engineering,
Sciences and Applications, vol. 1, no. 2, pp. 43–51, 2008.
28 D. P. S. Santos, A. F. B. A. Prado, L. Casalino, and G. Colasurdo, “Optimal Trajectories towards near-
earth-objects using solar electric propulsion SEP and gravity assisted maneuver,” Journal of Aerospace
Engineering, Sciences and Applications, vol. 1, no. 2, pp. 51–64, 2008.
doi:10.1155/2012/659396
Research Article
Semianalytic Integration of High-Altitude Orbits
under Lunisolar Effects
Martin Lara,1 Juan F. San Juan,2 and Luis M. López3

1
C/Columnas de Hércules 1, ES-11100 San Fernando, Spain
2
Departamento de Matemáticas y Computación, Universidad de La Rioja, 26004 Logroño, Spain
3
Departamento de Ingenierı́a Mecánica, Universidad de La Rioja, 26004 Logroño, Spain
Correspondence should be addressed to Martin Lara, mlara0@gmail.com
Academic Editor: Josep Masdemont
Copyright q 2012 Martin Lara et al. This is an open access article distributed under the Creative
Commons Attribution License, which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.
The long-term effect of lunisolar perturbations on high-altitude orbits is studied after a double
averaging procedure that removes both the mean anomaly of the satellite and that of the moon.
Lunisolar effects acting on high-altitude orbits are comparable in magnitude to the Earth’s
oblateness perturbation. Hence, their accurate modeling does not allow for the usual truncation
of the expansion of the third-body disturbing function up to the second degree. Using canonical
perturbation theory, the averaging is carried out up to the order where second-order terms in
the Earth oblateness coefficient are apparent. This truncation order forces to take into account
up to the fifth degree in the expansion of the lunar disturbing function. The small values of the
moon’s orbital eccentricity and inclination with respect to the ecliptic allow for some simplification.
Nevertheless, as far as the averaging is carried out in closed form of the satellite’s orbit eccentricity,
it is not restricted to low-eccentricity orbits.
1. Introduction
In an increasingly saturated space about the Earth, aerospace engineers confront the
mathematical problem of accurately predicting the position of Earth’s artificial satellites. This
is required not only for the correct operation of satellites but also for preserving the integrity
of space assets. Thus, operational satellites are threatened by the not so remote possibility of a
collision with a defunct satellite 1 but most probably by the impact with other uncontrolled
man-made space objects as spent rocket stages or collision fragments—all of them commonly
called space debris.
Precise predictions require the integration of complete force models including both
gravitational and nongravitational effects, like a high-degree and order geopotential,
ephemerides-based lunisolar perturbations, drag, solar radiation pressure including eclipses,

and so forth see 2, 3, for instance. The most accurate integration is expected from numeri-
cal methods, although precision ephemeris can also be obtained by means of semianalytical
integration 4. In fact, both approaches, numerical and semi-analytical, do not need to enter
a competition. Thus, for instance, while semi-analytical methods may be efficient in keeping a
running catalog of hundreds of thousands of space objects within a reasonable accuracy, if the
probability of impact with an operational satellite is detected to surpass a certain threshold,
then a more accurate numerical integration will help control engineers in deciding whether
a collision avoidance maneuver is required or, on the contrary, the integrity of the satellite is
not in risk 5.
In a semi-analytical approach the highest frequencies of the motion, which normally
have small amplitudes, are filtered analytically via averaging procedures. This filtering
allows the numerical integration of the averaged system to proceed with very long step sizes.
Then, the short-period terms are recovered analytically, if desired, at any step of the numerical
integration 6–8.
Averaging techniques are also used for exploring questions affecting stability, such
as those derived from tesseral resonances or third-body perturbations, in a reduced-phase
space 9. In this respect, much attention has been recently paid to the long-term evolution of
classical GNSS constellations, either for operational or disposal orbits 10.
While the noncentralities of the Earth gravitational potential play a key role in the
motion of low altitude satellites, third-body perturbations have also a decisive influence
in the long-term evolution of medium- and high-altitude Earth orbits. The third-body
disturbing function is commonly given by a series expansion in Legendre polynomials. Often,
the series is truncated to the first term in the expansion 11, but this early truncation is not
always accurate enough to represent the real dynamics 4, 12, 13. Nevertheless, recursions in
the literature allow to extend the Legendre polynomial expansion to any desired order either
in classical or nonsingular elements 14–16.
Because of the physical characteristics of the Earth gravitational potential, where the
second-order zonal coefficient J2 clearly dominates over all other harmonics, second-order
effects of J2 may be important and must be taken into account when the effect of higher-
order harmonics is studied. Correspondingly, the truncation in the expansion of third-body
perturbations must include terms of magnitude comparable to J2 -squared. Because the third-
body disturbing function is expanded in the ratio semi-major axis of the satellite’s orbit
to semi-major axis of the third-body’s orbit, the degree at which the expansion must be
truncated depends on the altitude of the satellites to be studied.
In this paper we study the effect of lunisolar perturbations on high-altitude orbits
about a noncentral Earth, which is assumed to be oblate although without equatorial symme-
try. More specifically, we are interested in the semi-analytical integration of satellites on alti-
tudes of classical global navigation satellite system GNSS constellations such as GPS, Glo-
nass, or Galileo. Note, however, that the assumption of an axisymmetric geopotential pre-
vents to tackle the tesseral resonance problem that commonly suffer this kind of orbits. With
respect to previous research 17, where we approached the general case of third-body pertur-
bations via double averaging, we release here the common simplification of assuming the
third-body in circular orbit. Also, we focus on the case of Earth’s artificial satellites deal-
ing with more real lunisolar perturbations. We do that because recent results 18 seem to con-
tradict the claim that taking the third-body in circular orbit does not produce any noticeable
degradation in the long-term propagation of real Earth orbits 19. Besides, for the actual
values of the orbits of the sun and moon, neglecting terms in the eccentricity is not
consistent with a higher-order expansion of the lunisolar disturbing function. Hence, both the
eccentricity of the orbits of the sun and moon are taken into account in the present work. For
the latter, the moon, the orbit is assumed to remain with constant inclination with respect to
the ecliptic plane, over which the longitude of the ascending node moves with linear motion.
The argument of the perigee of the moon is also assumed to evolve linearly, while we take
the apparent orbit of the sun to be purely Keplerian.
We use canonical perturbation theory by Lie transforms 20–22. The order of each
term of the disturbing function is determined by a virtual small parameter that is taken
proportional to the ratio of the satellite’s orbit semimajor axis to the moon’s orbit semi-major
axis. Then, we check that the magnitude of second order terms due to the Earth oblateness is
comparable to that of the fifth degree in the expansion of the moon’s third-body potential, and
hence we truncate the moon’s third-body potential up to the fifth degree. On the contrary, the
usual truncation up to the second degree is enough for modeling the sun disturbing function
assumed that we limit the theory to the order of J2 -squared terms. The Hamiltonian model
also takes into account the asymmetries in the Earth potential caused by the J3 term, whose
influence in the dynamics of the lower eccentricity orbits is clearly noted.
The initial Hamiltonian is of three degrees of freedom and time dependent. Note, how-
ever, that the time appears in different scales related to the mean motion of the moon, that of
the sun, and the rate of variation of the moon’s argument of the perigee and longitude of the
ascending node. In our averaging procedure, we eliminate the mean anomaly of the satellite
and that of the moon to obtain a two-degrees-of-freedom Hamiltonian, which still depends
on time albeit only through very slowly varying quantities. As far as we do not find reso-
nances between the mean motions of the satellite and the moon, the averaging can be done in
two steps: the mean anomaly of the satellite is removed first by means of a classical Delaunay
normalization 23; then, a similar transformation is used to eliminate the mean anomaly
of the moon. The splitting of the averaging has the advantage of simplifying computations.
Specifically, the generating function of each transformation is obtained from the solution of
simple quadratures.
The averaging is carried out not only in closed form of the satellite’s orbit eccentricity,
thus making the simplified Hamiltonian useful for studying the long-term evolution of any
orbit, but also in closed form of the eccentricity of the moon’s orbit. Nevertheless, because of
the small values of the eccentricity of the orbits of both perturbing bodies as well as the small
inclination of the orbit of the moon with respect to the ecliptic, further simplifications can be
done by neglecting terms of order higher than the truncation order of the theory.
Numerical experiments using the doubly averaged Hamiltonian demonstrate that the
inclusion in the model of the orbital eccentricities of the sun and moon does not cause
qualitative differences with respect to the circular orbit approximation. Besides, the re-
covery of the short-period effects by means of the analytical transformation equations of the
averaging provides a quite reasonable precision in the long-term integrations.
2. Model
We study the motion of an Earth artificial satellite of negligible mass whose Keplerian motion
is slightly distorted by the noncentralities of the geopotential due to the Earth’s oblateness
and latitudinal asymmetry, as represented by the harmonic coefficient J2 and J3 , respectively,
and under the point-mass attraction of the sun and moon. Thus, the motion of the satellite is
derived from the potential

μ μ α2 3 z2 1 μ α3 z 5z2 3
V− J2 − J3 − V V , 2.1
r r r2 2 r2 2 r r3 r 2r 2 2
where μ is the Earth’s gravitational parameter, α its equatorial radius, Ji is the zonal harmonic
coefficient of degree i, r is distance from the origin, and z is the satellite’s coordinate in the
direction of the symmetry axis of the Earth. In the mass-point approximation, the third-body
disturbing potential V , ∈ , , is

μ r r · r
V − − 2 , 2.2
r r − r r
where μ is the third-body’s gravitational parameter, r and r are the radius vector of the
satellite and of the third-body, respectively, of corresponding modulus r and r . If the
disturbing body is far away from the perturbed body, 2.2 can be expanded in power series
of the ratio r/r :

n2 a3 & r j
V −β Pj cos ψ , 2.3
r j≥2 r
where β m /m m, n is the mean motion of the third-body in its orbit of semi-major axis
a , Pj are Legendre polynomials, and ψ is the angle encompassed by r and r . The absence
of the Keplerian term −μ /r as a summand in 2.3 has no effect in the restricted problem
approximation, in which the mass of the satellite has negligible effects on the third-body’s
orbit.
In our model we assume an Earth-centered frame with the origin defined by the
intersection of the Earth’s equator and the ecliptic; we both of which consider fixed planes.
Whereas we take the sun’s apparent orbit about the Earth to be purely Keplerian, we assume
that the moon moves with Keplerian motion on a precessing ellipse that evolves with constant
rate of the argument of perigee. Besides, the moon’s orbital plane is assumed to have constant
inclination with respect to the ecliptic; yet it is also assumed to be precessing over the ecliptic
with linear variation of the longitude of the ascending node.
Calling a, e, i, ω, Ω, and M to semimajor axis, eccentricity, inclination, argument of
the perigee, longitude of the ascending node, and mean anomaly, respectively, we use the
approximate values
a 150 × 106 km, e 0.017, i ε 23.44 deg, 2.4

for the sun, where ε notes the obliquity of the ecliptic, we assume a fictitious epoch in which
ω Ω 0, and take n 2π/sidereal year. For the moon,
a 384 400 km, e 0.055, i J 5.1 deg 0.089, 2.5
and take Ω N N0 nN t, ω w w0 nw t, with
2π 2π 2π
nN − , nw − nN , n − nw − nN . 2.6
18.61 year 8.85 year 27.32 day
We recall that J and N are referred to the ecliptic. Besides, β 1 for the sun and β 1/28.8245
for the moon.
Remark that this simple model is not, of course, valid for computing precise ephemeris
of the moon, who may be clearly out of the predicted position because of the amplitude of
periodic oscillations in N and J, and also in e and i , comper 24. Nevertheless, it will be
enough for our purposes of investigating the qualitative features that lunisolar perturbations
introduce in the long-term behavior of Earth’s artificial satellites.
3. Perturbation Approach
By using canonical variables we can study the problem in Hamiltonian form. Besides, in order
to apply perturbation theory, we arrange the Hamiltonian as a power series in a small para-
meter:
& δm
H Hm,0 . 3.1
m≥0
m!
We are interested in high-altitude orbits in the range of 20 000 to 35 000 km, the so-
called upper mean earth orbit MEO region. In order to take account of all relevant lunisolar
perturbations, we take a virtual small parameter proportional to the ratio semimajor axis of
the satellite semimajor axis of the moon. For the altitudes of interest, this ratio is δ ∼ O10−1 .
Then, we find that the Earth’s oblateness coefficient is a fourth-order quantity, and the J3 effect
appears at the eighth order. With respect to the moon the consecutive terms of the Legendre
polynomials expansion of the moon disturbing function appear at consecutive orders of the
Hamiltonian, starting by the fifth order. We include up to the fifth degree, whose effect may
be comparable to second-order effects of J2 . For the sun, it is enough to take the first term in
the Legendre polynomials expansion. Then, the zero-order term is the Keplerian
1
H0,0 − n2 a2 . 3.2
2
Besides, Hi,0 0 for i < 4, and

H4,0 α2 a3 z2
2 J2 3 1−3 2 ,
4!H0 a r r
2 3
H5,0 n a n2 a3
β 2 3 M 2 M2 ,
5!H0 n r n2 r3

H6,0 n2 a4
β 2 4 M 3 , 3.3
6!H0 n r

H7,0 n2 a5
β 2 5 M 4 ,
7!H0 n r

H8,0 n2 a6 α3 a4 z z2
β 2 6 M 5 J3 3−5 2 ,
8!H0 n r a3 r 4 r r

where
ai−2 r i
Mi 2 i−2 ai
Pi cos ψ , 3.4
a
with
xx yy zz xx yy zz

cos ψ , cos ψ ,
rr rr
3.5
aη2
r ,
1 e cos f
√
where f is the true anomaly and η 1 − e2 is the usual eccentricity function. The equations
of motion are obtained from the Hamilton equations:

d
, g, h ∂H dL, G, H ∂H
, − . 3.6
dt ∂L, G, H dt ∂
, g, h
Recall that the Cartesian coordinates of the satellite are expressed in orbital elements
by means of simple rotations. Thus,
⎛ ⎞ ⎛ ⎞
x r
⎝y⎠ R3 −ΩR1 −iR3 −θ⎝0⎠, 3.7
z 0
where θ f ω is the argument of latitude and R1 and R3 are the usual rotation matrices
about the x- and z-axes, respectively.
Since the Hamiltonian must be expressed in canonical variables, we assume hereafter
that the orbital elements of the satellite are always expressed as function of the Delaunay
√
variables
, g, h, L, G, H, given by the known relations
M, g ω, h Ω, L μ a,
G L η, and H G cos i. Nevertheless, we use the orbital elements notation because of its
immediate physical insight.
Delaunay’s variables are singular for zero eccentricity and/or zero inclination. In
spite of that, the Lie transforms theory is naturally computed in Delaunay variables. Once
the generating function of the averaging is computed, it can be applied to any function of
the Delaunay’s variables. Specifically nonsingular variables are assumed to be functions of
Delaunay variables to obtain the transformation equations of the averaging in nonsingular
variables 25.
We note that the sun and moon coordinates are assumed to be known functions of
time. Therefore, 3.1 is a time-dependent Hamiltonian of three degrees of freedom. In our
perturbation approach, we avoid dealing with time by moving to an “extended” phase space
see, for instance, 26. Since the time-dependent variables evolve in quite different time
scales, we find convenient to introduce four new pairs of canonical conjugate variables: l, L
for the mean anomaly of the sun and its conjugate momentum λ, Λ the same for the moon,
w, W for the moon argument of perigee and its conjugate momentum, and N, Γ for the
moon longitude of the ascending node and its conjugate momentum. The specific values of
the introduced canonical momenta are irrelevant for our purposes, and we find convenient
to make the ordering
H1,0 Λn , H2,0 2!Ln , H3,0 3!Wnw ΓnN . 3.8
Once we have set the Hamiltonian order, we will apply perturbation theory by the
Lie transforms in order to filter the short-period effects in the potential equation 2.1. More
precisely, we base our computations on Deprit’s algorithm, which is specifically designed for
automatic computation by machine 21, 27.
3.1. The Delaunay Normalization
We compute the canonical transformation T :

, g, h, L, G, H →
, g , h , L , G , H that
removes the mean anomaly of the satellite from the Hamiltonian up to the eighth order.
Note that the mean anomaly does not appear explicitly but through its dependence on the
true anomaly. This fact introduces some subtleties in the averaging procedure, but the usual
differential relations between true, mean, and eccentric anomalies allow to carry out the
normalization on closed form of the eccentricity.
After normalization, we get the new Hamiltonian
& δm
H ≡ T : H H0,m , 3.9
m≥0
m!
where the H0,m terms are expressed in the prime variables. We find H0,m Hm,0 for m < 4,
and

H0,4 α2 1 3
2 J 2 3 1 − s2 ,
4!H0,0 a η 2
3 - .
H0,5
2
n a n2 a3 - .
β 2 3 M 2 M2 ,
5!H0,0 n r
n2 r 3

H0,6 n2 a4 - .
β 2 4 M 3 ,
6!H0,0 n r

H0,7 n2 a5 - .
3.10
β 2 5 M 4 ,
7!H0,0 n r

H0,8 n2 a6 - . α3 3e α4 2 3
β 2 6 M J 3 4 − 5s 2
s sin ω J
8!H0,0 n r 5
a3 4η5 a4 2 32η7

+ ,
1
× 8 5 2η − η2 − 8 10 6η − η2 s2 35 36η 5η2 s4
2

1 − η+ ,
− 2 15 30η 7η − 5 7 14η 3η s s cos 2ω ,
2 2 2 2
1 η
where s and c stand for sine and cosine of the inclination, respectively;
&
am−2 m/2 &m
Mm
m−2
Em,2j k pm,2j k,i tm,i cos φ dm,i

sin φ ,
a j0 i−m 3.11

φ 2j k ω iΩ, k m mod 2, ∈ , ,
where m/2 is an integer division. The eccentricity coefficients E, the inclination ones p, and
the third-body direction coefficients t and d are given in Tables 1, 2, and 3.
Note that except for J2 -squared terms the new Hamiltonian is obtained by the simple
average over the mean anomaly of all the terms in 3.1. Besides, we introduced Kozai’s
arbitrary constant 28 in the solution of the fourth-order generating function to keep the
prime variables as close as possible to the average value of corresponding osculating ones.
The semimajor axis of the satellite remains constant after averaging, as well as its
canonical partner, the Delaunay action L . Then, the time evolution of the mean anomaly
decouples from the two-degrees-of-freedom, time-dependent system

d g , h ∂H dG , H ∂H
, − . 3.12
dt ∂G , H dt ∂ g ,h
The numerical integration of this system can be done with longer step sizes than the
original one because of the filtering of short periodic effects via averaging. At each step of
the numerical integration, the osculating elements can be recovered analytically using the
transformation equations computed also by the Lie transforms method.
Table 1: Direction coefficients tj,k and dj,k and eccentricity coefficients Ej,k . The unit vector x, y, z defines
the direction of the center of mass of the third body.
j k tj,k dj,k Ej,|k|

0 1 − 3z2 0 2 3e2
2 ±1 yz −kxz
±2 x2 − y 2 kxy e2
0 0 z3 − 5z2
±1 x1 − 5z2 ky1 − 5z2 4 3e2 e
3
±2 4xyz −kx2 − y2 z
±3 3xx2 − 3y2 ky3x2 − y2 e3
0 3 − 30z2 35z4 0 8 40e2 15e4
±1 yz3 − 7z2 −kxz3 − 7z2
4 ±2 x − y2 1 − 7z2
2
kxy1 − 7z2 2 e2 e2
±3 3y3x2 − y2 z −kxx2 − 3y2 z
±4 x4 − 6x2 y2 y4 kxyx2 − y2 e4
0 0 z15 − 70z2 63z4
±1 x1 − 14z2 21z4 ky1 − 14z2 21z4 8 68e2 29e4 e
±2 4xyz1 − 3z2 −kx2 − y2 z1 − 3z2
5
±3 3xx2 − 3y2 1 − 9z2 ky3x2 − y2 1 − 9z2 8 7e2 e3
±4 16xyx2 − y2 z −kx4 − 6x2 y2 y4 z
±5 5xx4 − 10x2 y2 5y4 ky5x4 − 10x2 y2 y4 e5
Table 2: Inclination coefficients p2,j,k and p3,j,k ; for brevity, we use the notation γ ≡ 1 ± c.
k p2,0,k p2,2,k p3,1,k p3,3,k

0 1/82 − 3s2 15/8s2 15/64s1 − 5c2 −175/64s3
±1 3/4cs ∓15/4sγ 15/256γ1 ± 10c − 15c2 −525/256s2 γ
±2 3/16s2 15/16γ 2 ±75/256sγ1 ∓ 3c ±525/256sγ 2
±2 −25/256s2 γ −175/768γ 3
Table 3: Inclination coefficients p4,j,k and p5,j,k ; for brevity, we use the notation γ ≡ 1 ± c.
j k p4,j,k p5,j 1,k

0 3/20483 − 30c2 35c4 −15/4096s1 − 14c2 21c4
±1 −15/512sc3 − 7c2 −15/8192γ1 ± 28c − 42c2 ∓ 84c3 105c4
±2 −15/1024s2 1 − 7c2 ∓105/4096sγ1 ∓ 3c − 9c2 ± 15c3
0
±3 35/512s3 c 35/16384s2 γ1 ± 6c − 15c2
±4 105/4096s4 ±315/32768s3 γ1 ∓ 5c
±5 −63/16384s4 γ
0 −105/512s2 1 − 7c2 245/8192s3 1 − 9c2
±1 ±105/256sγ1 ± 7c − 14c2 735/16384s2 γ1 ± 6c − 15c2
±2 105/256γ 2 1 ∓ 7c 7c2 ±735/8192sγ 2 1 ∓ 12c 15c2
2
±3 ∓245/256sγ 2 1 ∓ 2c −245/98304γ 3 13 ∓ 54c 45c2
±4 735/1024s2 γ 2 ±2205/65536sγ 3 3 ∓ 5c
±5 −441/32768s2 γ 3
0 2205/2048s4 −6615/8192s5
±1 ∓2205/512s3 γ −33075/16384s4 γ
±2 2205/1024s2 γ 2 ±33075/8192s3 γ 2
4
±3 ∓735/512sγ 3 −11025/32768s2 γ 3
±4 2205/4096γ 4 ±33075/65536sγ 4
±5 −1323/32768γ 5
3.2. Elimination of the Mean Anomaly of the Moon

A new canonical transformation T :
, g , h , L , G , H →
, g , h , L , G , H removes
the mean anomaly from the Hamiltonian. The algorithm starts now by setting the new Hami-
ltonian
& δm
K Km,0 , 3.13
m≥0
m!
where Km,0 H0,m . Then, we express the direction of the moon x , y , z in terms of the
moon’s orbital elements by the composition of rotations:
⎛ ⎞ ⎛ ⎞
x r
⎝y ⎠ R1 −εR3 −NR1 −JR3 −θ ⎝ 0 ⎠, 3.14
z 0
where θ f ω . The components of the moon direction vector are then replaced
in coefficients t , d of Table 1. Finally, application of the Lie triangle provides the new
Hamiltonian
& δm
K ≡ T : K K0,m , 3.15
m≥0
m!
where the K0,m terms are expressed in the double prime variables, as well as the generating
function of the transformation.
The new averaging is similar to the preceding Delaunay normalization in the sense
that we base on the differential relations between the true and mean anomalies to perform the
averaging. As before, up to the computed order in the perturbation theory, there is no coupl-
ing between the different Hamiltonian terms, which, therefore, reduce to their averaging over
the mean anomaly of the moon λ. Thus, K0,m Km,0 for m < 5, and
@ A
K0,5 n 2
a3 - . n2 a3 - .
β 2 3
M
2 2 3 M
2
,
5!K0,0 n r
n r
λ
@ A
K0,6 n2 a4 - .
β 2 M ,
4 3
6!K0,0 n r
λ
@ A
K0,7 n 2
a5 - .
β 2 M ,
7!H0,0 5 4
n r
3.16
λ
@ A
K0,8 n2 a - 6. α3 3e α4 2 3
β 2 M J 4 − 5s 2
s sin ω J
3
8!K0,0 n 6
r 5
a3 4η5 a4 2 32η7
λ
+ ,
1
× 8 5 2η − η2 − 8 10 6η − η2 s2 35 36η 5η2 s4
2

1 − η+ ,
− 2 15 30η 7η − 5 7 14η 3η s s cos 2ω ,
2 2 2 2
1 η
Table 4: Averaged moon direction coefficients :tj,k . Here, c ≡ cos J, s ≡ sin J, and e ≡ e .
j k :tj,k

3 2 1 3
0 1− e 2 − 3σ 2 2 − 3s2 − 3σκcs cos N σ 2 s2 cos 2N
2 4 4

3 1 1 1
2 ±1 1 − e2 σκ2 − 3s2 1 − 2σ 2 cs cos N − σκs2 cos 2N
2 4 2 4

3 1 2 1
±2 1 − e2 σ 2 − 3s2 σκcs cos N 2 − σ 2 s2 cos 2N
2 4 4
0 0
1 5
±1 e 4 − 5σ cosN ω − σκs cos2N ω
2
4 2
3
±2 eσκ cosN ω 1 − 2σ 2 s cos2N ω

9 2 9
±3 e σ cosN ω σκs cos2N ω
4 2
3 15
0 8 − 40σ 2 35σ 4 − 4 − 7σ 2 σκs cos N
8 2
3 3
±1 4 − 7σ 2 σκ 4 − 29σ 2 28σ 4 s cos N
8 8
3 3
4 ±2 6 − 7σ 2 σ 2 3 − 7σ 2 σκs cos N
8 2
9 3 9
±3 σ κ 3 − 4σ 2 σ 2 s cos N
8 8
3 4 3 3
±4 σ σ κs cos N
8 2
where, from 3.11,
@ A
am 1 - . &
am−2 m/2 &m

M m Em,2j k p m,2j k,i :tm,i cos φ d:m,i sin φ ,
m 1
r
am−2
j0 i−m 3.17
λ

φ 2j k ω iΩ, k m mod 2,
with
@ m 1 A @ m 1 A
a a
:tm,i t
m,i , d:m,i d
m,i . 3.18
r r
λ λ
To get some further simplification, we note that e ∼ i ∼ Oδ. Thus, in our eighth-
m
order theory, we neglect higher-order terms factored by e sinn J δk such that m n k > 8.
Under these simplifying assumptions, the coefficients :tm,i and d:m,i are presented in Tables 4
and 5, respectively.
Table 5: Averaged moon direction coefficients d:j,k . Here, c cos J, s sin J, and e ≡ e .
j k d:j,k
0 0

3 1 1
2 ±1 k 1 − e2 κcs sin N − σs2 sin 2N
2 2 4

3 1 1
±2 k 1 − e2 σcs sin N κs2 sin 2N
2 2 4

3 3 15
0 e 2 − 5σ 2 κs sin ω 4 − 5σ 2 σ sinN ω − σ 2 κs sin2N ω
2 4 4

3 1 5
±1 ke − 4 − 5σ 2 σs sin ω 4 − 15σ 2 κ sinN ω − 2 − 3σ 2 σs sin2N ω
2 4 4
3
3 1 1
±2 ke σ 2 κs sin ω − 2 − 3σ 2 σ sinN ω − 2 − 3σ 2 κs sin2N ω
2 4 4

3 3 3
±3 ke − σ 3 s sin ω σ 2 κ sinN ω 2 − σ 2 σs sin2N ω
2 4 4
0 0
3
±1 k 4 − 7σ 2 κs sin N
8
3
±2 k 6 − 7σ 2 σs sin N
4 8
9
±3 k σ 2 κs sin N
8
3
±4 k σ 3 s sin N
8
As in the preceding averaging, adequate arbitrary constants have been introduced in

the computation of the generating function to guarantee its average to zero.
After the double averaging, the Hamiltonian only depends on long-period terms
related to the sun’s apparent motion and very long-period terms related to the precession of
the nodes and recession of the line of apsides of the moon’s orbit. The numerical integration
of corresponding Hamilton equations

d g , h ∂K dG , H ∂K
, − 3.19
dt ∂G , H dt ∂ g ,h
is, now, very much faster and efficient.
4. Numerical Experiments
For the numerical tests we used a higher-order Runge-Kutta method. Specifically, the
numerical integration was performed with the Dormand and Prince implementation of the
Runge-Kutta method coded in FORTRAN 77 by Hairer et al. 29.
a (m)
3000
1000
−1000
−3000
0 5 10 15 20 25 30
a (m)
300
100
−100
−300
0 5 10 15 20 25 30
60
a (m)
20
−20
0 5 10 15 20 25 30
a (m)
2
−2
0 5 10 15 20 25 30
a (m)
2
0
−2
0 5 10 15 20 25 30
a (m)
0.4
0
−0.4
0 5 10 15 20 25 30
Figure 1: Semianalytical theory versus numerical integration. From top to bottom, errors in semimajor axis
of the secular terms and after computing the fourth, fifth, sixth, seventh, and eighth order transformation
equations, respectively. Abscissas are days.
Table 6: Initial conditions in the single- top and double-averaged phase space bottom after different
truncation orders n of the semi-analytical theory. Distances are km, and angles are deg.
n 28560 − a 0.2 − e 56 − i 0−ω 72 − Ω 0−
4 3.06217 0.0000723 0.0014902 0.0000000 0.0000000 0.0000000

5 3.06923 0.0000722 0.0014713 −0.0016703 0.0000132 0.0016800
6 3.06073 0.0000720 0.0014808 −0.0017529 0.0000081 0.0018807
7 3.06077 0.0000720 0.0014820 −0.0017477 0.0000075 0.0018817
8 3.06085 0.0000721 0.0014819 −0.0017510 0.0000042 0.0018783
4 3.06073 0.0000453 −0.0030387 −0.0149534 −0.0027445 0.0167622
5 3.06923 0.0000431 −0.0026903 −0.0153934 −0.0028968 0.0172767
6 3.06073 0.0000432 −0.0026846 −0.0153427 −0.0029047 0.0173255
7 3.06077 0.0000432 −0.0026744 −0.0153287 −0.0028995 0.0173220
8 3.06085 0.0000433 −0.0026745 −0.0153319 −0.0029027 0.0173187
To illustrate the significance of recovering the short-period effects up to higher orders,

we first show in Figure 1 a sequence of the errors obtained in the semimajor axis after one
month semi-analytical propagation. For this example, we take the initial osculating elements
a 28 560 km, e 0.2, i 56 deg., ω 0, Ω 72 deg., and
0; besides we assumed
that both the mean anomalies of the moon and the sun are zero at the origin of time.
Corresponding elements in the single- and double-averaged phase space depend on the
truncation order of the perturbation theory and are presented in the top and bottom parts
of Table 6, respectively.
0.2
58
0.15 57
i (deg)
56
e 0.1
55
0.05 54
53
0 20 40 60 80 100 0 20 40 60 80 100
Years Years
a b
Figure 2: Sample long-term propagations for ω 0 and Ω 0 full line, 90 dashed, 180 dotted, and 270 deg.
dash-dotted. a Eccentricity variation. b Inclination variations.
In reference to Figure 1, the top plot shows a direct comparison between the
numerical integration of the Hamilton equations of the original, nonaveraged problem—
whose disturbing potential is given in 2.1—and that of the long-term Hamiltonian after the
double averaging up to the eight order of the small parameter. Then, from top to bottom, we
show the errors obtained at each step of the integration when recovering short-period terms
by computing the transformation equations of the averaging up to the fourth, fifth, sixth,
seventh, and eighth orders. For the latter, the amplitude of the periodic errors is reduced to
several centimeter; note that, in addition to short-period errors related to the orbital period of
the satellite, we can appreciate a two-week modulation related to the moon’s mean motion.
Remark that the amplitude of the periodic errors roughly divides by ten with each order of
the transformation equations, which is consistent with the assumed magnitude of the virtual
small parameter δ ∼ 0.1.
The short-period effects can be ignored in the study of the long-term orbital behavior,
where the simple propagation of the double-averaged equations appears very fast and
efficient. Sample propagations are shown in figures below, which show the important effects
that have the initial right ascension of the ascending node and argument of perigee of the
satellite’s orbit in the long term and specifically in the eccentricity and inclination evolution
of the satellite’s orbit. Thus, Figure 2 shows the notably different evolution of the satellite’s
eccentricity and inclination for different initial nodes; for the other initial conditions we have
taken the same as in the preceding short-term propagations but now assuming directly that
they are mean elements. In fact, the more relevant parameter is the difference between the
node of the satellite’s orbit and that of the moon’s orbit, as illustrated in Figure 3 where the
initial longitude of the ascending node of the moon’s orbit over the ecliptic has been taken as
N 180 instead of N 0 deg.
Figure 4 shows that the effect of the initial argument of the perigee of the satellite’s
orbit is also important in the long term, eccentricity evolution. The effect is almost negligible
in the orbital inclination in the long-term and it is not presented.
Finally, we must mention that further tests demonstrated that there are no important
qualitative differences in the long term when using lower-order truncations of the theory,
resulting in faster numerical integration of the mean elements. So the fifth-order truncation
should be the preferred long-term Hamiltonian. Besides, we also checked for a variety of
orbits that making e e 0 does not either introduce qualitative differences in the long-
term, in agreement with 19.
0.3 60
0.25 59
0.2 58
i (deg)
57
e 0.15
56
0.1
55
0.05 54
0 53
0 20 40 60 80 100 0 20 40 60 80 100
Years Years
a b
Figure 3: The same as Figure 2 but now N 180 instead of N 0 deg.
0.12
0.1
0.08
e 0.06
0.04
0.02
0 20 40 60 80 100
Years
Figure 4: Sample long-term propagations for Ω 0 and ω 0 full line, 30 dashed, 60 dotted, and 90 deg.
dash-dotted. Eccentricity variation.
5. Conclusions
Modeling lunisolar perturbations on high-altitude Earth orbits requires to retain high degrees
in the Legendre polynomials expansion of the third-body disturbing function of the moon. In
consequence the eccentricity of the third-body’s orbit cannot be neglected in the case of either
the moon or the sun.
The long-term behavior of high-altitude Earth orbits is approached in a semianalytical
way via averaging procedures, in which we take advantage of the different scales in which
appears the time to do the averaging in the extended phase space. In addition, up to second-
order terms in the Earth’s oblateness coefficient, the averaging has been computed in closed
form of the eccentricity, and, therefore, the semianalytical integration can be applied to any
orbit.
Sample numerical propagations of test cases show that the more relevant parameter on
the long-term behavior is the difference between the right ascension of the ascending node of
the satellite’s orbit and the longitude of the ascending node of the moon’s orbit over the eclip-
tic, having an apparent effect manifested by the almost-secular growing of the orbit eccentric-
ity and also by very-long-period oscillations of the inclination with an amplitude of several
degrees. Also the initial argument of the perigee of the satellite’s orbit has notable effects in
the satellite’s orbit, but only in what respects to the long-term evolution of the eccentricity.
Acknowledgments
Part of this research has been supported by the Government of Spain Projects AYA 2009-
11896, AYA 2010-18796, and Grant Gobierno de La Rioja Fomenta 2010/16.
References
1 T. S. Kelso, “Analysis of the Iridum 33—Cosmos 2251 Colission,” Paper AAS 09-368, American Astro-
nautical Society, 2009.
2 O. Montenbruck and E. Gill, Satellite Orbits: Models, Methods, and Applications, Springer, Berlin, Ger-
many, 2000.
3 D. A. Vallado, Fundamentals of Astrodynamics and Applications, Microcosm Press and Kluwer Academic,
El Segundo, Calif, USA, 2007.
4 P. J. Cefola, V. Yurasov, Z. Folcik, E. Phelps, R. J. Proulx, and A. Nazarenko, “Comparison of the DSST
and the USM Semi-Analytical Orbit Propagators,” Paper AAS 03-236, American Astronautical Soci-
ety, 2003.
5 E. J. Fletcher, “Status and progress in the space surveillance and tracking segment of ESA’s space situa-
tional awareness programme,” in Proceedings of the AMOS Conference, Maui, Hawaii, USA, September
2010.
6 W. D. McClain, A Recursively Formulated First-Order Semianalytic Artificial Satellite Theory Based on the
Generalized Method of Averaging : The Generalized Method of Averaging Applied to the Artificial Satellite
Problem, vol. 1, Computer Sciences Corporation CSC/TR-77/6010, 1977.
7 S. Valk, A. Lemaı̂tre, and F. Deleflie, “Semi-analytical theory of mean orbital motion for geosynchr-
onous space debris under gravitational influence,” Advances in Space Research, vol. 43, no. 7, pp. 1070–
1082, 2009.
8 J. F. San-Juan, L. M. López, and R. López, “Higher-order analytical attitude propagation of an oblate
rigid body under gravity-gradient torque,” Mathematical Problems in Engineering, vol. 2012, Article ID
123138, 15 pages, 2012.
9 O. Winter, D. Mourão, C. de Melo, E. Macau, J. Ferreira, and J. Carvalho, “Controlling the eccentricity
of polar lunar orbits with low-thrust propulsion,” Mathematical Problems in Engineering, vol. 2009,
Article ID 159287, 10 pages, 2009.
10 D. Merguizo Sanchez, T. Yokoyama, P. I. De Oliveira Brasil, and R. R. Cordeiro, “Some initial condi-
tions for disposed satellites of the systems GPS and Galileo constellations,” Mathematical Problems in
Engineering, vol. 2009, Article ID 510759, 22 pages, 2009.
11 R. A. Broucke, “Long-term third-body effects via double averaging,” Journal of Guidance, Control, and
Dynamics, vol. 26, no. 1, pp. 27–32, 2003.
12 G. Metris and P. Exertier, “Semi-analytical theory of the mean orbital motion,” Astronomy and Astro-
physics, vol. 294, pp. 278–286, 1995.
13 A. F. B. A. Prado, “Third-body perturbation in orbits around natural satellites,” Journal of Guidance,
14 W. M. Kaula, “Development of the lunar and solar disturbing functions for a close satellite,” Astro-
nomical Journal, vol. 67, pp. 300–303, 1962.
15 G. Giacaglia, “Lunar perturbations of artificial satellites of the earth,” Celestial Mechanics, vol. 9, pp.
239–267, 1974.
16 S. K. Collins and P. J. Cefola, “Double Averaged Third Body Model for Prediction of Super-Syn-
chronous Orbits over Long Time Spans,” Paper AAS 79-135, American Astronautical Society, 1979.
17 M. Lara, J. F. San-Juan, L. M. Lopez, and P. Cefola, “On third-bodyperturbations on high-altitude
orbits,” submitted to. Celestial Mechanics and Dynamical Astronomy.
18 R. C. Domingos, R. Vilhena de Moraes, and A. F. B. D. A. Prado, “Third-body perturbation in the case
of elliptic orbits for the disturbing body,” Mathematical Problems in Engineering, vol. 2008, Article ID
763654, 14 pages, 2008.
19 C. C. Chao and R. A. Gick, “Long-term evolution of navigation satellite orbits: GPS/GLON-
ASS/GALILEO,” Advances in Space Research, vol. 34, no. 5, pp. 1221–1226, 2004.
20 G.-I. Hori, “Theory of general perturbation with unspecified canonical variable,” Publications of the
Astronomical Society of Japan, vol. 18, no. 4, pp. 287–296, 1966.
21 A. Deprit, “Canonical transformations depending on a small parameter,” Celestial Mechanics, vol. 1,
pp. 12–30, 1969.
22 J. A. Campbell and W. H. Jefferys, “Equivalence of the perturbation theories of Hori and Deprit,” Cele-
stial Mechanics, vol. 2, no. 4, pp. 467–473, 1970.
23 A. Deprit, “Delaunay normalisations,” Celestial Mechanics, vol. 26, no. 1, pp. 9–21, 1982.
24 J. M. A. Danby, Fundamentals of Celestial Mechanics, Willmann-Bell, Richmond, VA, USA, 1992.
25 A. Deprit and A. Rom, “The main problem of artificial satellite theory for small and moderate
eccentricities,” Celestial Mechanics, vol. 2, no. 2, pp. 166–206, 1970.
26 D. Brouwer and G. M. Clemence, Methods of Celestial Mechanics, Academic Press, New York, NY, USA,
1961.
27 J. F. San-Juan, L. M. López, and R. López, “MathATESAT: a symbolic-numeric environment in astro-
dynamics and celestial mechanics,” Lecture Notes in Computer Science, vol. 6783, part 2, pp. 436–449,
2011.
28 Y. Kozai, “Second-order solution of artificial satellite theory without air drag,” Astronomical Journal,
vol. 67, no. 7, pp. 446–461, 1962.
29 E. Hairer, S. P. Nørsett, and G. Wanner, Solving Ordinary Differential Equations. I. Non-Stiff Problems,
vol. 8, Springer, Berlin, Germany, 2nd edition, 1993.
doi:10.1155/2012/140301
Research Article
Error Modeling and Analysis for
InSAR Spatial Baseline Determination of
Satellite Formation Flying
Jia Tu, Defeng Gu, Yi Wu, and Dongyun Yi

Department of Mathematics and Systems Science, College of Science,
National University of Defense Technology, Changsha 410073, China
Correspondence should be addressed to Jia Tu, tu jia jia@yahoo.com.cn
Received 30 September 2011; Revised 9 December 2011; Accepted 12 December 2011
Copyright q 2012 Jia Tu et al. This is an open access article distributed under the Creative
Spatial baseline determination is a key technology for interferometric synthetic aperture radar
InSAR missions. Based on the intersatellite baseline measurement using dual-frequency GPS,
errors induced by InSAR spatial baseline measurement are studied in detail. The classifications
and characters of errors are analyzed, and models for errors are set up. The simulations of single
factor and total error sources are selected to evaluate the impacts of errors on spatial baseline
measurement. Single factor simulations are used to analyze the impact of the error of a single type,
while total error sources simulations are used to analyze the impacts of error sources induced
by GPS measurement, baseline transformation, and the entire spatial baseline measurement,
respectively. Simulation results show that errors related to GPS measurement are the main error
sources for the spatial baseline determination, and carrier phase noise of GPS observation and
fixing error of GPS receiver antenna are main factors of errors related to GPS measurement. In
addition, according to the error values listed in this paper, 1 mm level InSAR spatial baseline
determination should be realized.
1. Introduction
Close formation flying satellites equipped with synthetic aperture radar SAR antenna
could provide advanced science opportunities, such as generating highly accurate digital
elevation models DEMs from Interferometric SAR InSAR 1, 2. Compared to a single
SAR satellite system, the performance of two SAR satellites flying in close formation can be
greatly enhanced. Nowadays, close satellite formation flying has become the focus of space
technology and geodetic surveying.
In order to realize the advanced space mission goal of InSAR mission, the high-
precision determination of inter-satellite interferometric baseline 3 is a fundamental issue.
Take the TanDEM-X mission for instance. TanDEM-X mission is the first bistatic single-pass
SAR satellite formation, which is formed by adding a second TanDEM-X, almost identical
spacecraft, to TerraSAR-X and flying the two satellites in a closely controlled formation. The
primary mission goal is the derivation of a high-precision global DEM according to high-
resolution terrain information HRTI level 3 accuracy 4–6. The generation of accurate
InSAR-derived DEMs requires a precise knowledge of the interferometric baseline with an
accuracy of 1 mm 1D, RMS 7. Therefore high-precision determination of inter-satellite
interferometric baseline is a prerequisite for InSAR mission.
The interferometric baseline is defined as the separation between two SAR antennas
that receive echoes of the same ground area 8. Based on this definition, the interferometric
baseline can be denoted as the resultant vector of temporal baseline and spatial baseline, that
is,
S2 t2 − S1 t1 S2 t2 − S1 t2 S1 t2 − S1 t1 , 1.1
where t1 , t2 are epochs that two SAR antennas receive echoes of the same ground area, S1 t,
S2 t represent the positions of SAR antenna phase centers of satellite 1 and satellite 2 at
epoch t in International Terrestrial Reference Frame ITRF, respectively, S2 t2 − S1 t2 is
the spatial baseline, S1 t2 − S1 t1 is the temporal baseline which is the velocity integral of
satellite 1. For close formation flying 1 km-2 km with single-pass bistatic acquisitions, the
deviation of epochs that two SAR antennas receive echoes of the same ground area is typically
on the millisecond level. When the velocity is determined on the mm/s level, its influence in
the temporal baseline can be neglected. Therefore, the accuracy of interferometric baseline
is mainly determined by the accuracy of spatial baseline. Note that only spatial baseline is
considered in this paper.
The spaceborne dual-frequency GPS measurement scheme 9–11 is widely used
for inter-satellite baseline determination currently. This scheme for spatial baseline deter-
mination consists of two steps. Firstly, the relative position of two formation satellites is
determined by dual-frequency GPS measurement, and then spatial baseline is transformed
from inter-satellite relative position. The relative position here is the vector that links the
mass centers of two formation satellites.
In our research, impacts of the errors introduced by spatial baseline measurement are
analyzed. This paper starts with a description of spatial baseline measurement using dual-
frequency GPS. The baseline transformation from the relative position to spatial baseline
is given. In a second step, errors are classified into two groups: errors related to GPS
measurement and errors related to baseline transformation. The error characters are studied,
and the impact of each error on spatial baseline determination is analyzed from theoretical
aspect. Then the impacts of each error and total errors on spatial baseline determination are
analyzed by single factor simulations and total error sources simulations. At last, conclusions
are shown.
2. Generation of Spatial Baseline

In preparation for latter description some coordinate systems are introduced at first, which
are illustrated in Figure 1. Coordinate systems employed in this paper contain Conventional
Inertial Reference Frame CIRF, ITRF, satellite body coordinate system, and satellite orbit
XBody
ZCIRF
XOrbit
ZITRF ZBody
OS
ZOrbit
Greenwich YOrbit
meridian Satellite
OE YITRF orbit YBody
tG
XITRF
Equator
XCIRF
Vernal equinox YCIRF
Figure 1: Definitions of coordinate systems employed in this paper. CIRF, ITRF, satellite body coordinate
system, and satellite orbit coordinate system are denoted as OE -XCIRF YCIRF ZCIRF , OE -XITRF YITRF ZITRF , OS -
XBody YBody ZBody , and OS -XOrbit YOrbit ZOrbit , respectively. OE is the geocenter, and OS is the mass center of
satellite.
G1 G2
Satellite 1 Satellite 2
O1 Relative position O2
Spatial baseline
S1 S2
Figure 2: Geometric relation for spatial baseline determination. G1 and G2 are GPS receiver antenna phase
centers, O1 and O2 are mass centers, and S1 and S2 are SAR antenna phase centers.
coordinate system. CIRF used here is J2000.0 inertial system and ITRF is ITRF2000 system.
The definitions of these coordinate systems can be found in 12.
As the spatial baseline is determined by spaceborne dual-frequency GPS measurement
scheme, the entire process of spatial baseline determination consists of relative positioning
and baseline transformation. Figure 2 is the geometric relation for spatial baseline determina-
tion.
Relative positioning is the determination of O1 O2 by dual-frequency GPS observation
data. As the real position of signal reception is the phase center Gi i 1, 2 of GPS receiver
antenna, GPS observation data has to be revised to the mass center Oi i 1, 2 of satellite
using the phase center data of GPS receiver antenna during relative positioning.
From Figure 2, baseline transformation can be described as follows:
S1 S2 O1 O2 M1 · S1 O1 − M2 · S2 O2 , 2.1
where S1 S2 is the spatial baseline in ITRF, O1 O2 is the relative position of two satellites
in ITRF, Si Oi i 1, 2 is a vector that links SAR antenna phase center to mass center of
satellite in body coordinate system of Satellite i, Mi i 1, 2 is a transformation matrix of
Satellite i from satellite body coordinate system to ITRF. The flow chart of spatial baseline
determination is shown in Figure 3.
Position and SAR antenna phase center

GPS observation Baseline
velocity of position of satellite 1 in
data of satellite 1 transformation ITRF
satellite 1 in ITRF
Subtracting at same
Spatial baseline
GPS relative epoch
positioning
Position and SAR antenna phase center

velocity of Baseline
GPS observation position of satellite 2 in
data of satellite 2 transformation
satellite 2 in ITRF ITRF
Figure 3: Flow chat of spatial baseline determination.
3. Errors of Spatial Baseline Measurement

According to the generation of spatial baseline in Section 2, the errors of spatial baseline
measurement can be classified into two groups: errors related to GPS measurement, which
are introduced by relative positioning using dual-frequency GPS measurement, and errors
related to baseline transformation, which are generated by the transformation from relative
position to spatial baseline.
3.1. Errors Related to GPS Measurement

The relative positions of two satellites are determined by the reduced dynamic carrier
phase differential GPS approach. In this approach, the absolute orbits of one reference
satellite Satellite 1 are fixed, which are determined by the zero-difference reduced dynamic
batch least squares approach based on GPS measurements of single satellite. Only the
relative positions are estimated by reduced dynamic batch least-squares approach based
on differential GPS measurements. The integer double difference ambiguities for relative
positioning are resolved by estimating wide-lane and narrow-lane combinations 13. The
well-known Least-Squares Ambiguity Decorrelation Adjustment LAMBDA method 14,
15 is implemented for the integer estimate.
By differenced GPS observation, common errors can be eliminated or reduced.
International GNSS Service IGS final GPS ephemeris product orbit product and clock
product 16 is often adopted for orbit determination based on GPS observation. The
accuracy of GPS final orbit product is presently on the order of 2.5 cm. For 2 km separation of
satellite formation, the impact of GPS ephemeris error on single-difference GPS observation
is about 0.0025 mm 17, which can be neglected. The impact of GPS clock error can be
well cancelled out by differential GPS observation. Due to the close separation 1 km-
2 km and similar materials, configuration, and in-flight environment of formation satellites,
near-field multipath, thermal distortions of satellites, and other external perturbations can
also be effectively reduced by differential GPS observation. In addition, the influence of
differential ionospheric path delay is mainly from the first order, which can be eliminated by
constructing ionosphere free differential GPS observation. Therefore, the errors related to GPS
measurement that have to be considered consist of noise of GPS carrier phase measurement,
ground calibration error of GPS receiver antenna phase center, error of satellite attitude
measurement, and fixing error of GPS receiver antenna.
3.1.1. Noise of GPS Carrier Phase Measurement

The quality of GPS carrier phase observation data used is of utmost importance for relative
positioning. The noise of GPS carrier phase measurement belongs to random error, which
cannot directly be eliminated by GPS differential observation. Take the BlackJack receiver
and its commercial Integrated GPS and Occultation Receiver IGOR version, for example,
which are widely used for geodetic grade space missions and exhibit a representative noise
level of 1 mm for L1 and L2 carrier phase measurements 18. The reduced dynamic relative
positioning approach makes use of dynamical models of the spacecraft motion to constrain
the resulting relative position estimates, which allows an averaging of measurements from
different epochs. The influence of GPS carrier phase noise can be effectively reduced by
reduced dynamic relative positioning approach.
3.1.2. Ground Calibration Error of GPS Receiver Antenna Phase Center

The phase center location accuracy of the GPS receiver antenna will directly affect the veracity
of GPS observation modeling. GPS receiver antenna phase center is the instantaneous location
of the GPS receiver antenna where the GPS signal is actually received. It depends on intensity,
frequency, azimuth, and elevation of GPS receiving signal.
The phase center locations can be described by the mechanical antenna reference point
ARP, a phase center offset PCO vector, and phase center variations PCVs. The PCO
vector describes the difference between the mean center of the wave front and the ARP.
PCVs represent direction-dependent distortions of the wave front, which can be modeled
as a consistent function that depends on azimuth and elevation of the observation from
the position indicated by the PCO vector. The position of GPS receiver antenna phase
center can be measured by ground calibration, such as using an anechoic chamber and
using field calibration techniques 18, 19. Take the SEN67-1575-14 CRG antenna system
for instance. It is a dual-frequency GPS receiver antenna and has been used for TanDEM-
X mission. Its phase center has been measured by automated absolute field calibration 20.
The mean value of calibration result is shown in Figure 4 that the pattern of PCVs has obvious
character of systematic deviation. The maximum value for the mean PCVs on ionosphere-
free combination can reach to 1.5 cm. In addition, there also exist random errors in the same
direction of different receptions. The random errors are similar to the noise of GPS carrier
phase measurement and can also be smoothed by reduced dynamic relative positioning
approach.
As there is a slim difference between the line of sight LOS vectors of two satellites
during close formation flying, the common systematic errors of GPS receiver antenna phase
center and near-field multipath can be eliminated by differential GPS observation. Therefore,
the same type of GPS receiver antenna has to be selected for both formation satellites in order
to reduce the impact of these errors.
3.1.3. Error of Satellite Attitude Measurement

Satellite attitude data are obtained from star camera observations and provided as
quaternion. The error of satellite attitude measurement consists of a slowly varying bias and
a random error. Its impact on GPS relative positioning appears on the correction for GPS
observation data of single satellite, that is, the reference point of GPS observation data has to
X (azimuth = 0)
Y (azimuth = 90)
0
30
60
90
−12 −6 0 6 12
Antenna phase center (mm)
Figure 4: Ground calibrated mean PCVs result of SEN67-1575-14 CRG antenna on ionosphere-free
combination.
be corrected from GPS receiver antenna phase center to the mass center of satellite by satellite
attitude data and GPS receiver antenna phase center data. Take Satellite 1 for instance. The
j
correction in direction of LOS vector e1 in CIRF is given by
j T
δOffset,1 − e1 · MBody CIRF · O1 G1 ,
3.1
MBody CIRF MOrbit CIRF · MBody Orbit ,
where O1 G1 is GPS receiver antenna phase center location in body coordinate system of
Satellite 1, MBody CIRF is the transformation matrix from body coordinate system of Satellite
1 to CIRF and can be obtained by attitude quaternion data, MBody Orbit is the transformation
matrix from body coordinate system to orbit coordinate system of Satellite 1, and MOrbit CIRF
is the transformation matrix from orbit coordinate system of Satellite 1 to CIRF.
Assuming that the Euler angles are ϕ, θ, and ψ respectively, we can get

MBody Orbit RX ϕ · RY θ · RZ ψ , 3.2
where RX ϕ, RY θ, RZ ψ are rotation matrices around roll axis, pitch axis, and yaw axis,
respectively.
Assuming that the errors of Euler angle measurements are εϕ , εθ , and εψ , respectively,
and the corresponding error matrix of MBody CIRF is εM , the relation between εM and εϕ , εθ , εψ
can be expressed as

∂RX ∂RY ∂RZ
εM MOrbit CIRF · RY RZ · εϕ RX RZ · εθ RX RY · εψ . 3.3
∂ϕ ∂θ ∂ψ
Furthermore, the impact of Euler angle errors on MBody CIRF · O1 G1 in 3.1 can be
obtained as
εMBody CIRF ·O1 G1 MOrbit CIRF

∂RX ∂RY ∂RZ
· RY RZ · O1 G1 · εϕ RX RZ · O1 G1 · εθ RX RY · O1 G1 · εψ .
∂ϕ ∂θ ∂ψ
3.4
εMBody CIRF ·O1 G1 is a three-dimensional random vector and its magnitude can be described
as the mean value of space radius, that is
% %2
% %
σ 2MBody CIRF ·O1 G1 E %εMBody CIRF ·O1 G1 % E εTMBody CIRF ·O1 G1 · εMBody CIRF ·O1 G1 , 3.5
where |·| denotes the magnitude of a vector, E· denotes the expectation of a random variable.
Assuming Euler angle errors of different axes are independent, we can get
% %2
% ∂RX % 2
σ 2MBody CIRF ·O1 G1 %%MOrbit CIRF · RY RZ · O1 G1 %% · Var εϕ E εϕ
∂ϕ
% %2
% ∂RY %
%
%MOrbit CIRF · RX RZ · O1 G1 %% · Varεθ Eεθ 2 3.6
∂θ
% %2
% ∂RZ % 2
%
%MOrbit CIRF · RX RY · O1 G1 %% · Var εψ E εψ ,
∂ψ
where Var · denotes the variation of a random variable.

As RX ϕ, RY θ, RZ ψ, and MOrbit CIRF are orthogonal matrices, for any v ∈ R3 , we
can get
% %
% ∂RX %2
2
|RX · v| |v| , 2 % % 2
% ∂ϕ · v% ≤ |v|
% %2
% ∂RY %
|RY · v|2 |v|2 , % · v% ≤ |v|2
% ∂θ % 3.7
% %
% ∂RZ %2
|RZ · v|2 |v|2 , % · v % ≤ |v|2
% ∂ψ %
|MOrbit CIRF · v|2 |v|2 .
Taking 3.7 into 3.6, we can get
+ 2 2 ,
σ 2MBody CIRF ·O1 G1 ≤ |O1 G1 |2 · Var εϕ E εϕ Varεθ Eεθ 2 Var εψ E εψ .
3.8
Hence,
% %2 %
% j T % %2
% %
σδ2Offset,1 E |εδOffset | E %% e1 · εMBody CIRF ·O1 G1 %%
2
≤ E %εMBody CIRF ·O1 G1 % . 3.9
For differential GPS observation, the impact of attitude determination error on two
satellites can be given as follows
σΔδOffset ≤ σδ2Offset,1 σδ2Offset,2 . 3.10
According to the TanDEM-X missions, the attitude determination accuracy has a

slowly varying bias of ±0.005◦ in the yaw, pitch, and roll components plus a 0.003◦ sigma
random error 21. From 3.8, 3.9, and 3.10, we can get

0.005 2 0.003 2
σΔδOffset 2 2
≤ 3 · |O1 G1 | |O2 G2 | · π π . 3.11
180 180
Take the GPS receiver antenna ARP location of TanDEM-X mission for instance, that
is,
|O1 G1 | |O2 G2 | 1.8976 m, 3.12
we can get
σΔδOffset ≤ 0.47 mm. 3.13
3.1.4. Fixing Error of GPS Receiver Antenna

The fixing error of GPS receiver antenna is caused by the inaccuracy of the fixed position
of antenna onboard the satellite. This error is a random error for multiple repeated satellite
missions. But for a single launch, it is considered to be a fixed bias vector in satellite body
coordinate system during satellite flying.
The fixing errors of GPS receiver antenna in body coordinate system of two satellites
are assumed as follows:

ΔE1 Δx1 ; Δy1 ; Δz1 ,
3.14
ΔE2 Δx2 ; Δy2 ; Δz2 .
For a mutually observed GPS satellite j, the LOS vectors of two formation satellites
j j
are assumed to be e1 and e2 . The impact of fixing errors of GPS receiver antenna for both
formation satellites on GPS observation data can be denoted as
j
j T
Δδ1 e1 · MBody CIRF,1 · ΔE1 ,
j T 3.15
j
Δδ2 e2 · MBody CIRF,2 · ΔE2 .
The impact of fixing error of GPS receiver antenna on differential GPS observation is
j
j T j T
Δδ12 e2 · MBody CIRF,2 · ΔE2 − e1 · MBody CIRF,1 · ΔE1 . 3.16
Due to the close separation of two satellites, we can assume
j j
e1 ≈ e2 . 3.17
From 3.16 and 3.17, we can get
j
j T
Δδ12 ≈ e2 · MBody CIRF,2 · ΔE2 − MBody CIRF,1 · ΔE1
j T j T 3.18

e2 · MBody CIRF,2 · ΔE2 − ΔE1 e2 · MBody CIRF,2 − MBody CIRF,1 · ΔE1 .
As the magnitudes of ΔE1 and ΔE2 are small generally less than 0.5 mm and the
j T
difference between MBody CIRF,1 and MBody CIRF,2 is insignificant; therefore, the impact of e2 ·
MBody CIRF,2 − MBody CIRF,1 · ΔE1 in 3.18 can be neglected and the main influence is from
j T
e2 ·MBody CIRF,2 ·ΔE2 −ΔE1 . If the magnitude of GPS receiver antenna fixing error is 0.5 mm
for each formation satellite, the maximum 3-dimensional impact on relative positioning can
reach to 1 mm.
In addition, we can also draw a conclusion from the aforementioned analysis that the
GPS receiver antenna bias caused by thermal distortions of satellites can be cancelled out by
differential GPS observation.
3.2. Errors Related to Baseline Transformation

From 2.1, errors related to baseline transformation consist of two parts: one part is
introduced by transformation matrices M1 and M2 , which is mainly caused by the satellite
attitude measurement error; the other part is introduced by S1 O1 and S2 O2 , which is caused
by the inconsistency of two SAR antenna phase centers.
3.2.1. Error of Satellite Attitude Measurement

Take M1 for instance,
M1 MCIRF ITRF · MBody CIRF , 3.19
where MCIRF ITRF is a transformation matrix from CIRF to ITRF, MBody ITRF has been defined
in 3.2.
Note that the transformation from CIRF to ITRF is in accordance with IERS 1996
conventions 22 and this transformation error can be neglected. The errors of M1 and M2 are
also introduced by satellite attitude measurement errors. Similar to the analysis of satellite
attitude measurement error related to GPS measurement, from 3.8, we can obtain
+ 2 2 ,
2 2
2
σM 1 ·S1 O1
≤ |S1 O1 | · Var εϕ E ε ϕ Varεθ Eεθ Var εψ E εψ . 3.20
Hence, the impact of attitude determination errors on baseline transformation is given

as follows:

σAtt ≤ 2
σM 1 ·S1 O1
σM
2
2 ·S2 O2
. 3.21
Take the attitude determination accuracy of TanDEM-X mission for instance and select
the magnitudes of S1 O1 and S2 O2 as follows
|S1 O1 | |S2 O2 | 2 m, 3.22
we can get
σAtt ≤ 0.50 mm. 3.23
3.2.2. Consistency Error of SAR Antenna Phase Center

Unlike GPS receiver antenna, active phased array antenna is selected for SAR antenna.
The phase center of the SAR antenna describes the variation of the phase curve within the
coverage region against a defined origin, here the origin of the antenna coordinate system
18. For two formation satellites of InSAR mission, the same type of SAR antenna should be
selected. As the identical processes of the scheme designing, manufacturing, and testing are
selected for SAR antennas of the same type, theoretically the consistency in configuration
and electric performance of SAR antennas should be well achieved. But factually there
exist the errors during manufacturing, fixing, and deploying of SAR antenna, therefore, the
consistency error of the SAR antenna phase center corresponding to the same beam occurs. It
is mainly caused by two factors:
1 The inconsistency between receiver channels, which is introduced by manufac-

turing process, such as the instrument difference, machining art level, module
assembling level and the work temperature difference, et al.
Table 1: Orbit elements of formation satellites.
Parameters of satellites Satellite 1 Satellite 2

Semimajor axis 6886478 m 6886478 m
Inclination 97.4438◦ 97.4438◦
Eccentricity 0.00117 0.001073
Argument of perigee 90◦ 90◦
Right ascension of ascending node 0◦ 0.01171◦
True anomaly 269.99677◦ 270.00622◦
2 The inconsistency between the locations of apertures, which is mainly caused
by the fixing flatness difference, relative dislocation difference, the deployment
inconsistency of SAR antennas and the configuration distortions caused by different
thermal circumstances, and others.
According to current ability of engineering, the phase inconsistency between T/R
modules at X-band can be constrained to 15◦ 3σ and the inconsistency between the
locations of apertures can be constrained to λ/10 3σ 23 that equals to 36◦ 3σ of
phase inconsistency. Assuming that the number of T/R modules
of an SAR antenna is
N, the synthetic phase consistency error can be constrained to 15◦ 2 36◦ 2 /N
√
39◦ / N 3σ. Hence, the consistency error of two SAR antenna phase center locations can
be constrained to
√
39◦ / N √ λ
◦
· 2 · λ 0.153 · √ 3σ. 3.24
360 N
Take the TanDEM-X mission, for example. Setting N 384, λ 0.032 m, the
consistency error of SAR antenna phase center location can be constrained to 0.25 mm 3σ.
4. Simulations for InSAR Spatial Baseline Determination

4.1. Simulation Settings
The HELIX satellite formation is selected for the simulations and the orbit elements of two
satellites are shown in Table 1. The spaceborne SAR is assumed to work at X-band with a
wavelength of 0.032 m and consist of 384 T/R modules.
The entire simulation consists of GPS measurement simulation and baseline transfor-
mation simulation. The flow chart of GPS observation data simulation is shown in Figure 5.
The International Reference Ionosphere 2007 IRI2007 model is used to simulate ionospheric
delay, Allan variation is used to simulate the clock offset of GPS receiver, and the ARP data,
PCO data 18 and PCVs data of GPS receiver antenna system SEN67-1575-14 CRG are
used to simulate the GPS receiver antenna phase center locations. The PCVs data contains
the mean values and RMS values corresponding to frequency, azimuth, and elevation of
received signal. The attitude data of formation satellite is generated as follows: at first, a
transformation matrix from CIRF to satellite orbit coordinate system is obtained from orbit
data of a formation satellite in CIRF; second, assuming the real Euler angles are 0◦ , that
is, satellite orbit coordinate system and satellite body coordinate system are the same, the
Orbit elements of formation satellite, CODE final orbit and clock product for
orbit dynamical parameters GPS satellites
Orbit integral Lagrange interpolation
Standard positions and velocities of Positions, velocities and

formation satellite clock offsets of GPS satellites
Visibility analysis of GPS satellites
The real distances between visible GPS

satellites and formation satellite
Measurement noises of
GPS receiver antenna
o sets of formation
Phase center data of

GPS receiver clock
formation satellite
Ionospheric delay
Relativistic e ect
code and phase

Attitude data of
satellite
GPS observation data file
Figure 5: Flow chart of GPS observation data simulation.
simulating data of Euler angles are generated by attitude measurement error model list in
Table 2; third, the transformation matrix from satellite orbit coordinate system to satellite
body coordinate system can be obtained by the simulating data of Euler angles; at last, the
attitude quaternion is obtained by the transformation matrix from CIRF to satellite body
coordinate system.
Baseline transformation simulation is the process that the spatial baseline in ITRF is
obtained by mass center data of formation satellites in ITRF, attitude simulation data, and
SAR antenna phase center simulation locations in satellite body coordinate system. The
real SAR antenna phase center simulation location in satellite body coordinate system is
1.2278 m, 1.5876 m, 0.0223 m. The error accuracies and models in the simulations are shown
in Table 2.
4.2. Simulations of Errors Related to GPS Measurement

Each error related to GPS measurement is analyzed by single factor simulation, which is
intended to obtain its impact on relative positioning based on dual-frequency GPS. The
impact of each error is drawn by the comparison residuals between the relative position
Table 2: Error accuracies and modeling descriptions in simulations.
Error type Error accuracy Modeling description

GPS code measurement Gaussian white noise model with mean value of
0.5 m 1σ
noise 0 m and standard deviation of 0.5 m
GPS carrier phase AR2 model with mean value of 0 m and standard
0.002 m 1σ
measurement noise deviation of 0.002 m
Ground calibration error
ARP data, PCO data, and PCVs data mean value
of GPS receiver antenna —
data and RMS data of SEN67-1575-14 CRG
phase center
Fixed bias of 0.005◦
in the yaw, pitch, Gaussian white noise model with mean value of
Attitude measurement
and roll components 0.005◦ and standard deviation of 0.003◦ in the yaw,
error
plus a 0.003◦ 1σ pitch, and roll components
random error
A fixed vector with direction randomly drawn in
Fixing error of GPS
0.5 mm 3σ unit ball and magnitude of 0.5 mm in each satellite
receiver antenna
body coordinate system
A fixed vector with direction randomly drawn in
Consistency error of SAR
0.25 mm 3σ unit ball and magnitude of 0.25 mm in body
antenna phase center
coordinate system of Satellite 1
solutions determined by GPS observation data and relative positions obtained by standard
orbits of formation satellites. The relative position solutions are implemented in the separate
software tools as part of the NUDT Orbit Determination Software 1.0. The GPS observation
data processing consists of GPS observation data preprocessing 24, reduced dynamic
precise orbit determination for single satellite 25, GPS observation data editing 17, 24, and
reduced dynamic precise relative positioning. The RMS values of KBR comparison residuals
of GRACE relative position solution are about 1-2 mm implemented by this software.
4.2.1. Simulations for GPS Carrier Phase Measurement Noise

The noises of GPS carrier phase L1 and L2 measurements are separately simulated by
second-order autoregressive model AR2 as follows
j j j j
eL ti eL ti−1 − 0.67 · eL ti−2 εL ti , 4.1
j j
where εL ti is the noise of carrier phase L measurement for GPS satellite j at epoch ti , εL ti
is the Gaussian white noise.
From the following formula
j j
E eL ti 0 m; σ eL ti 0.002 m, 4.2
where σ· denotes the standard deviation of a random variable, we can get
j j
E εL ti 0 m; σ εL ti 0.0012 m. 4.3
One instance of carrier phase noise simulation is shown in Figure 6.

×10−3 Mean value = 0 m; standard deviation = 0.002 m

8
Carrier phase noise (m)

2
−2
−4
−6
−8
0 2000 4000 6000 8000
Number of epochs
Figure 6: One instance of carrier phase measurement noise simulation.
50 groups of 24 h GPS observation data interval of 30 s for two formation satellites are
simulated by only adding the noises of GPS carrier phase L1 and L2 measurements. By the
processing of relative positioning, the mean RMS values of comparison residuals of relative
position solutions in ITRF Figure 7 are 0.340 mm of x-axis, 0.333 mm of y-axis, 0.288 mm of
z-axis, and 0.560 mm of 3 dimensions. It is shown by simulation results that the GPS carrier
phase noise can be well smoothed by reduced dynamic relative positioning approach.
4.2.2. Simulations for Ground Calibration Error of GPS Receiver Antenna Phase Center
Ground calibration error of GPS receiver antenna phase center is mainly caused by PCVs.
The PCVs values are described by the mean value and RMS value corresponding to the
direction of received signal. The PCV value corresponding to the direction of received signal
is simulated by Gaussian white noise with mean value and RMS value obtained from ground
calibration result of GPS receiver antenna system SEN67-1575-14 CRG.
The GPS observation data are simulated only considered ground calibration error of
GPS receiver antenna phase center. By the precise orbit determination for single satellite, the
mean RMS values of comparison residuals of orbit solutions in ITRF are 4.018 mm of x-axis,
4.154 mm of y-axis, 2.427 mm of z-axis, and 6.269 mm of 3 dimensions. The impacts of PCVs
on single satellite orbit solutions are mainly made by the mean value part of PCVs, while the
impacts of RMS part in ITRF are only 0.119 mm of x-axis, 0.094 mm of y-axis, 0.116 mm of
z-axis, and 0.191 mm of 3 dimensions, and the RMS part of PCVs can nearly be smoothed.
By the processing of relative positioning, the mean RMS values of comparison residuals of
relative position solutions in ITRF are 0.067 mm of x-axis, 0.070 mm of y-axis, 0.056 mm of z-
axis, and 0.112 mm of 3-dimensions. As the nearly equal models of ground calibration errors
of GPS receiver antenna phase centers for two formation satellites are selected and the LOS
vectors are nearly the same for close satellite formation, the impacts of mean value part of
PCVs can nearly be cancelled out by differential GPS observation and impacts of RMS part
can be well smoothed by the constraints of orbit dynamical models. It is shown by the results
Average RMS value = 0.34 mm

0.5
x (mm)
0
0 10 20 30 40 50
Number of simulations
a
0.5
y (mm)
0
0 10 20 30 40 50
b
0.5
z (mm)
0
0 10 20 30 40 50
c
Figure 7: Simulation results of GPS carrier phase measurement noise for relative positioning.
of single satellite orbit solutions and relative position solutions that the characters of GPS
receiver antenna phase centers onboard two formation satellites must have great consistency.
4.2.3. Simulations of Satellite Attitude Measurement Error for GPS Relative Positioning
The Euler angle errors are simulated by Gaussian white noise with 0.005◦ of mean value and
0.003◦ of standard deviation, and 50 groups of 24 h GPS observation data for two formation
satellites are simulated by only adding the attitude measurement errors. The mean RMS
values of comparison residuals of relative position solutions in ITRF Figure 8 are 0.069 mm
of x-axis, 0.075 mm of y-axis, 0.081 mm of z-axis, and 0.128 mm of 3 dimensions. The 3
dimensional maximum of comparison residuals in these 50 simulations is 0.219 mm, which is
less than 0.47 mm and is well consistent with aforementioned analysis in Section 3.1.3.
4.2.4. Simulations for Fixing Error of GPS Receiver Antenna

The fixing error of GPS receiver antenna belongs to systematic error and it is a fixed
bias vector in satellite body coordinate system. At first, four representatively “extreme”
circumstances of fixing errors of GPS receiver antennas onboard two formation satellites
are simulated. The so-called “extreme” circumstance is that the directions of two fixed bias
vectors are opposite. Four representatively “extreme” circumstances of fixing errors of GPS
receiver antennas here are directions along X-axis, Y -axis, Z-axis, and diagonal of X-axis,
0.1
x (mm)
0.05
0
0 10 20 30 40 50
a
0.2
y (mm)
0.1
0
0 10 20 30 40 50
b
0.2
z (mm)
0.1
0
0 10 20 30 40 50
c
Figure 8: Simulation results of satellite attitude measurement error for relative positioning.
Table 3: Relative positioning results for four representatively “extreme” circumstances of fixing errors of
GPS receiver antenna.
x/mm y/mm z/mm 3 dimension/mm

X-axis 0.512 0.499 0.701 1.002
Y-axis 0.677 0.694 0.133 0.979
Z-axis 0.072 0.083 0.081 0.136
Diagonal 0.502 0.503 0.429 0.830
Y -axis, Z-axis in satellite body coordinate system, respectively. All the magnitudes of fixed
bias vectors are selected 0.5 mm. 24 h GPS observation data for two formation satellites are
simulated by only considering the four representatively “extreme” circumstances of fixing
errors of GPS receiver antennas. The results of relative positioning are shown in Table 3.
From Table 3, it is shown that the fixing errors of GPS receiver antenna along X-axis
and Y -axis will mainly be absorbed by relative position solutions and the impact can reach
to 1 mm, but the error along Z-axis can be smoothed by the constraints of orbit dynamical
models.
In practice, the occurrence of “extreme” circumstances is extremely low and they are
just analyzed as the ultimate circumstances. For multiple repeated satellite missions, the
fixing error of GPS receiver antenna is a random error. So this error can be simulated as a
fixed vector with direction randomly drawn from unit ball and magnitude of 0.5 mm in each
satellite body coordinate system. 50 groups of 24 h GPS observation data for two formation
satellites are simulated by only adding the simulations of fixing error of GPS receiver

1
x (mm)
0.5
0
0 10 20 30 40 50
a
1
y (mm)
0.5
0
0 10 20 30 40 50
b

1
z (mm)
0.5
0
0 10 20 30 40 50
c
Figure 9: Simulation results of fixing error of GPS receiver antenna for relative positioning.
antenna. The mean RMS values of comparison residuals of relative position solutions in ITRF
Figure 9 are 0.295 mm of x-axis, 0.294 mm of y-axis, 0.249 mm of z-axis, and 0.495 mm of 3
dimensions.
From aforementioned simulations of each error related to GPS measurement, it is
shown that the impacts of GPS carrier phase measurement noise and fixing error of GPS
receiver antenna on GPS relative positioning are much bigger than other errors related to GPS
measurement and these two errors are the main factors of errors related to GPS measurement.
4.3. Simulations of Errors Related to Baseline Transformation

In this section, the impact of each error on baseline transformation is obtained by single factor
simulation. Each impact is given by the comparison between the spatial baseline solutions
obtained with and without errors.
4.3.1. Simulations of Satellite Attitude Measurement Error for Baseline Transformation

The satellite attitude simulation data used here are the same as Section 4.2.3. By baseline
transformation with attitude simulation data, the mean RMS values of comparison residuals
of spatial baseline solutions in ITRF Figure 10 are 0.115 mm of x-axis, 0.115 mm of y-
axis, 0.133 mm of z-axis, and 0.210 mm of 3 dimensions. The 3 dimensional maximum of

0.2
x (mm)
0.1
0
0 10 20 30 40 50
a
0.2
y (mm)
0.1
0
0 10 20 30 40 50
b
0.2
z (mm)
0.1
0
0 10 20 30 40 50
c
Figure 10: Simulation results of satellite attitude measurement error for baseline transformation.
comparison residuals in these 50 simulations is 0.213 mm, which is less than 0.50 mm and is
consistent with aforementioned analysis in Section 3.2.1.
4.3.2. Simulations for Consistency Error of SAR Antenna Phase Center

It is shown by the analysis in Section 3.2.2 that the accuracy of consistency error of SAR
antenna phase center is better than 0.25 mm 3σ in current simulation circumstances. This
error is only added to the SAR antenna phase center of Satellite 1 and can be simulated as
a fixed vector with direction randomly drawn from unit ball and magnitude of 0.25 mm in
body coordinate system of satellite 1. By 50 groups of simulations, the mean RMS values of
comparison residuals of spatial baseline solutions in ITRF Figure 11 are 0.142 mm of x-axis,
0.142 mm of y-axis, and 0.153 mm of z-axis.
4.4. Simulations of Total Error Sources

In this section, all the errors are added to the flow of spatial baseline determination
simulations according to the error models listed in Table 2. By 50 groups of total error sources
simulations, the mean RMS values of comparison residuals of spatial baseline solutions in
ITRF Figure 12 are 0.500 mm of x-axis, 0.500 mm of y-axis, 0.452 mm of z-axis, and 0.845 mm
of 3 dimensions.

0.2
x (mm)
0.1
0
0 10 20 30 40 50
a
0.2
y (mm)
0.1
0
0 10 20 30 40 50
b
0.2
z (mm)
0.1
0
0 10 20 30 40 50
c
Figure 11: Simulation results of consistency error of SAR antenna phase center for baseline transformation.

1
x (mm)
0.5
0
0 10 20 30 40 50
a
1
y (mm)
0.5
0
0 10 20 30 40 50
b
1
z (mm)
0.5
0
0 10 20 30 40 50
c
Figure 12: Simulation results of total error sources for spatial baseline determination.

1
x (mm)
0.5
0
0 10 20 30 40 50
a
1
y (mm)
0.5
0
0 10 20 30 40 50
b
1
z (mm)
0.5
0
0 10 20 30 40 50
c
Figure 13: Simulation results of total errors related to GPS measurement for relative positioning.
In addition, the impact of total errors related to GPS measurement on GPS relative
positioning in ITRF Figure 13 is 0.454 mm of x-axis, 0.452 mm of y-axis, 0.388 mm of z-
axis, and 0.755 mm of 3 dimensions, and the impact of total errors related to baseline
transformation in ITRF Figure 14 is 0.185 mm of x-axis, 0.185 mm of y-axis, 0.206 mm of
z-axis, and 0.334 mm of 3 dimensions.
It is shown by the simulations of total error sources that errors related to GPS
measurement are the main error sources for the spatial baseline determination and 1 mm
level InSAR spatial baseline determination can be realized according to current simulation
conditions.
5. Conclusions
In this paper, the errors introduced by spatial baseline measurement for InSAR mission are
deeply studied. The impacts of errors on spatial baseline determination are analyzed by single
factor simulations and total error sources simulations. The main conclusions are drawn as
follows.
1 The spatial baseline measurement errors can be classified into two groups: errors
related to GPS measurement and errors related to baseline transformation. By
simulations, the three-dimensional impacts of these errors on spatial baseline
determination in ITRF are 0.755 mm and 0.334 mm, respectively. It is shown that

0.4
x (mm)
0.2
0
0 10 20 30 40 50
a
0.4
y (mm)
0.2
0
0 10 20 30 40 50
b
0.4
z (mm)
0.2
0
0 10 20 30 40 50
c
Figure 14: Simulation results of total errors related to baseline transformation for baseline transformation.
the errors related to GPS measurement are the main influence on spatial baseline
determination.
2 By the results of single factor simulations, the three dimensional impacts of GPS
carrier phase measurement noise and the fixing error of GPS receiver antenna on
GPS relative positioning in ITRF are 0.560 mm and 0.495 mm, respectively. These
two errors are the main factors of errors related to GPS measurement.
3 It is shown by total error sources simulations that the impact of all the errors on
spatial baseline determination in ITRF is 0.500 mm of x-axis, 0.500 mm of y-axis,
0.452 mm of z-axis, and 0.845 mm of 3 dimensions. Therefore, 1 mm level InSAR
spatial baseline determination can be realized.
Acknowledgments
The mean antenna phase center description for the Sensor Systems SEN67157514 antenna has
been contributed by the German Space Operations Center GSOC, Deutsches Zentrum für
Luft- und Raumfahrt DLR, Wessling, to enable the simulation of antenna phase center data
of dual-frequency GPS receiver. Precise GPS ephemerides for use within this study have been
obtained from the Center for Orbit Determination in Europe at the Astronomical Institute of
the University of Bern AIUB. The authors extend special thanks to the support of the above
institutions. This paper is supported by the National Natural Science Foundation of China
Grant no. 61002033 and no. 60902089 and Open Research Fund of State Key Laboratory of
Astronautic Dynamics of China Grant no. 2011ADL-DW0103.
References
1 G. Krieger, I. Hajnsek, K. P. Papathanassiou, M. Younis, and A. Moreira, “Interferometric syn-
thetic aperture radar SAR missions employing formation flying,” Proceedings of the IEEE, vol. 98,
no. 5, pp. 816–843, 2010.
2 M. L. Jiao, “A review on latest Interferometric Synthetic Aperture Radar researches,” in WRI World
Congress on Software Engineering (WCSE ’09), pp. 387–390, May 2009.
3 W. Wang, “Optimal baseline design and error compensation for bistatic spaceborne InSAR,” in Pro-
ceedings of Fringe Workshop, November-December 2005.
4 G. Krieger, A. Moreira, H. Fiedler et al., “TanDEM-X: a satellite formation for high-resolution SAR
interferometry,” IEEE Transactions on Geoscience and Remote Sensing, vol. 45, no. 11, pp. 3317–3340,
2007.
5 M. Zink, H. Fiedler, I. Hajnsek, G. Krieger, A. Moreira, and M. Werner, “The TanDEM-X mission
concept,” in IEEE International Geoscience and Remote Sensing Symposium (IGARSS ’06), pp. 1938–1941,
August 2006.
6 R. Werninghaus and S. Buckreuss, “The TerraSAR-X mission and system design,” IEEE Transactions
on Geoscience and Remote Sensing, vol. 48, no. 2, pp. 606–614, 2010.
7 M. Wermuth, O. Montenbruck, and A. Wendleder, “Relative navigation for the TanDEM-X mission
and evaluation with DEM calibration results,” in the 22nd International Symposium on Space Flight
Dynamics, Sao Jose dos Campos, Brazil, 2011.
8 H. Xu, Y. Zhou, and C. Li, “Analysis and simulation of spaceborne SAR interferometric baseline,” in
Proceedings of the CIE International Conference on Radar, pp. 639–643, Beijing, China, October 2001.
9 R. Kroes, O. Montenbruck, W. Bertiger, and P. Visser, “Precise GRACE baseline determination using
GPS,” GPS Solutions, vol. 9, no. 1, pp. 21–31, 2005.
10 O. Montenbruck, P. W. L. van Barneveld, Y. Yoon, and P. N. A. M. Visser, “GPS-based precision
baseline reconstruction for the TanDEM-X SAR-formation,” in the 20th International Symposium on
Space Flight Dynamics, pp. 24–28, 2007.
11 S. D’Amico and O. Montenbruck, “Differential GPS: an enabling technology for formation flying
satellites,” in the 7th IAA Symposium on Small Satellites for Earth Observation, pp. 457–464, 2009.
12 O. Montenbruck and E. Gill, Satellite Orbits: Models, Methods and Applications, Springer, Heidelberg,
Germany, 2000.
13 P. W. Binning, Absolute and relative satellite to satellite navigation using GPS, Ph.D. dissertation, Univer-
sity of Colorado, 1997.
14 P. J. G. Teunissen, “The least-squares ambiguity decorrelation adjustment: a method for fast GPS
integer ambiguity estimation,” Journal of Geodesy, vol. 70, no. 1-2, pp. 65–82, 1995.
15 P. J. G. Teunissen, P. J. De Jonge, and C. C. J. M. Tiberius, “The least-squares ambiguity decorrelation
adjustment: its performance on short GPS baselines and short observation spans,” Journal of Geodesy,
vol. 71, no. 10, pp. 589–602, 1997.
16 J. Kouba, A guide using International GPS Service (IGS) products, Jet Propulsion Laboratory, Pasadena,
Calif, USA, 2009.
17 R. Kroes, Precise relative positioning of formation flying spacecraft using GPS, Ph.D. dissertation, Delft
University of Technology, The Netherlands, 2006.
18 O. Montenbruck, M. Garcia-Fernandez, Y. Yoon, S. Schön, and A. Jäggi, “Antenna phase center cali-
bration for precise positioning of LEO satellites,” GPS Solutions, vol. 13, no. 1, pp. 23–34, 2009.
19 A. Jäggi, R. Dach, O. Montenbruck, U. Hugentobler, H. Bock, and G. Beutler, “Phase center modeling
for LEO GPS receiver antennas and its impact on precise orbit determination,” Journal of Geodesy, vol.
83, no. 12, pp. 1145–1162, 2009.
20 M. Garcia and O. Montenbruck, “TerraSAR-X/TanDEM-X GPS antenna phase center analysis and
results,” German Space Operations Center, Germany, 2007.
21 J. H. Gonzalez, M. Bachmann, G. Krieger, and H. Fiedler, “Development of the TanDEM-X calibration
concept: analysis of systematic errors,” IEEE Transactions on Geoscience and Remote Sensing, vol. 48, no.
2, pp. 716–726, 2010.
22 D. D. McCarthy, “IERS conventions,” IERS Technical Note 21, Observatoire de Paris, Paris, France, pp.
20–39, 1996.
23 W. T. Wang and Y. H. Qi, “A new technique to compensate for error in SAR antenna power pattern,”
Chinese Space Science and Technology, vol. 17, no. 3, pp. 65–70, 1997.
24 D. F. Gu, The spatial states measurement and estimation of distributed InSAR satellite system, Ph.D. disser-
tation, National University of Defense Technology, China, 2009.
25 O. Montenbruck, T. Van Helleputte, R. Kroes, and E. Gill, “Reduced dynamic orbit determination
using GPS code and carrier measurements,” Aerospace Science and Technology, vol. 9, no. 3, pp. 261–
271, 2005.
doi:10.1155/2012/596396
Research Article
Analysis of Attitude Determination Methods Using
GPS Carrier Phase Measurements
Leandro Baroni1 and Hélio Koiti Kuga2

1
Institute of Science, Engineering and Technology, Federal University of Jequitinhonha and
Mucuri Valleys (UFVJM), Rua do Cruzeiro, 1, 39803-371 Teófilo Otoni, MG, Brazil
2
Space Mechanics and Control Division (DMC), National Institute for Space Research (INPE),
12227-010 São José dos Campos, SP, Brazil
Correspondence should be addressed to Hélio Koiti Kuga, hkakinha@hotmail.com
Copyright q 2012 L. Baroni and H. K. Kuga. This is an open access article distributed under
If three or more GPS antennas are mounted properly on a platform and differences of GPS
signals measurements are collected simultaneously, the baselines vectors between antennas can
be determined and the platform orientation defined by these vectors can be calculated. Thus, the
prerequisite for attitude determination technique based on GPS is to calculate baselines between
antennas to millimeter level of accuracy. For accurate attitude solutions to be attained, carrier
phase double differences are used as main type of measurements. The use of carrier phase
measurements leads to the problem of precise determination of the ambiguous integer number
of cycles in the initial carrier phase integer ambiguity. In this work two algorithms LSAST and
LAMBDA were implemented and tested for ambiguity resolution allowing accurate real-time
attitude determination using measurements given by GPS receivers in coupled form. Platform
orientation was obtained using quaternions formulation, and the results showed that LSAST
method performance is similar to LAMBDA as far as the number of epochs which are necessary
to resolve ambiguities is concerned, but with processing time significantly higher. The final result
accuracy was similar for both methods, better than 0.1◦ to 0.2◦ , when baselines are considered in
decoupled form.
1. Introduction
Global Navigation Satellite Systems GNSSs are satellite-based radionavigation systems,
providing to worldwide users precise position and timing. System satellites transmit radio-
frequency signals containing information required for the user equipment to compute its
navigation solution position, velocity, and time. GNSS can also be used to determine
attitude of a platform, in which three or more antennas are needed to calculate attitude
parameters 1.
If three or more GNSS antennas properly mounted on a platform and differences of

GNSS signals measurements are collected simultaneously, baselines vectors formed between
antennas can be determined, and orientation of the platform defined by these vectors can be
calculated. Thus, the prerequisite for the attitude determination technique based on GNSS
systems is to calculate the baselines between the antennas.
Accurate attitude solutions can be obtained using carrier phase double difference
observables as the main type of measurements, including all independent combinations
of antenna positions. Baselines between antennas must be determined in millimeter level
of accuracy. Typically, the distance between the antennas is a few meters or less, and all
spatially correlated errors between the antennas are almost eliminated in differencing single
and double process, including orbital, ionospheric, and tropospheric errors. Therefore, main
error sources affecting attitude determination are the multipath, receiver internal noise, and
antenna phase center variation 2.
The use of carrier phase measurements leads to the problem of determining precisely
the ambiguous initial carrier phase integer number of cycles integer ambiguity. To increase
confidence and accelerate the process by limiting the search space, three types of restrictions
can be established from prior knowledge of the antenna fixed geometry: i length of the
baseline, ii angle between baselines, and iii knowledge of the geometry of the antennas
as a network in which double difference ambiguities must satisfy a closed loop condition 2.
Recently, several studies have focused on increasing the success rate of resolving ambiguity
process using the restrictions, as in 3, 4. References 5, 6 show phase measurements
usage and performance of solving integer ambiguity methods in positioning applications.
7 Reference develops a procedure to determine satellite attitude in three axes with GPS
associated with a gyro, and resolving ambiguities with the method described in 8. Thus,
the resolving ambiguity process is an important step to determine an accurate baseline vector,
resulting in an accurate attitude determination.
A common attitude representation is done by Euler angles. This parameterization
has difficult computation in general, because of the use of trigonometric functions
and the appearance of a singularity in the motion modeling. Another way to attitude
parameterization is using quaternions. This parameterization has some advantages over
Euler angles; it is computationally efficient, there is no singularity, and it does not depend
on trigonometric functions 9.
So, in this work an implementation and analysis of algorithms for integer ambiguity
resolution allowing accurate attitude determination in real time, using measurements
provided by GPS receivers, will be tested. Algorithm tests, using quaternions for attitude
representation, will be implemented with real data, collected at INPE and described in 7. In
this experiment, three antennas were fixed on a structure with known baseline lengths 1 m
and angle between baselines of 90◦ . The tests were executed with the methods LAMBDA and
LSAST for ambiguity resolution.
least-squares ambiguity decorrelation adjustment LAMBDA method is a procedure
for integer ambiguity estimation in carrier phase measurements. This method executes the
integer ambiguity estimation through a Z transform, in which ambiguities are decorrelated
before the integer values search process. Then, a minimization problem is approached as
a discrete search inside an ellipsoidal region defined by decorrelated ambiguities, which is
smaller than original ones. As a result, integer least-squares estimates for the ambiguities are
obtained. This method was introduced in 10, 11. References 12, 13 show computational
implementation aspects and ambiguity search space reducing performance.
Least-Squares Ambiguity Solution Technique LSAST method, also known as LSAST

method, was proposed in 14. This method involves a modified sequential least-squares
technique, in which ambiguity parameters are divided into two groups: primary ambiguities
typically three double difference ambiguities, and the secondary ambiguities. Only the
primary ambiguities are fully searched, in ±5 cycles around the corresponding float
ambiguity, after rounded to the nearest integer. For each set of the primary ambiguities, there
is a unique set of secondary ambiguities. Therefore, the search dimension is smaller and the
computation time is significantly shorter than the full search approach. The choice of primary
group measurements is based on GDOP value. geometric dilution of precision GDOP is a
quantity which measures the influence of satellite geometry on positioning errors. Satellites
with low GDOP will lead to a search with less-potential solutions. However, GDOP cannot
be very low, in order to avoid the position uncertainty including more than one solution for
secondary group measurements. The procedure is to choose primary group of satellites which
have a reasonable GDOP.
2. Attitude Representation
Euler angles are a means of representing the spatial orientation of any coordinate system
as a composition of rotations from a frame of reference. These angles uniquely determine
the orientation of a rigid body in three-dimensional space. There are several conventions for
defining the Euler angles, depending on the choice of axes and the order in which rotations
about these axes are performed. A matrix expression can be found for any frame given its
Euler angles, performing three rotations in sequence. Here, Euler angles are denoted as θ is
pitch, φ is roll, and ψ is yaw. The resulting rotation matrix is given in convention

R θ, ψ, φ Rx φ Ry θRz ψ . 2.1
Expanding the rotation matrix,

⎡ ⎤
cos θ cos ψ cos θ sin ψ − sin θ
R ⎣sin φ sin θ cos ψ − cos φ sin ψ sin φ sin θ sin ψ cos φ cos ψ cos θ sin φ ⎦. 2.2
cos φ sin θ cos ψ sin φ sin ψ cos φ sin θ sin ψ − sin φ cos ψ cos θ cos φ
9 and is
Quaternion q is determined in function of rotation angle ϕ and rotation axis n
composed by a scalar term q4 and a vector term q 15
> ?T > ?T
q q1 q2 q3 q4 qT q4 , 2.3
with
> ?T ϕ ϕ
q q1 q2 q3 sin 9,
n q4 cos . 2.4
2 2
Rotation matrix is given in terms of quaternions as

% % % %
Rq %q4 % − %q% I3 2
2 2
qqT 2q4
q, 2.5
where I3 is an third-order identity matrix, and

⎡ ⎤
0 q3 −q2
q ⎣−q3 0 q1 ⎦.
2.6
q2 −q1 0
Or explicitly,
⎡ 2 ⎤
q1− q22 − q32 q42 2 q1 q2 q3 q4 2q1 q3 − q2 q4
Rq ⎣ 2q1 q2 − q3 q4 −q12 q22 − q32 q42 2 q2 q3 q1 q4 ⎦. 2.7
2 q1 q3 q2 q4 2 q2 q3 − q1 q4 −q12 − q22 q32 q42
Euler angles can be calculated from rotation matrix as

⎡ ⎤ ⎡ > ?⎤
φ atan2 2 q2 q3 >q1q4 , −q21 − ?
q22 q32 q42
⎣θ ⎦ ⎣ ⎦
> asin 2 q2 q4 − q21 q3 2 ? . 2.8
ψ atan2 2 q1 q2 q3 q4 , q1 − q2 − q3 q4
2 2
3. Ambiguity Resolution
When distance between receivers is short until 10 km, ionospheric and tropospheric
residuals are small compared to multipath and internal receiver noise errors. Thus, for a short
ij
baseline, carrier phase double-difference measurements φub are 16
ij
j
ij ij
φub 1ib − 1b · xub λNub ub,φ , 3.1
ij
where 1ib is unit vector pointing from base to satellite i, xub is baseline vector, Nub is the
integer ambiguity, and is a vector representing unmodeled errors. If m satellites were in
view simultaneously, then there are m − 1 double difference measurements. In matrix form,
taking M as master satellite, 3.1 becomes
⎡ ⎤ ⎡ 1 ⎤ ⎡ ⎤
1M
φub 1b − 1M b
Nub 1M
⎢ .. ⎥ ⎢ .. ⎥ ⎢ .. ⎥
⎢ ⎥⎢ ⎥xub λ⎢ ⎥ , 3.2
⎣ . ⎦ ⎣ . ⎦ ⎣ . ⎦
m−1M m−1 m−1M
φub 1b − 1M
b
N ub
or
y Hxub λN . 3.3
Equation 3.3 is the measurement model considered for ambiguity resolution. Results were
obtained for ambiguity resolution from the same estimation process using a Kalman filter,
and processing code and carrier phase measurements. Ambiguity resolution is done epoch to
epoch, applying LAMBDA and LSAST methods on real-valued float ambiguities given by
the Kalman filter.
Nort
h
Ant
enna
2
Antenn
a3
Antenna East
4
Figure 1: Antenna configuration in data acquisition.
4. Results
Data used in this study were originally collected for use in 7 on March 30th, 2005. Three
GPS receivers were used, and their antennas were fixed on a frame allowing the knowledge
of the baseline lengths 1 m and the angle between them 90◦ . The sampling rate was
1 Hz, collected only in L1 carrier frequency. Figure 1 illustrates that the antenna mounting
configuration and equipments used are listed below:
i 3 AllStar CMC GPS receivers Canadian Marconi Space Company and
ii 3 AllStar CMC model AT 575-70 GPS antennas.
GPS measurements were processed to obtain a precise baseline length. Ambiguity

resolution was made using LAMBDA and LSAST methods. Each baseline was independently
determined to form a frame, whose attitude is calculated referred to the east-north-up
reference system. Rotation matrix, obtained by quaternions, is transformed to Euler angles
using 2.8, in order to give a geometric view of orientation. All algorithms were implemented
in Matlab language, using a computer with 4 Gb RAM, Intel Core i3 processor, and Windows
7 operating system. Three data sets were tested.
Data Set 1
In this test, same 8 satellites were kept in view SV04, SV08, SV13, SV19, SV23, SV27, SV28 e
SV31, resulting in 7 double difference measurements. SV13 was chosen as master satellite,
because of its high elevation. Ambiguity resolution methods use a Kalman filter to obtain
real-valued float ambiguities. Kalman filter needs some epochs to converge to an ambiguity
solution, and thus to a given baseline. Data were free of cycle slips.
In this way, LAMBDA method presents a solution with correct values of ambiguities
after 107 epochs, while LSAST method takes 157 epochs to reach the correct values. Graphs
in Figure 2 show Euler angles values after the correct solution is reached.
Table 1 shows the statistics for the angle values in each method after ambiguities
are correctly solved. These results show that baseline orientation could be determined with
accuracy better than 0.1◦ in both methods although LSAST method has taken longer to deliver
the correct solution.
LSAST method sweeps a fixed number of cycles in satellite primary set, leading to
a slower search. This method had a mean processing time of 140 ± 32 ms for resolving 7
ambiguities. Mean processing time of LAMBDA method was 11 ± 1 ms after ambiguities are
4 4
2 2
0 0
Angle (◦ )
Angle (◦ )
−2 −2
−4 −4
−6 −6
−8 −8
100 125 150 175 100 125 150 175
Time (s) Time (s)
Yaw Yaw
Pitch Pitch
Roll Roll
a LAMBDA b LSAST
Figure 2: Euler angles for data set 1.
4 4
2 2
0
Angle (◦ )
0
Angle (◦ )
−2 −2
−4 −4
−6 −6
−8 −8
100 150 200 250 300 350 100 150 200 250 300 350
Time (s) Time (s)
Yaw Yaw
Pitch Pitch
Roll Roll
a LAMBDA b LSAST
solved due to small search space of candidate ambiguities. Considering whole set, processing
time was 18 ± 25 ms because search space was larger. Processing time is shown on Figure 5a.
Data Set 2
In this set, 7 satellites were visible SV08, SV13, SV19, SV23, SV27, SV28, and SV31, resulting
in 6 double difference measurements. SV13 was taken as master satellite. Data were free of
cycle slips.
Graphs in Figure 3 show platform roll, pitch, and yaw angles. For this set, both
LAMBDA and LSAST methods converged to correct ambiguity values after 141 epochs.
Table 2 shows mean and standard deviation values for Euler angles after a stable solution
is reached. The accuracy is very similar in both methods, once number of epochs necessary
for getting ambiguity solution is the same.
4 4
2 2
0 0
Angle (◦ )
Angle (◦ )
−2 −2
−4 −4
−6 −6
−8 −8
300 450 600 750 900 1050 1200 300 450 600 750 900 1050 1200
Time (s) Time (s)
Yaw Yaw
Pitch Pitch
Roll Roll
a LAMBDA b LSAST
0.2 0.35
0.3
Processing time (s)
Processing time (s)
0.15
0.25
0.2
0.1
0.15
0.05 0.1
0.05
0 0
100 125 150 175 100 150 200 250 300 350
Time (s) Time (s)
LAMBDA LAMBDA
Hatch Hatch
a Data set 1 b Data set 2
0.4
0.35
Processing time (s)
0.3
0.25
0.2
0.15
0.1
0.05
0
300 450 600 750 900 1050 1200
Time (s)
LAMBDA
Hatch
c Data set 3
Figure 5: Processing time for all data sets.

Table 1: Mean and standard-deviation SD for data set 1.
LAMBDA LSAST
◦
Angle Mean SD Mean SD
Roll 0.136 0.132 0.097 0.014
Pitch 0.482 0.060 0.520 0.028
Yaw −6.488 0.045 −6.478 0.001
Table 2: Mean and standard-deviation SD for data set 2.
LAMBDA LSAST
◦
Angle Mean SD Mean SD
Roll 1.839 0.190 1.839 0.190
Pitch −0.820 0.079 −0.820 0.079
Yaw −5.608 0.056 −5.608 0.056
Table 3: Mean and standard deviation SD for data set 3.
LAMBDA LSAST
Angle ◦ Mean SD Mean SD
Roll 1.049 0.306 1.049 0.306
Pitch 0.557 0.116 0.557 0.116
Yaw −5.612 0.178 −5.612 0.178
In this test, mean processing time was 11 ± 17 ms for LAMBDA method in whole set,
and 8 ± 1 ms after reaching correct solution. For LSAST, processing time was 154 ± 55 ms to
solve 6 ambiguities Figure 5b.
Data Set 3
This data set had the same visible satellites as Data Set 2. Both LAMBDA and LSAST
methods present solutions for ambiguities after 320 epochs. Graphs in Figure 4 show the
Euler angles obtained by both ambiguities resolution methods. Table 3 shows mean and
standard deviation values for roll, pitch, and yaw angles.
The average processing time for ambiguity resolution step, whenever a successful
solution is obtained, is 7 ± 2 ms, and whole set processing time was 9 ± 11 ms with LAMBDA
method, and 137 ± 39 ms with LSAST method. Processing time in LSAST method increases
along the time due to changing satellite geometry Figure 5c.
5. Conclusions
Due to short baseline 1 m, both methods have a tendency to solve ambiguities for the same
number of epochs, as filter converges. Both LAMBDA and LSAST methods had a similar
performance in number of epochs needed to give the correct solution; however, processing
time of LSAST method was significantly longer. This is due to the fact that LSAST method
performs a systematic search throughout the primary set of measurements, while LAMBDA
optimizes the search space as a whole. Processing time for LAMBDA method is search space
size dependant 13. The smaller search space, the shorter processing time. This can be
verified by mean time to process the whole set compared to partial solution.
The accuracy of final result is also similar for both methods, better than 0.1◦ to 0.2◦ ,
once they find the same ambiguity set. The small variations are due to difference in number
of epochs to obtain the solution. Data Set 2 and Data Set 3 showed same mean values and
standard deviation for Euler angles. This occurs because the only difference in algorithms is
resolution ambiguity routine. Once ambiguities were resolved to the same values, statistics
must be equal. LAMBDA method also has an extensive series of studies documented in the
literature.
Attitude estimation using quaternions led to the same values when using a rotation
matrix based on Euler angles, as results showed in 17. The main difference is processing
time, slightly longer with quaternions. This is probably due to conversion of results to Euler
angles, which uses trigonometric functions. However, quaternions can be useful to avoid
singularity situations.
Using dual frequency measurements is not necessary for the mitigation of ionospheric
errors since the baselines are short but can be useful for a faster ambiguity resolution, for
example, using widelane technique. If compared with the use of GPS alone, measurements
from two different GNSS systems can certainly improve the attitude determining accuracy
as result of increased number of measurements and improved satellite geometry. This can be
accomplished with the use of GPS/GLONASS and future GPS/Galileo receivers.
References
1 B.W. Parkinson and J.J. Spilker Jr., Global Positioning System: Theory and Applications, vol. 1-2, 1996.
2 A. El-Mowafy and A. Mohamed, “Attitude determination from GNSS using adaptive Kalman
filtering,” Journal of Navigation, vol. 58, no. 1, pp. 135–148, 2005.
3 J. Pinchin, “Enhanced integer bootstrapping for single frequency GPS attitude determination,” in
Proceedings of the 21st International Technical Meeting of the Satellite Division of the Institute of Navigation
(ION GNSS ’08), pp. 1524–1532, Savannah, Ga, USA, September 2008.
4 G. Zheng and D. Gebre-Egziabher, “Enhancing ambiguity resolution performance using attitude
determination constraints,” in Proceedings of the 22nd International Technical Meeting of the Satellite
Division of the Institute of Navigation (ION GNSS ’09), pp. 3598–3610, Savannah, Ga, USA, September
2009.
5 L. Baroni, H. K. Kuga, and K. O’Keefe, “Analysis of three ambiguity resolution methods for real time
static and kinematic positioning of a GPS receiver,” in Proceedings of the 22nd International Technical
Meeting of the Satellite Division of the Institute of Navigation (ION GNSS ’09), pp. 2392–2400, Savannah,
GA, USA, September 2009.
6 L. Baroni., Algoritmos de navegação em tempo real para um sistema GPS de posicionamento relativo de precisão
(Real time navigation algorithms for a precise GPS relative positioning system), Ph.D. thesis, INPE, São José
dos Campos, Brazil, 2009, INPE-16598-TDI/1584.
7 A. C. Louro, Determinação autônoma de atitude de satélites utilizando GPS (Autonomous satellite attitude
determination using GPS), Ph.D. thesis, INPE, São José dos Campos, Brazil, 2006, INPE-14091-
TDI/1074.
8 R. V. F. Lopes and H. K. Kuga, “Determinação de atitude em 3 eixos por interferometria GPS Three-
axis attitude determination by GPS interferometry,” in Plataforma Integrada Sensores Inerciais/GPS
(Inertial sensor/GPS integrated platform), O. S. C. Durão, V. R. Schad, H. K. Kuga, R. V. F. Lopes, H. C.
Carvalho, and M. Esper, Eds., INPE, São José dos Campos, Brazil, 2002, INPE-9293-PRP/234.
9 R. V. Garcia, Filtro não linear de Kalman sigma-ponto com algoritmo unscented aplicado a estimativa dinâmica
da atitude de satélites artificiais (Sigma point nonlinear Kalman filter with unscented algorithm applied to
attitude dynamics estimation of artificial satellites), Ph.D. thesis, INPE, São José dos Campos, Brazil, 2011,
10.16.14.32-TDI.
10 P. J. G. Teunissen, “Least-squares estimation of the integer GPS ambiguities,” in Proceedings of the IAG
General Meeting (IAG ’93), Beijing, China, 1993.
11 P. J. G. Teunissen, “New method for fast carrier phase ambiguity estimation,” in Proceedings of the
IEEE Position Location and Navigation Symposium, pp. 562–573, Las Vegas, Mev, USA, April 1994.
12 P. J. de Jonge and C. C. J. M. Tiberius, “The LAMBDA method for integer ambiguity estimation:
implementation aspects,” Tech. Rep., Delft University of Technology, 1996.
13 P. J. de Jonge, C. C. J. M. Tiberius, and P. J. G. Teunissen, “Computational aspects of the LAMBDA
method for GPS ambiguity resolution,” in Proceedings of the 9th International Technical Meeting of the
Satellite Division of the Institute of Navigation (ION GPS ’96), pp. 935–944, Kansas City, Mo, USA, 1996.
14 R. R. Hatch, “Instantaneous ambiguity resolution,” in Proceedings of the KIS ’90, pp. 299–308, Springer,
Banff, Canada, 1990.
15 R. V. Garcia, H. K. Kuga, and M. C. Zanardi, “Unscented Kalman filter for spacecraft attitude
estimation using quaternions and Euler angles,” in Proceedings of the 22 International Symposium on
Space Flight Dynamics, São José dos Campos, Brazil, 2011.
16 P. Misra and P. Enge, Global Positioning System: Signals, Measurements and Performance, Ganga-Jamuna
Press, Lincoln, Mass, USA, 2001.
17 L. Baroni and H. K. Kuga, “Integer ambiguity resolution in attitude determination using GPS
measurements,” in Proceedings of the 8th International Conference on Mathematical Problems in
Engineering, Aerospace and Sciences (ICNPAA ’10), São José dos Campos, Brazil, 2010.
doi:10.1155/2012/769376
Research Article
Numerical Analysis of Constrained, Time-Optimal
Satellite Reorientation
Robert G. Melton
Department of Aerospace Engineering, Pennsylvania State University, 229 Hammond Building.,
University Park, PA 16802, USA
Correspondence should be addressed to Robert G. Melton, rgmelton@psu.edu
Received 11 July 2011; Accepted 11 October 2011
Copyright q 2012 Robert G. Melton. This is an open access article distributed under the Creative
Previous work on time-optimal satellite slewing maneuvers, with one satellite axis sensor axis
required to obey multiple path constraints exclusion from keep-out cones centered on high-
intensity astronomical sources reveals complex motions with no part of the trajectory touching
the constraint boundaries boundary points or lying along a finite arc of the constraint boundary
boundary arcs. This paper examines four cases in which the sensor axis is either forced to follow
a boundary arc, or has initial and final directions that lie on the constraint boundary. Numerical
solutions, generated via a Legendre pseudospectral method, show that the forced boundary arcs
are suboptimal. Precession created by the control torques, moving the sensor axis away from
the constraint boundary, results in faster slewing maneuvers. A two-stage process is proposed
for generating optimal solutions in less time, an important consideration for eventual onboard
implementation.
1. Introduction
The problem of reorienting a spacecraft in minimum time, often through large angles so-
called slew maneuvers and subject to various constraints, can take a number of forms. For
example, the axis normal to the solar panels may be required to lie always within some
specified minimum angular distance from the sun-line. In cases where the control authority is
low and the slew requires a relatively long time, certain faces of the vehicle may benefit from
being kept as far as possible from the sun-line to avoid excessive solar heating. For many
scientific missions, observational instruments must be kept beyond a specified minimum
angular distance from high-intensity light sources to prevent damage.
Before addressing the time-optimal, constrained problem, it is useful to review what is
known about the unconstrained problem. In a seminal paper, Bilimoria and Wie 1 consider
time-optimal slews for a rigid spacecraft whose mass distribution is spherically symmetric,
and which has equal and independently limited control-torque authority for all three axes.
Despite the symmetry of the system, the intuitively obvious time-optimal solution is not a
λ-rotation about the eigenaxis the appendix contains a discussion of λ-rotations and the
eigenaxis of a direction cosine matrix. Indeed, the fallacy here is that one easily confuses the
minimum-angle of rotation problem i.e., the angle about the eigenaxis with the minimum-
time problem, ignoring the constraints imposed by Euler’s equations of rigid-body motion.
Bilimoria and Wie find that the time-optimal solution includes precessional motion to achieve
a lower time approximately 10% less than that obtained with an eigenaxis maneuver.
Further, they determine that the control history is bang-bang, with a switching structure that
changes depending upon the magnitude of the angular maneuver referring here to the final
orientation in terms of a fictitious λ-rotation, with associated angle θ: for values of θ less
than 72 degrees, the control history is found to contain seven switches between directions of
the control torque components; larger values of θ require only five switches.
Several subsequent papers have revisited the unconstrained problem, including
such modifications as axisymmetric mass distribution and only two-axis control 2,
asymmetric mass distribution 3, small reorientation angles 4, and combined time and
fuel optimization 5. Recently, Bai and Junkins 6 have reconsidered the original problem
spherically symmetric mass distribution, three equal control torques and find that at least
two locally optimum solutions exist for reorientations of less than 72 deg. one of which
requires only six switches if the controls are independently limited. Further, they prove
that if the total control vector is constrained to have a maximum magnitude i.e., with
the orthogonal control components not independent, then the time-optimal solution is the
eigenaxis maneuver.
Hablani 7 and Mengali and Quarta 8 consider constrained maneuvers, but focus
upon generating feasible solutions without attempting to find optimal solutions. Melton 9
considers time-optimal, constrained slewing maneuvers for cases involving multiple path
constraints. That work uses the Swift spacecraft 10 as an example of a vehicle that must
be rapidly reoriented to align two telescopes at a desired astronomical target, namely, a
gamma-ray burst. The satellite’s burst alert telescope wide field of view first detects the
gamma-ray burst and the spacecraft then must reorient to allow the X-ray and UV/optical
instruments to capture the rapidly fading afterglow of the event. To prevent damage to these
instruments, the slewing motion is constrained to prevent the telescopes’ common axis from
entering established “keep-out” zones, defined as cones with central axes pointing to the Sun,
Earth, and Moon, with specified half-angles. A somewhat surprising result is that all of the
cases studied yield trajectories of the sensor axis that neither travel along the boundary of
the keep-out cone nor even touch it so-called boundary arcs and boundary points. Figure 1
shows an example of this behavior, in which the sensor axis traverses a narrow gap 0.1 deg.
between the Sun and Moon cones, but does not intersect either cone.
This paper presents a preliminary study of boundary arcs and boundary points in
this same problem. A full analytical solution is not possible; however, some insight to the
problem can be gained by examining instances where the sensor axis is constrained to follow
the constraint boundary, and those where the initial and final sensor axis directions lie exactly
on the boundary. The paper also addresses the practical challenge of implementing onboard
optimal control for this type of maneuver and proposes a means for reducing computation
time for the reference trajectory.
1.5
Earth
Moon
1
0.5 Sensor axis path
0 Sun
−0.5 Final position
−1 Initial position
−1.5
−1 −0.5
0 0.5
1 1.5 −1 0.5 1
−0.5 0
Figure 1: Trajectory of sensor axis between Sun yellow and Moon gray cones.
2. Problem Statement
The problem is formulated as a Mayer optimal control problem, with performance index
J tf , 2.1
where tf is the final time to be minimized. Euler’s equations of rigid-body motion describe
the system dynamics
M1 − ω2 ω3 I3 − I2
ω̇1
I1
M2 − ω3 ω1 I1 − I3
ω̇2 2.2
I2
M3 − ω1 ω2 I2 − I1
ω̇3 .
I3
Note that the control torques are assumed to be independently bounded, and no assumption
is made about the type of control actuator being used. Euler’s equations must be augmented
with kinematic relationships in order to determine the orientation of the body over time.
In this work, a formulation that uses Euler parameters is employed, with the relationship
between the Euler parameters εi and the angular velocity components ωj given by
⎡ ⎤ ⎡ ⎤⎡ ⎤
ε̇1 ε4 −ε3 ε2 ε1 ω1
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ε̇2 ⎥ 1 ⎢ ε3 ε4 −ε1 ε2 ⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥⎢ω2 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥. 2.3
⎢ε̇3 ⎥ 2 ⎢−ε2 ε1 ε4 ε3 ⎥⎢ ⎥
⎣ ⎦ ⎣ ⎦⎣ω3 ⎦
ε̇4 −ε1 −ε2 −ε3 ε4 0
The appendix discusses the relationship between the Euler parameters, λ-rotations,
and direction cosine matrices. The problem to be considered is a rest-to-rest maneuver, with
initial conditions
ω1 0 ω2 0 ω3 0 0

2.4a
ε1 0 ε2 0 ε3 0 0, ε4 0 1,
and two possible sets of final conditions

ω1 tf ω2 tf ω3 tf 0
2.4b
ε1 tf ε1f , ε2 tf ε2f , ε3 tf ε3f , ε4 tf ε4f ,
for which the final orientation at the final time is completely specified, or

ω1 tf ω2 tf ω3 tf 0
f1 ε1 , ε2 , ε3 , ε4 0
.. 2.4c
.
fn ε1 , ε2 , ε3 , ε4 0,
for which only some aspect of the final orientation is specified this is discussed further in
Section 2.1.
For the unconstrained optimal control problem formulated using 2.1–2.3, the
Hamiltonian is
H λω1 ω̇1 λω2 ω̇2 λω3 ω̇3 λε1 ε̇1 λε2 ε̇2 λε3 ε̇3 λε4 ε̇4 . 2.5
For spacecraft of the type being modeled here Swift, or other astronomical missions, one
or more sensors are fixed to the spacecraft bus; these sensors all have the same central axis
for their fields of view and this axis is designated here with the unit vector σ9 and referred
to as the sensor axis. This axis must be kept at least a minimum angular distance αA from
each of several high-intensity astronomical sources. Denoting the directions to these sources
as σ9A , where the subscript A can be S Sun, E Earth, or M Moon, the so-called keep-out
constraints are then written as follows:
αA ≤ cos−1 9
σ · σ9A . 2.6
Without loss of generality, σ9 is assumed to lie along the body-fixed x-axis and its orientation
with respect to the inertial frame is then determined from A.6, with A as the inertial frame
and B as the body-fixed frame

σ1 1 − 2 ε12 ε32
σ2 2ε1 ε2 ε3 ε4 2.7
σ3 2ε3 ε1 − ε2 ε4 .
It is further assumed that the reorientation maneuver can occur quickly enough that the
spacecraft’s orbital position remains essentially unchanged, and that therefore, the inertial
directions to the high-intensity sources also remain constant during the slew maneuver.
Analytically, the path constraint given by 2.6 can be adjoined to the Hamiltonian by
creating the following constraint function:
C cos−1 9
σ · σ9A − αA , 2.8
then substituting for σ9 using 2.7. The result must then be differentiated twice with respect
to time and using 2.2 and 2.3 in order to get a form in which the control torques
appear 11. Finally, a new Hamiltonian H B is formed by adjoining C̈ to H with Lagrange
multiplier μ,
B H μC̈,
H 2.9
with the conditions

μ ≥ 0 if C̈ 0
2.10
μ 0 if C̈ > 0.
Further, the so-called tangency conditions 11 must also be applied i.e., for any part of
the trajectory on the boundary, not only C, but also its first and second time derivatives
must be zero. The resulting form is analytically intractable for assessing whether necessary
conditions are met along a boundary arc or at a boundary point; however, by using a direct
method Legendre pseudospectral, it is possible to obtain numerical solutions that meet the
necessary conditions of optimality. Fleming and Ross 12 show that a pseudospectral method
is effective for solving the unconstrained minimum-time reorientation problem.
2.1. Semifree Final Orientation

For some missions, the reorientation strategy may be altered if the final orientation is not
completely specified. An example would be the need to reorient the sensor axis to a desired
target direction in minimum time, with no constraints on the orientation of the other body-
fixed axes at the final time. In practice, some subsequent rotation about the sensor axis
might be required to optimize some other parameter e.g., maximizing illumination of solar
panels, or minimizing solar heating of sensitive components, but the principal reorientation
maneuver could be achieved faster. The corresponding optimal control problem is the same
λ = σA
σf
Sensor axis path along

the constraint boundary
σi
Figure 2: Constrained rotation about the keep-out cone axis.
as before, but with the final conditions on the Euler parameter values given in 2.4c specified
by
+ ,
σ1 tf σ1,f 1 − 2 ε12 tf ε32 tf
> ? 2.11
σ2 tf σ2,f 2 ε1 tf ε2 tf ε3 tf ε4 tf
> ?
σ3 tf σ3,f 2 ε3 tf ε1 tf − ε2 tf ε4 tf .
2.2. Constrained Rotation Axis

Consider now the suboptimal approach of a simple λ-rotation that carries the sensor axis σ9
from initial to final orientation along the constraint boundary, with the other body-fixed axes
undergoing the same λ-rotation This amounts to a λ-rotation about the keep-out cone axis σ9A
Figure 2. Such a constrained rotation is a simple one-dimensional problem whose solution
is bang-bang, with the maximum torque being applied along the direction of λ. 9 Because the
control torque components are each limited to a maximum magnitude Mmax , the maximum
torque along λ9 has magnitude
% %
% % % M λ9 %
% % % max %
%M λ % % %, 2.12
% λmax %

where λmax maxλ1 , λ2 , λ3 . Equation 2.12 thus maximizes Mλ while obeying the limits on
the individual control torques M1 , M2 , and M3 . Assuming that a feasible rotation axis λ9 can
be identified, then the rotation angle θ can be determined from A.5, and the rotation time is
given by

Iλ θ
tf,λ 2 , 2.13
|Mλ |
9
where Iλ is the moment of inertia about the λ-axis. This provides a useful benchmark to which
the optimal solution can be compared. As a practical matter, such a forced-axis rotation could
be an acceptable suboptimal solution if the optimal path required too much computing time
or resources to calculate.
3. Results
Four cases are considered: two of these cases BA1 and BA2 constrain the sensor axis to
move along the constraint boundary; the initial and final sensor axis directions also lie on
the boundary. These cases serve as proxies for boundary arcs, at least in that they provide
some indication of the slewing time and other qualitative properties of the motion. The other
spacecraft axes are unconstrained at the final time. The other two cases BP1 and BP2 have the
sensor axis beginning and terminating on the constraint boundary and of course, prohibited
from entering the constraint cone; the other spacecraft axes are unconstrained at the final
time.
The numerical results are generated using a Legendre pseudospectral method,
implemented in the software package DIDO 13. In all cases, the discrete solution for the
control history produced by the pseudospectral method has been subsequently employed
to numerically propagate the dynamics from the given initial conditions; the results give
solutions that match the discrete solutions to within an absolute error of 10−16 corresponding
units at each node. These cases all require significant computation time as much as 72 hours
on a computer with an Intel Core 2 2.0 GHz processor, with the number of pseudospectral
nodes in the range of 100–250.
Note that the Hamiltonian and costate values are not evaluated directly during the
pseudospectral solution process, but rather are reconstructed via the covector mapping
principle 14 after the problem solution is found. In all cases here, the Hamiltonian is found
to be reasonably constant given the relatively small number of nodes and level of accuracy
specified for the nonlinear programming aspect of the calculations.
A system of nondimensional units is employed, partly to provide somewhat more
general results, but chiefly because this system of units provides the kind of scaling needed
for the pseudospectral method to perform well. In physical units, the angular velocities,
moments of inertia, and control torques about the spacecraft’s principal axes are denoted
as ωi , I i , and Mi , respectively; the corresponding nondimensional quantities are defined as

I max
ξ
Mmax
ωi ξωi
3.1
Ii
Ii
I max
Mi
Mi ,
Mmax
where ξ is the time unit, and I max and Mmax are the maximum values of the principal inertias
and control torques, respectively. In all of the cases presented, the spacecraft is assumed to
have a spherically symmetric mass distribution i.e., all three principal moments of inertia
are equal and three-axis control capability, with equal and independently limited control
authority about all three axes.
3.1. Case BA1

This is a forced boundary arc trajectory of the sensor axis σ9 . The constraint cone has half-angle
of 45 deg. corresponding approximately to the Sun keep-out cone for the Swift spacecraft
and the sensor axis must traverse an arc of −90 deg. about the cone’s central axis. This solution
uses 151 nodes in the pseudospectral method even a modest increase in the number of nodes
to 171 resulted in nonconvergence with the available computing resources.
A notable feature of the motion Figure 3 is its qualitative similarity to the
unconstrained time-optimal solution shown in Figure 4. Referring to the angular velocity
components, it is evident that the motion includes precession, with portions of the trajectory
including fully saturated control torques. Some finite time intervals have intermediate
torques, necessitated by the boundary arc constraint. This solution yields a final time of tf
1.9480 nondimensional units. For comparison, if the rotation axis is constrained to lie along
the constraint cone’s axis, the final time would be tf,λ 2.1078.
Figure 5 depicts the Hamiltonian and costate histories. It is evident that the
Hamiltonian is fairly constant, giving some confidence that the optimal solution has been
determined.
3.2. Case BA2
For this forced boundary arc, the constraint cone has half-angle αA 23 deg. corresponding
to the Moon cone for Swift, and the sensor axis must traverse an angle of −70 deg. about
the cone’s central axis. The final orientation of the sensor axis σ9f is calculated via A.1, with

a σ9i 1, 0, 0, b σ9f , θ −70 deg., and λ9 cosαA , sinαA , 0.

For this case, the solution uses 100 nodes and has a final time of tf 1.3020. For
comparison, if the rotation axis is constrained to lie along the constraint cone’s axis, the
final time would be tf,λ 2.0966. Figure 6 depicts the dynamic response and control
torques; Figure 7 shows the Hamiltonian and costate histories. The motion and controls are
qualitatively similar to those in case BA1 . It should be noted that in both BA1 and BA2 , the
path of the sensor axis is verified to lie within 10−16 radians of the constraint boundary.
3.3. Case BP1

This case is identical to case BA1 , except that only the initial and final directions of the sensor
axis are constrained to lie on the constraint boundary, corresponding to two forced boundary
points. Two of the control torques M1 and M2 exhibit bang-bang behavior whereas M3
shows some chatter even with 250 nodes employed, as seen in Figure 8. The somewhat
larger variation in the Hamiltonian Figure 9 occurs near the initial time; however, this
happens at a finite time t 0.025 corresponding to 22 pseudospectral nodes after t 0 after
the sensor axis has moved away from the constraint boundary Figure 10. Such behavior
is therefore not the theoretically expected discontinuity in the Hamiltonian and costates at
ωi
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
ω1
ω2
ω3
a
2
1
εi
0
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
ε1 ε3
ε2 ε4
b
1
Mi
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
M1
M2
M3
c
Figure 3: Dynamic response and controls for the case BA1 .
a point where the trajectory leaves the constraint boundary 11; indeed, such discontinuities
may not be observable in numerical solutions where the equality condition in 2.10 is
unlikely to occur. The numerical solutions may be improved by using a Bellman chain, based
upon a sequence of embedded optimal solutions each of which can use a relatively small
number of nodes 15.
Nevertheless, the trend is clear: precession created by the control torques, works to
reduce the final time to 1.9258, approximately 1% faster than the solution in BA1 .
3.4. Case BP2

This case is identical to case BA2 , except that only the initial and final directions of the sensor
axis are constrained to lie on the constraint boundary. Unlike case BP1 , the control torques
display more intermediate behavior Figure 11; solution accuracy is not as good here since
convergence could not be obtained for more than 100 nodes in the 72 hours of computer
time available. This is also evident in the slight variation in the Hamiltonian Figure 12. As
with case BP1 , the only points where the sensor axis contacts the constraint boundary see
Figure 13 are at the initial and final times. The final time achieved is tf 1.2967.
Angular velocity
2
1
ωi
0
−1
0 0.5 1 1.5 2 2.5 3 3.5
Time
ω1
ω2
ω3
a
Euler parameters
2
1
εi
0
−1
0 0.5 1 1.5 2 2.5 3 3.5
Time
ε1 ε3
ε2 ε4
b
Control torques
1
0
Mi
−1
0 0.5 1 1.5 2 2.5 3 3.5
Time
M1
M2
M3
c
Figure 4: Dynamic response and controls 9 for the motion shown in Figure 1.
−0.96
−0.98
Hamiltonian
−1
−1.02
−1.04
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
a
500
0
Costates
−500
−1000
−1500
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
λω1 λε2
λω2 λε3
λω3 λε4
λε1
b
Figure 5: Hamiltonian and costates for the case BA1 .
0.5
0
ωi
−0.5
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
ω1
ω2
ω3
a
2
1
εi
0
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
ε1 ε3
ε2 ε4
b
1
Mi
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
M1
M2
M3
c
Figure 6: Dynamic response and controls for the case BA2 .
4. Practical Consideration
Regardless of whether boundary points or boundary arcs exist as part of an optimal trajectory,
the numerical determination of the solution via a pseudospectral method frequently requires
considerable computation time; indeed, the actual slewing maneuver of the spacecraft can
be accomplished in seconds or minutes depending upon the control authority while
the pseudospectral solver requires from 20 minutes to 72 hours to compute the solution.
Experience shows that providing even a rudimentary estimate of the states and controls as
an initial guess for the pseudospectral solver can reduce the computation time significantly.
A two-stage process is proposed, wherein a random-process algorithm, such as a particle
swarm optimizer PSO which can rapidly explore the solution space, provides the initial
guess to the pseudospectral algorithm. The literature abounds with hybrid methods used in
related control problems. For example, Ahmed et al. 16 successfully apply just a PSO to the
problem of tuning a satellite’s attitude controller while Sentinella and Casalino 17 examine
a hybrid evolutionary algorithm that comprises differential evolution, genetic algorithms,
and a PSO applied to the problem of spacecraft trajectory optimization.
−0.96
−0.98
Hamiltonian
−1
−1.02
−1.04
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
a
100
0
Costates
−100
−200
−300
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
λω1 λε2
λω2 λε3
λω3 λε4
λε1
b
Figure 7: Hamiltonian and costates for the case BA2 .
Initial efforts that employ a two-stage process for optimal slewing maneuvers show
promising results. A PSO produces a low-quality approximation of the states, controls,
and node times for the solution, followed by the pseudospectral solver, which takes the
approximate solution as its initial starting point. An efficient method for generating the first-
stage solution is to represent each control torque component Mi as a sum of N Chebyshev
polynomials of the form 18

&
N−1
1
Mi t ≈ ck Tk t − c0 , 4.1
k0
2
where Tk is the Chebyshev polynomial of degree k, and the coefficients ck are determined by
the PSO using an explicit integration of the equations of motion 2.2, 2.3. Implementation
of 4.1 makes use of the Clenshaw recursion relation 18 for rapid evaluation of the
Chebyshev polynomials.
As an example, the control torque for a one-dimensional slewing maneuver with no
keep-out cones is shown in Figure 14a. The PSO runs for only 50 iterations requiring 64
seconds of computation time, producing a crude approximation to the bang-bang control
solution that is known to be optimal. That approximate solution is then employed by the
DIDO pseudospectral solver to calculate the optimal solution Figure 14b, requiring 12
seconds. Using this two-stage method, the total computation time PSO plus pseudospectral
solver requires approximately half the time needed using only the pseudospectral method
ωi
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
ω1
ω2
ω3
a
2
1
εi
0
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
ε1 ε3
ε2 ε4
b
1
Mi
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
M1
M2
M3
c
Figure 8: Dynamic response and controls for the case BP1 .
148 seconds with no initial guess for the solution. Future study should examine the utility
of the two-stage method for the full three-dimensional, constrained problem.
5. Conclusion
This preliminary study indicates that, although boundary arcs and boundary points may exist
in time-optimal spacecraft slewing maneuvers with path constraints, they are at best part of
a suboptimal solution. The numerical calculations completed via a Legendre pseudospectral
method show that even if the initial and final states are boundary points, the solution moves
away from the constraint boundary, resulting in a lower final time than if the motion is forced
to move along the boundary a forced boundary arc. The necessary conditions lead to an
unwieldy set of relations, making it impossible to determine analytically if boundary points
or boundary arcs are excluded. Further examination of the problem using a Bellman chain
to improve the numerical accuracy may provide additional insight. A two-stage method for
generating optimal solutions in less time than that required by the pseudospectral method
alone shows some promise, but further work is needed to determine its utility for the three-
dimensional constrained problem.
−0.98
−1
Hamiltonian
−1.02
−1.04
−1.06
−1.08
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
a
2000
1000
Costates
0
−1000
−2000
−3000
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
λω1 λε2
λω2 λε3
λω3 λε4
λε1
b
Figure 9: Hamiltonian and costates for the case BP1 .
0.035
Distance from “keep-out” cone (rad)
0.03
0.025
0.02
0.015
0.01
0.005
−0.005
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
Figure 10: Distance from the sensor axis to the constraint boundary case BP1 .
Appendix
λ-Rotations and the Eigenaxis
Consider a dextral orthonormal basis set a9j , fixed in reference frame A. A copy of this set,
denoted b9i , is rotated in a right-handed sense, through an angle θ, about a unit vector λ,
9
which has fixed orientation with respect to A. The new orientation is denoted as reference
frame B and the rotation axis λ9 will have the same orientation with respect to B as it has to A.
0.5
0
ωi
−0.5
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
ω1
ω2
ω3
a
2
1
εi
0
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
ε1 ε3
ε2 ε4
b
1
Mi
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
M1
M2
M3
c
Figure 11: Dynamic response and controls for the case BP2 .
This representation of the orientation of B with respect to A is called a λ-rotation 19. For

two vectors a and b , fixed in frames A and B, respectively, if b a initially, and B undergoes

a λ-rotation, then the subsequent relationship between a and b is given by

b a cos θ− a × λ9 sin θ a · λ9 λ1
9 − cos θ,

A.1
a result named after Olinde Rodrigues Goldstein 20 claims that the relationship was known
before Rodrigues, and that the vector form used here was first published by Gibbs 21. This
form is useful in describing rotations of a sensor axis about an axis fixed in both the inertial
and spacecraft frames.
If one uses a direction cosine matrix CAB to express the orientation of B with respect
to A, then the constituent basis vectors are related by
⎡ ⎤ ⎡ ⎤
b91 a91
⎢ ⎥ ⎢ ⎥
⎢9 ⎥ AB ⎢ ⎥
⎢b2 ⎥ C ⎣a92 ⎦, A.2
⎣ ⎦
b93 a93
−0.9
−0.95
Hamiltonian
−1
−1.05
−1.1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
a
5000
0
Costates
−5000
−10000
−15000
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
λω1 λε2
λω2 λε3
λω3 λε4
λε1
b
Figure 12: Hamiltonian and costates for the case BP2 .
0.012
Distance from “keep-out” cone (rad)
0.01
0.008
0.006
0.004
0.002
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time
Figure 13: Distance from the sensor axis to the constraint boundary case BP2 .
and the elements of CAB are given by
Cij b9i · a9j . A.3
Every direction cosine matrix has one unity-valued eigenvalue with corresponding
eigenvector in matrix form e, that is,
CAB e e, A.4
1D slew maneuver Dido result—200 nodes, accurate mode

1.5
1 1
Control torque
0.6
0.5 0.2
Control torque
0 −0.2
−0.5 −0.6
−1
−1
0 0.5 1 1.5 2 2.5 3
−1.5
Time
−2
0 0.5 1 1.5 2 2.5
Time
PSO values
Chebyshev approx
a b
Figure 14: a Approximate solution for 1D slewing maneuver, generated by 50 iterations of a particle
swarm optimizer. b Optimal control solution for the same maneuver, generated by DIDO running in
accurate mode. Total required cpu time for solution: 148 sec using only DIDO, with no initial guess; 76 sec
for the combined PSO-DIDO solution.
and the elements of e clearly constitute the components of the associated λ-rotation vector.
The vector e is commonly referred to as the eigenaxis for the rotation that generates B from A.

By taking dot products of both sides of A.1 with a and then with b , and using
appropriate vector identities, it can be shown that
2 2
a · b − a · λ9 −1 b · a cos θ a · λ9 1 − cos θ

cos θ 2 , sin θ , A.5

1 − a · λ9 b · a × λ9

9 Another useful relationship gives the
which permits the calculation of θ for given a, b , and λ.
direction cosine matrix CAB in terms of the Euler parameters
⎡ ⎤
1 − 2 ε22 ε32 2ε1 ε2 ε3 ε4 2ε1 ε3 − ε2 ε4
⎢ 2 ⎥
CAB ⎢
⎣2ε2 ε1 − ε3 ε4 1 − 2 ε1 ε3 2ε2 ε3 ε1 ε4 ⎦.
2 ⎥ A.6
2
2ε3 ε1 ε2 ε4 2ε3 ε2 − ε1 ε4 1 − 2 ε1 ε22
Nomenclature
C: Constraint function
H: Hamiltonian without the path constraint
B
H: Hamiltonian with the path constraint
Ii : Principal moment of inertia nondimensional
I i: Principal moment of inertia in physical units
J: Performance index
Mi : Control torque nondimensional

Mi : Control torque in physical units
tf : Final time
αA : Half-angle of the keep-out cone for object A
εi : Euler parameter
9
λ: Rotation axis
λx : Costate corresponding to state x
μ: Lagrange multiplier associated with constraint function C
θ: 9
Rotation angle about the λ-axis
ξ: Time unit
σ9 : Direction of the sensor axis
σ9A : Direction of the central axis of the keep-out cone for object A
ωi : Angular velocity component nondimensional
ωi : Angular velocity component in physical units.
Acknowledgment
This paper has been presented at the 6th International Workshop and Advanced School
“Spaceflight Dynamics and Control,” Covilha, Portugal, March 28–30, 2011, http://www
.aerospace.ubi.pt/workshop2011/.
References
1 K. D. Bilimoria and B. Wie, “Time-optimal three-axis reorientation of a rigid spacecraft,” Journal of
2 H. Shen and P. Tsiotras, “Time-optimal control of axisymmetric rigid spacecraft using two controls,”
Journal of Guidance, Control, and Dynamics, vol. 22, no. 5, pp. 682–694, 1999.
3 R. J. Proulx and I. M. Ross, “Time-optimal reorientation of asymmetric rigid bodies,” in Proceedings
of Advances in the Astronautical Sciences AAS/AIAA Astodynamics Conference, vol. 109, pp. 1207–1227,
August 2002.
4 S. L. Scrivener and R. C. Thompson, “Time-optimal reorientation of a rigid spacecraft using
collocation and nonlinear programming,” in Proceedings of Advances in the Astronautical Sciences
AAS/AIAA Astodynamics Conference, vol. 85, pp. 1905–1924, August 1993.
5 S. W. Liu and T. Singh, “Fuel/time optimal control of spacecraft maneuvers,” Journal of Guidance,
6 X. Bai and J. L. Junkins, “New results for time-optimal three-axis reorientation of a rigid spacecraft,”
7 H. B. Hablani, “Attitude commands avoiding bright objects and maintaining communication with
ground station,” Journal of Guidance, Control, and Dynamics, vol. 22, no. 6, pp. 759–767, 1999.
8 G. Mengali and A. A. Quarta, “Spacecraft control with constrained fast reorientation and accurate
pointing,” Aeronautical Journal, vol. 108, no. 1080, pp. 85–91, 2004.
9 R. G. Melton, “Constrained time-optimal slewing maneuvers for rigid spacecraft,” in Proceedings of
the Advances in the Astronautical Sciences AAS/AIAA Astrodynamics Specialist Conference, vol. 135, pp.
107–126, Univelt, San Diego, Calif, USA, 2010.
10 Swift, http://heasarc.nasa.gov/docs/swift/swiftsc.html.
11 A. E. Bryson and Y.-C. Ho, Applied Optimal Control: Optimization, Estimation and Control, Hemisphere
Publishing, New York, NY, USA, 1975.
12 A. Fleming and I. M. Ross, “Minimum-time maneuvering of CMG-driven spacecraft,” in Proceedings
of the Advances in the Astronautical Sciences AAS/AIAA Astrodynamics Specialist Conference, vol. 129, pp.
1623–1644, 2007.
13 I. M. Ross, “A beginner’s guide to DIDO ver 7.3: a MATLAB application package for solving optimal
control problems,” Tech. Rep. TR-711, Elissar LLC, Monterey, Calif, USA, 2007.
14 Q. Gong, I. M. Ross, W. Kang, and F. Fahroo, “Connections between the covector mapping theorem
and convergence of pseudospectral methods for optimal control,” Computational Optimization and
Applications, vol. 41, no. 3, pp. 307–335, 2008.
15 I. Michael Ross, Q. Gong, and P. Sekhavat, “Low-thrust, high-accuracy trajectory optimization,”
16 R. Ahmed, H. Chaal, and D. W. Gu, “Spacecraft controller tuning using particle swarm optimization,”
in Proceedings of the International Joint Conference 2009 (ICCAS-SICE ’09), pp. 73–78, August 2009.
17 M. R. Sentinella and L. Casalino, “Cooperative evolutionary algorithm for space trajectory
optimization,” Celestial Mechanics and Dynamical Astronomy, vol. 105, no. 1, pp. 211–227, 2009.
18 W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes in C++, The Art of
Scientific Computing, Cambridge University Press, 2nd edition, 2002.
19 T. R. Kane, P. W. Likins, and D. A. Levinson, Spacecraft Dynamics, McGraw-Hill, 1983.
20 H. Goldstein, Classical Mechanics, Addison-Wesley, 2nd edition, 1981.
21 J. W. Gibbs., Vector Analysis, edited by E. B. Wilson, Scribner, 1901; Yale University Press, London, UK,
1931.
doi:10.1155/2012/532708
Research Article
Low-Thrust Out-of-Plane Orbital Station-Keeping
Maneuvers for Satellites
Vivian M. Gomes and Antonio F. B. A. Prado

DMC, National Institute for Space Research (INPE), Avenida dos Astronautas 1758,
Correspondence should be addressed to Vivian M. Gomes, vivian.gomes@uol.com.br
Received 30 October 2011; Revised 30 November 2011; Accepted 8 January 2012
Copyright q 2012 V. M. Gomes and A. F. B. A. Prado. This is an open access article distributed
This paper considers the problem of out of plane orbital maneuvers for station keeping of satellites.
The main idea is to consider that a satellite is in an orbit around the Earth and that it has its orbit
is disturbed by one or more forces. Then, it is necessary to perform a small amplitude orbital
correction to return the satellite to its original orbit, to keep it performing its mission. A low thrust
propulsion is used to complete this task. It is important to search for solutions that minimize the
fuel consumption to increase the lifetime of the satellite. To solve this problem a hybrid optimal
control approach is used. The accuracy of the satisfaction of the constraints is considered, in
order to try to decrease the fuel expenditure by taking advantage of this freedom. This type of
problem presents numerical difficulties and it is necessary to adjust parameters, as well as details
of the algorithm, to get convergence. In this versions of the algorithm that works well for planar
maneuvers are usually not adequate for the out of plane orbital corrections. In order to illustrate
the method, some numerical results are presented.
1. Introduction
The present paper studies the orbital maneuvers required by a spacecraft that needs to
perform an orbital correction maneuver in order to remain with its orbital elements inside
a region where it can achieve the goals of the mission. It is assumed that the orbit of the
satellite is given, as well as a nominal orbit for this satellite that allows it to be useful
for the planned activities. Then, it is necessary to maneuver this satellite from its current
position to the nominal specified orbit. Emphasis will be given in out-of-plane maneuvers,
because this situation was not much explored in previous studies, like in 1, 2, and it has a
high cost in terms of fuel consumption. Another contribution of the present paper is that it
studies the problem of station-keeping maneuvers. In terms of theoretical definition, orbital
transfers and station keeping are quite similar. But, when facing practical problems, from
the engineering point of view, they are different. This fact can be seen in the literature that
presents many methods for station-keeping and orbital transfers, considering them as two
different problems. For station-keeping, it is necessary much lower values for the parameters
that control the convergence of the method. It means that convergence is much more difficult
in the present case. To solve this problem, it was necessary to divide the convergence of the
method in several steps. First a larger number for the tolerance was used and then, using the
solution obtained in this step as first guess, we searched for new values of the parameters
α1 and β1, in order to make the steps in the direction of the convergence smaller than in
the previous case. This technique is repeated until the goal defined for the convergence
is reached. So, the present paper shows that this method is also valid for the important
application of station keeping maneuvers.
The control available to perform this maneuver is the application of a low thrust,
and the objective is to perform this maneuver with minimum fuel consumption. An optimal
approach will be used. There is no time restriction involved here, and the spacecraft can leave
from any point in the initial orbit and arrive at any point in the final orbit. The stochastic
version of the projection of the gradient method is used. This version allows us to include
the fact that the constraints do not need to be exactly satisfied see 1, 2. This is done
to realistically treat the numerical inaccuracies and/or flexibilities in terms of tolerance in
mission requirements, leading to situations where the final state is constrained to lie inside a
given region, instead of having an exact value.
2. Review of Orbital Maneuvers

A very important result in this field was obtained by Hohmann 3. He solved the problem
of minimum ΔV transfers between two circular coplanar orbits. The Hohmann transfer
would later be generalized to the elliptic case transfer between two coaxial elliptic orbits
by Marchal 4.
After that, can be found in the literature the problem of a two-impulse transfer, where
the magnitude of the two impulses are fixed, like in Jin and Melton 5 and Jezewski and
Mittleman 6.
Then, the three-impulse concept was introduced in the literature by Hoelker and Silber
7 and Shternfeld 8. They showed that a bielliptical transfer between two circular orbits
has a ΔV lower than the one required by the Hohmann transfer, for some combinations of
the initial and final orbits. Continuing those studies, Ting 9 showed that the use of more
than three impulses does not lower the ΔV , considering only situations where impulsive
maneuvers are used.
Some other researchers worked on methods where the number of impulses is a
parameter to be optimized, and not a value fixed in advance. It is the case of the papers
written by Jezewski and Rozendaal 10 and Eckel 11. Most of the research done in this
situation is based on the “Primer-Vector” theory, developed by Lawden and showed in
12, 13.
Later, two more techniques were introduced in the orbital maneuvers field, using the
concepts of swing-by and gravitational capture. Those techniques are based on the use of the
gravitational forces of a third body to increase or decrease the energy of the spacecraft, so
reducing the fuel consumption of the maneuver. References 14–20 describe this problem
and show some applications of those techniques in more details.
3. Definition of the Problem

The objective of the problem considered here is to change the orbit of a spacecraft. The
amplitude of this orbital maneuver has to be compatible with the requirement that the
spacecraft needs to return to its nominal orbit that was affected by some perturbations
21, 22, but this is done in a small amplitude framework. Note that the third-body
perturbation, when the perturbed body is not in the same plane of the perturbing body, can
change the inclination of the perturbed body, as shown in 21. Then, an initial and a final
orbit around the Earth are completely specified. The problem is to find how to transfer the
spacecraft between those two orbits and make this transfer in a form that minimizes the fuel
consumption. Time restrictions will not be considered in the present investigation, and the
spacecraft can leave and arrive at any point in the given initial and final orbits. The maneuver
is made by an engine that can deliver a thrust with constant magnitude and variable direction.
Although the algorithm is generic regarding the orbital elements changed, some adaption is
made in order to search for faster convergence in the case of a maneuver that changes the
inclination of the orbit.
The spacecraft is assumed to be travelling in a Keplerian orbit without perturbation
forces during the maneuver. This orbit is controlled by the thrusts, so the orbit is no longer
keplerian when the engines are turned on. In a situation like this, there are two types of
motion:
i a Keplerian orbit, assuming that the Earth is a point of mass and that it dominates
the motion of the spacecraft. This motion occurs when the engine is not being
applied to the satellite;
ii the motion governed by the two forces: the gravity of the Earth and the force
delivered by the thrusts. This motion occurs when the engine is working.
The thrusts have the following characteristics:
i fixed magnitude, so intermediate values are not allowed and the thrusts can deliver
the full power or are not working;
ii constant ejection velocity that has to occur to be consistent with the assumption that
the velocity of the gases ejected from the thrusts is constant;
iii free angular motion: the direction of the thrusts is not constant during the transfer.
This direction is specified by the angles u1 and u2 , called pitch the angle between
the direction of the thrust and the perpendicular to the line Earth-spacecraft and
yaw the angle with respect to the orbital plane;
4. Optimal Control Formulation

This approach is based on optimal control theory. First-order necessary conditions for a local
minimum are used to obtain the adjoint equations the Pontryagin’s maximum principle,
which allow us to obtain the control angles at each instant of time in fact, we search for
those control angles as a function of the range angle that is an independent variable that
replaces the time and is equivalent to the true anomaly. These assumptions lead us to
a Two-Point Boundary Value Problem TPBVP, where the difficulty is to find the initial
values of the Lagrange multipliers. The approach used in the present research is the hybrid
approach of guessing a set of values, numerically integrating all the differential equations,
and then searching for a new set of values, based on a nonlinear programming algorithm.
With this approach, the problem is reduced to a parametric optimization, where the Lagrange
multipliers are variables to be optimized. This idea follows the same principles used by Biggs
in 23, 24.
The method also showed by Biggs 24, where the “adjoint-control” transformation
is performed and, instead of the initial values of the Lagrange multipliers, one guesses the
control angles and their rates at the beginning of the “burning arc,” is used here. Then, it is
easier to find a good initial guess, and the convergence is faster. This hybrid approach has the
advantage that, since the Lagrange multipliers remain constant during the “ballistic arcs,” it
is necessary to guess values of the control angles and their rates only for the first “burning
arc.” This transformation reduces very much the number of variables to be optimized and, as
a consequence, the time of convergence.
The next paragraphs show how to write this problem using an optimal control
approach.
Objective Function
Let Mf be the final mass of the vehicle. It has to be maximized with respect to the control
u·;
Subject to

ẋ f x, u, s ,

Ce x, u, s Ee,
4.1
Cd x, u, s ≤ Ed,

h x tf , tf Eh, t0 , xt0 given,
where x is the state vector, f· is the right-hand side of the equations of motion, in the same
way that was used by Biggs 23, 24 and by Prado and Rios-Neto 25 and Bryson and Ho
26 s is the independent variable that replaces the time s0 ≤ s ≤ sf, Ce· and Cd· are the
algebraic dynamic constraints on state and control that have dimensions me and md , h· are
the boundary constraints of dimension mh , and Ee, Ed, Eh are error vectors satisfying
|Eei | ≤ EeTi , i 1, 2, 3, . . . , me ,
|Edi | ≤ EdTi , i 1, 2, 3, . . . , md , 4.2
|Ehi | ≤ EhTi , i 1, 2, 3, . . . , mh ,
where the fixed given values EeTi , EdTi , and EhTi characterize the region around zero within
which errors are considered tolerable.
To avoid singularities or close approaches to them, we use the following variables:
1/2
a 1 − e2
X1 ,
μ

X2 e cos ω − φ ,

X3 e sin ω − φ ,
Fuel Consumed
X4 ,
m0
X5 t,
4.3
i Ω φ
X6 cos cos ,
2 2

i Ω−φ
X7 sin cos ,
2 2

i Ω−φ
X8 sin sin ,
2 2

i Ω φ
X9 cos sin ,
2 2
where: a semi-major axis, e eeccentricity, i inclination, Ω argument of the ascending

node, ω argument of perigee, f true anomaly, s range angle, φ f ω − s,μ
gravitational constantl, m0 initial mass of the spacecraft.
In those variables, the equations of motion are the following ones:
dX1
f1 SiX 1 F1 ,
ds
dX2
f2 Si{Ga 1 coss X2 F1 GaF 2 sins},
ds
dX3
f3 Si{Ga 1 sins X3 F1 − GaF 2 coss},
ds
dX4 SiGaF1 − X4
f4 ,
ds X1 W
dX5 SiGa1 − X4 m0
f5 , 4.4
ds X1
dX6 X7 coss X8 sins
f6 − SiF 3 ,
ds 2
dX7 X6 coss − X9 sins
f7 SiF 3 ,
ds 2
dX8 X9 coss X6 sins
f8 SiF 3 ,
ds 2
dX9 X7 sins − X8 coss
f9 SiF 3 ,
ds 2
where
Ga 1 X2 coss X3 sens,
μX14 4.5
Si + ,.
Ga3 m0 1 − X4
F, F1 , F2 , F3 are the forces generated by the thrust and they are given by
F F1 F2 F3 ,

% %
% %
%F % F,
4.6
F1 F cosα cos β ,

F2 F senα cos β ,

F3 F sen β ,
where α is the angle between the direction of the thrust and the line that is perpendicular to
the radius vector and β is the angle between the direction of the thrust and the orbital plane.
The equations for the Lagrange multipliers are given as follows:
69
dp1 4 j1 pj fj p1 f1 − p4 f4 − p5 f5
− ,
ds X1
⎡ ⎤
dp2 coss & 9
⎣3 pj fj − p4 f4 − p5 f5 ⎦
ds Ga j1

− Si p2 F1 − Si cos2 s p2 F1 − p3 F2 − Si cosssens p2 F2 p3 F1 ,
⎡ ⎤
dp3 sens & 9
⎣3 pj fj − p4 f4 − p5 f5 ⎦
ds Ga j1

− Si p3 F1 − Si coss sens p2 F1 − p3 F2 − Si sen2 s p2 F2 p3 F3 ,
⎡6 ⎤
9
dp4 j1 pj fj − p4 f4 − p5 f5 4.7
−⎣ ⎦,
ds m0 1 − X4
dp5
0,
ds > ?
dp6 −SiF3 p7 coss p8 sens
,
ds > 2 ?
dp7 SiF3 p6 coss − p9 sens
,
ds > 2 ?
dp8 SiF3 p6 sens p9 coss
,
ds > 2 ?
dp9 −SiF3 p8 coss − p7 sens
.
ds 2
The control that is applied to the spacecraft also needs a substitution of variables, with
the objective of solving numerical problems. In this situation we use the following set of
variables:
u1 s0 ,

u2 sf − s0 cos β0 cosα0 ,

u3 sf − s0 cos β0 sinα0 ,
4.8
u4 sf − s0 sin β0 ,
u5 α ,
u6 β .
The first-order necessary conditions of the optimal problem, which are the conditions
that allow us to obtain the optimal control, can be obtained, at every instant of time, by
extremizing the Hamiltonian of the system. The equations are given as follows:
q2
sinα ,
S
q3
sinB ,
S
q1 4.9
cosα ,
S
S
cosB ,
S
where
+ ,1/2
S ± q12 q22 ,
> ?1/2
S ± q12 q22 q32 ,
q1 p1 X1 p2 X2 Ga 1 coss p3 X3 Ga 1 sins,
4.10
q2 p2 Ga sins − p3 Ga coss,
X7 coss X8 sins

q3 − p6 p7 X6 coss − X9 sins
2
p8 X6 sins X9 coss p9 X7 sins − X8 coss.
It is possible to include constraints in this type of problem. They can be represented by the
following equations:
S· ≥ 0,
a − a∗
0,
|a0 − a∗ |
a1 e − a∗ 1 e∗
0,
|a0 1 e0 − a∗ 1 e∗ |
i − i∗
0,
|i0 − i∗ |
Ω − Ω∗
0,
|Ω0 − Ω∗ |
ω − ω∗
0.
|ω0 − ω∗ |
4.11
In those constraints, the first equation represents a generic inequality constraint and the other
five represent the important constraint of having an orbit specified.
5. Numerical Method
To solve this problem, the stochastic version of the projection of the gradient method Rios-
Neto and Pinto 1 can be used. It is showed in the next paragraphs.
We start by assuming an initial value for p, the vector of parameters p that represents
the variables that we are searching. It may come from an initial guess or from a value that
belongs to an immediately previous iteration. Then, a first-order direct search approach is
adopted in a typical iteration to obtain an approximate solution for the increment Δp:

Minimize: J p Δp 5.1

Subject to: Ce p Δp α1Ce p Ee, 5.2

Cd p Δp β1Cd p Ed, 5.3
where Jp is the objective function, Cep is the equality constraints, Cdp is the active
inequality constraints at p, and 0 ≤ α1 < 1, 0 ≤ β1 < 1 are parameters that have to be chosen
close enough to one to make the increments Δp be of the first order of magnitude.
Linearizing the left-hand sides of 5.2 and 5.3 and using a stochastic interpretation
for the errors Ee and Ed, we have
⎛ + , ⎞
d Ce p
⎜ ⎟
α1 − 1Ce p ⎝ ⎠Δp Ee, 5.4
dp
⎛ + , ⎞
d Cd p
⎜ ⎟
β1 − 1 Cd p ⎝ ⎠ Δp Ed, 5.5
dp
where Ed and Ee are zero mean uniformly distributed errors, given by

+ ,
E EeEeT diagei , i 1, 2, . . . , me ,
+ , 5.6
E EdEdT diagdi , i 1, 2, . . . , md ,
where E· indicate the expected value of its argument.

The condition shown in 5.1 is approximated by

−g · ∇ J T p Δp n, 5.7
where g ≥ 0 needs to be adjusted to guarantee that Δp is small enough to permit the use of
the linearized representation of Jp Δp, and n is taken as a zero mean uniformly distributed
random vector, modeling the a priori searching error in the direction of the gradient ∇Jp,
with
+ ,
E nnT P 5.8
as its diagonal covariance matrix. The values of the variances in P are chosen such as to
characterize an “adequate order of magnitude” for the dispersion of n. The diagonal form
adopted is to model the assumption that it does not impose any a priori correlation between
the errors in the gradient components.
The simultaneous consideration of conditions of 5.4, 5.5, and 5.7 characterizes a
problem of parameter estimation, which can be written as
U U n, 5.9
Y MU V , 5.10
where UΔ − g · ∇J T p is the “a priori information”; U ΔΔp; Y Δ α1 − 1CeT p : β1 −
1CdT p is the observation vector; MT ΔdCep/dpT : dCdp/dpT ; V T EeT :
EdT .
Adopting a criterion of linear, minimum variance estimation, the optimal search
increment can be determined using the classical Gauss-Markov estimator, which in Kalman
form gives

9 U K Y −MU ,
U 5.11
P P − KM P , 5.12
−1
K P MT M P MT R , 5.13
where P is defined as before, R ΔEV V T diag Rk , k 1, 2, . . . , me md , and P has the

meaning of being the covariance matrix of the errors in the components estimates of U, that
is,
T
P E 9
U−U 9
U−U . 5.14
To build a numerical algorithm using the proposed procedure, the following types of
iterations are considered:
i initial phase of acquisition of constraints: when starting from a feasible point
that satisfies the inequality constraints, the search is done to capture the equality
constraints, including those inequality constraints that eventually became active
along this phase;
ii search of the minimum: when from a point that satisfies the constraints in the limits
of the tolerable errors V in 5.10, the search is done to take the objective function
5.1 to get closer to the minimum; this search is conducted by relaxing the order of
magnitude of the error bounds around the constraints;
iii restoration of the constraints: when from a point that resulted from a type ii
iteration, the search is done to restore constraints satisfaction, within the limits
imposed by the error V in 5.10.
Rios-Neto and Pinto 1 suggest how to choose good values for the numerical
parameters that must be different for each type of iteration.
6. Simulations and Numerical Tests

To verify if the algorithm proposed is useful for this particular type of problem, three
maneuvers of station keeping were simulated, all of them having modifications in the
inclination. These results were compared with the ones obtained by the deterministic version,
without flexibility in the constraint’s satisfaction. Similar problems can be found in 27–29.
The first maneuver will occur with the data given in Table 1. The thrust level is 2.0 N. Table 2
shows the errors allowed in the final Keplerian elements of the orbit. The goal is to combine
a modification in semimajor axis, eccentricity, and inclination in a single maneuver.
Several numbers of “burning arcs” were used for the same maneuver, in order to
obtain some information about the importance of this parameter.
The consumptions found are showed in Table 3, as well as comparisons with
deterministic methods that we also simulated for comparison.
The second maneuver will occur with the data given in Table 4. Now, only the
inclination is changed. The thrust level is again 2.0 N. Table 5 shows the errors allowed in
the final Keplerian elements of the orbit.
The choice of the number of “burning arcs” was done for several different values, in
the same way that was made in the first maneuver.
The consumptions found are showed in Table 6, as well as comparisons with
deterministic methods.
The third maneuver occurs with the data given in Table 7. Now, inclination and
eccentricity are changed. The thrust level is again 2.0 N. Table 8 shows the errors allowed
in the final Keplerian elements of the orbit.
Table 1: Data for Maneuver 1.

Orbits Initial Final
Semimajor axis 7060.00 7090.00
Inclination degrees 1.00 0.50
Ascending node degrees 0.00 Free
Argument of perigee degrees 0.00 Free
Mean anomaly degrees 0.00 Free
Table 2: Errors allowed for final keplerian elements for Maneuver 1.
Semimajor axis 1.0 Km

Eccentricity 0.005
Inclination 0.001 deg
Table 3: Fuel expenditure comparisons kg for Maneuver 1.
Approach Stochastic Deterministic

2 Arcs 0.301 0.303
3 Arcs 0.300 0.301
4 Arcs 0.298 0.299
5 Arcs 0.297 0.298
6 Arcs 0.296 0.297
7 Arcs 0.295 0.296
8 Arcs 0.294 0.296
Table 4: Data for transfer Maneuver 2.

The number of “burning arcs” was again varied from two to eight, in the same way
that was made in the other maneuvers. The consumptions found are showed in Table 9, as
well as comparisons with deterministic methods.
Next, a new maneuver is considered to emphasize the effects of the errors allowed
for the final Keplerian elements. This maneuver considers the data shown in Table 1 for the
initial and final orbits, but increases the tolerable errors to the values shown in Table 10.
Several numbers of “burning arcs” were also used for this maneuver, to be compatible
with the previous studies. The consumptions found are showed in Table 11, as well as
the usual comparisons with the deterministic methods. Note that the difference in savings
increased, when larger values for the errors are allowed.

Eccentricity 0.005

2 Arcs 0.130 0.132
3 Arcs 0.128 0.130
4 Arcs 0.127 0.129
5 Arcs 0.127 0.128
6 Arcs 0.127 0.128
7 Arcs 0.127 0.128
8 Arcs 0.127 0.128
Table 7: Data for transfer Maneuver 3.


Eccentricity 0.005

2 Arcs 0.170 0.172
3 Arcs 0.167 0.169
4 Arcs 0.165 0.167
5 Arcs 0.163 0.166
6 Arcs 0.162 0.160
7 Arcs 0.162 0.160
8 Arcs 0.162 0.160

Eccentricity 0.05

2 Arcs 0.295 0.303
3 Arcs 0.294 0.301
4 Arcs 0.292 0.299
5 Arcs 0.291 0.298
6 Arcs 0.290 0.297
7 Arcs 0.290 0.296
8 Arcs 0.289 0.296
7. Conclusions
Optimal control was used to build an algorithm to search for solutions for the problem of
minimum fuel consumption to make orbital maneuvers for a satellite that needs to return to
its nominal orbit after deviations caused by perturbations forces.
In this research, emphasis was given in the problem considering station keeping with
out-of-plane maneuvers. The adjustments made in the algorithm, as well as in the parameters
used, allow us to get convergence in most of the cases for this version of the problem.
This problem took into account the accuracy in the constraint’s satisfaction by using
the nonlinear programming algorithm proposed by Rios Neto and Pinto 1.
The results showed that it is possible to reduce the costs by exploring tolerable errors
shown in Tables 2, 5, 8 and 10 in the constraint’s satisfaction. The amount saved can be
important in many cases.
It is also clear that increasing of the number of propulsion arcs decreases the fuel
costs. It can be seen from Tables 3, 6, 9 and 11, because the fuel consumption is smaller
for larger numbers of arcs. The reason is that increasing this number causes an increase in
the degrees of freedom available for the optimization technique. Since the maneuvers have
small amplitudes, this increase in the burning arcs has a limit in the savings, so the fuel
consumption reaches a constant value after a certain value.
Acknowledgments
The authors are grateful to Fundação de Amparo a Pesquisa do Estado de São Paulo,
FAPESP, Conselho Nacional de Desenvolvimento Cientifico e Tecnológico CNPq, and
Coordenação de Aperfeiçoamento de Pessoal de Nı́vel Superior CAPES for supporting this
research.
References
1 A. Rios Neto and R. L. U. Pinto, A Stochastic Approach to Generate a Projection of the Gradient Type
Method, INPE-4060-PRE/1019, INPE, Sao Jose dos Campos, Brazil, 1986.
2 A. F. B. A. Prado and A. Rios-Neto, “A stochastic approach to the problem of spacecraft optimal
maneuvers,” Revista Brasileira de Ciências Mecânicas, vol. 16, no. 3, pp. 268–278, 1994.
3 W. Hohmann, Die Erreichbarkeit der Himmelskorper, Munique, Oldenbourg, Germany, 1925.
4 C. Marchal, “Transferts optimaux entre orbites elliptiques coplanaires Durée indifférente,”
Astronautica Acta, vol. 11, no. 6, pp. 432–445, 1965.
5 H. Jin and R. G. Melton, “Transfers between circular orbits using fixed impulses,” in Proceedings of the
AAS/AIAA Spaceflight Mechanics Meeting, vol. 75 of Advances in the Astronautical Sciences, pp. 833–842,
Houston, Tex, USA, February 1991.
6 D. J. Jezewski and D. Mittleman, “An analytic approach to two-fixed-impulse transfers between
Keplerian orbits,” Journal of Guidance, Control, and Dynamics, vol. 5, no. 5, pp. 458–464, 1982.
7 R. F. Hoelker and R. Silber, “The bi-elliptic transfer between circular co-planar orbits,” Technical
Memorandum 2-59, Army Ballistic Missile Agency, Redstone Arsenal, Ala, USA, 1961.
8 A. Shternfeld, Soviet Space Science, Basic Books, New York, NY, USA, 1959.
9 L. Ting, “Optimum orbital transfer by several impulses,” Astronautica Acta, vol. 6, no. 5, pp. 256–265,
1960.
10 D. J. Jezewski and H. L. Rozendaal, “An efficient method for calculating optimal free-space N-
impulsive trajectories,” AIAA Journal, vol. 6, no. 11, pp. 2160–2165, 1968.
11 K. G. Eckel, “Optimal impulsive transfer with time constraint,” Acta Astronautica, vol. 9, no. 3, pp.
139–146, 1982.
12 D.F. Lawden, “Minimal rocket trajectories,” Journal of the American Rocket Society, vol. 23, no. 6, pp.
360–382, 1953.
13 H. B. Griffiths, “The fundamental group of two spaces with a common point,” Quarterly Journal of
Mathematics, vol. 5, no. 1, pp. 175–190, 1954.
14 V. M. Gomes and A. F. B. A. P. Prado, “Effects of the variation of the periapsis velocity in a swing-
by maneuver of a cloud of particles,” in Proceedings of the Recent Advances in Applied and Theoretical
Mechanics, pp. 106–108, Tenerife, Spain, 2009.
15 V. M. Gomes and A. F. B. A. Prado, “A numerical study of the dispersion of a cloud of particles,” in
Proceedings of the 20th International Congress of Mechanical Engineering (COBEM ’09), 2009.
16 V. M. Gomes and A. F. B. D. A. Prado, “The use of gravitational capture for space travel,” in Proceedings
of the 60th International Astronautical Congress (IAC ’09), vol. 2, pp. 1066–1071, Daejeon, South Korea,
2009.
17 V. M. Gomes and A. F. B. De Almeida Prado, “A study of the close approach between a planet and a
cloud of particles,” in Proceedings of the Recent Advances in Signal Processing, Robotics and Automation,
vol. 109, pp. 1941–1958, Cambridge, UK, 2002.
18 A. F. B. De Almeida Prado, “Numerical and analytical study of the gravitational capture in the
bicircular problem,” Advances in Space Research, vol. 36, no. 3, pp. 578–584, 2005.
19 A. F. B. D. A. Prado, “Numerical study and analytic estimation of forces acting in ballistic gravitational
capture,” Journal of Guidance, Control, and Dynamics, vol. 25, no. 2, pp. 368–375, 2002.
20 A. F. B .A. Prado and R. Broucke, “Transfer orbits in restricted problem,” Journal of Guidance Control
21 A. F. B. De Almeida Prado, “Third-body perturbation in orbits around natural satellites,” Journal of
22 A. P. M. Chiaradia, H. K. Kuga, and A. F. B. A. Prado, “Single frequency GPS measurements in real-
time artificial satellite orbit determination,” Acta Astronautica, vol. 53, no. 2, pp. 123–133, 2003.
23 M. C. B. Biggs, The Optimization of Spacecraft Orbital Manoeuvres. Part I: Linearly Varying Thrust Angles,
The Hatfield Polytechnic, Numerical Optimization Centre, 1978.
24 M. C. B. Biggs, The Optimisation of Spacecraft Orbital Manoeuvres. Part II: Using Pontryagin’s Maximun
Principle, The Hatfield Polytechnic, Numerical Optimisation Centre, 1979.
25 A. F. B. A. Prado and A. Rios-Neto, “Suboptimal and hybrid numerical solution schemes for orbit
transfer maneuvers,” SBA Controle & Automação, vol. 4, no. 2, pp. 82–88, 1975.
26 A. E. Bryson and Y. C. Ho, Applied Optimal Control, Wiley, New York, NY, USA, 1975.
27 V. M. Gomes, A. F. B. A. P. Prado, and H. K. Kuga, “Orbital maneuvers using low thrust,”
in Proceedings of the Recent Advances in Signal Processing, Robotics and Automation, pp. 120–125,
Cambridge, UK, 2009.
28 A. F. B. A. Prado, V. M. Gomes, and I. M. Fonseca, “Low thrust orbital maneuvers to insert a satellite in
a constellation,” in Proceedings of the 4th International Workshop on Satellite Constellations and Formation
Flying, Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil, February 2005.
doi:10.1155/2012/429479
Research Article
An Adaptive Remeshing Procedure for Proximity
Maneuvering Spacecraft Formations
Laura Garcia-Taberner1 and Josep J. Masdemont2

1
Departament d’Informàtica i Matemàtica Aplicada, Universitat de Girona, C/Lluı́s Santaló s/n,
17071 Girona, Spain
2
IEEC and Departament de Matemàtica Aplicada I, Universitat Politècnica de Catalunya, Diagonal 647,
08028 Barcelona, Spain
Correspondence should be addressed to Laura Garcia-Taberner, laura.garcia@ima.udg.edu
Copyright q 2012 L. Garcia-Taberner and J. J. Masdemont. This is an open access article

distributed under the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.
We consider a methodology to optimally obtain reconfigurations of spacecraft formations. It

is based on the discretization of the time interval in subintervals called the mesh and the
obtainment of local solutions on them as a result of a variational method. Applied to a libration
point orbit scenario, in this work we focus on how to find optimal meshes using an adaptive
remeshing procedure and on the determination of the parameter that governs it.
1. Introduction
Formation flying concepts have been growing in interest during the last years. In different
scenarios, formations or clusters of small satellites can perform like a virtual larger telescope,
obtaining equivalent information, but with a reduced cost. Mission concepts like the NASA
Terrestrial Planet Finder 1, the ESA Darwin 2, the NASA MAXIM 3, or the ESA XEUS 4
are just few examples that remark the importance of this new technology for space telescopes.
Nevertheless, formation flying still demands many new technologies to be successfully
implemented. Usually the spacecraft must be located and maintained within a very narrow
range of relative distances, and severe constraints like this, increasing the already high
complexity of the mission design see for instance the works of Farrar et al. 5 and references
therein. Also there are many other issues that need to be addressed as well. For instance, a
main one is collision avoidance when maneuvering or reconfiguring the formation. Typically,
from one observation to the next one, the formation needs to change the pointing goal and
eventually change its pattern. To this end, some representative techniques considered are
the computation of proximity maneuvers by means of artificial potential functions in the

works of Badawy and McInnes 6, rotation techniques introduced by Wang et al. 7, 8, or the
FEFF methodology based on a finite element implementation introduced by Garcia-Taberner
and Masdemont in 9.
In this paper we consider the FEFF methodology Finite Elements for Formation
Flight for the reconfiguration and proximity maneuvering of a formation about a libration
point. With this methodology, individual trajectories of the spacecraft inside the formation
are computed by means of solving a direct optimal control problem which is formulated
in terms of the discretization given by a finite element procedure. We briefly summarize
this procedure in Section 2. Then in Section 3 we focus on an adaptive remeshing strategy
that produces optimal discretizations for the trajectories, in the sense that the error in the
trajectories obtained is kept below a given threshold with the coarsest possible mesh. Finally,
the paper ends with some numerical examples and conclusions.
2. The FEFF Methodology

In this section we present a brief summary of the basics of the FEFF methodology that can be
found fully developed in 9, 10. This methodology was made with the purpose of system-
atically computing reconfigurations of spacecraft formations located in libration point orbits.
However, it could also be generalized for formations about the Earth or for in free space.
As it is well known, the vicinity of the Lagrangian points L1 and L2 is a very convenient
place for space observatories L1 for the Sun and L2 for deep space. In this paper we consider
a formation of spacecraft located in a halo orbit about L2 . We assume that the formation
is made of N spacecraft flying with a particular pattern and our objective is to perform a
reconfiguration in a fixed time T . We also assume that the spacecraft are in a small formation.
This is, the distance between them is only of the order of few hundreds of meters, both in the
initial and the final configurations. The objective of the FEFF methodology is to find a suitable
trajectory for each of the spacecraft that delivers it to the goal position, with minimum fuel
consumption and avoiding collisions with other spacecraft.
Since the formations are small with respect to the amplitude of the halo orbit, we
consider the linearized equations of motion about the nonlinear orbit. In 11 we have
studied the impact of the nonlinear part, concluding that, for orbits with a diameter of a few
hundreds of meters i.e., the usual length for a formation, these linearized equations model
the dynamics of the formation in a very good way.
Then, according to these hypotheses, associated to each spacecraft in the formation,
we have an equation of the form
Ẋt AtXt, 2.1
where At is a 6 × 6 matrix and X refers to the state of the spacecraft see The appendix.
Usually the origin of the reference frame for the X coordinates is the nominal point on the
base halo orbit at time t being the orientation of the coordinate axis parallel to the ones of the
RTBP model.
In order to perform the reconfigurations, we consider a control function for each of the
spacecraft and also we include the boundary conditions corresponding to the initial and final
states:
Ẋi t AtXi t Ui t,
Xi 0 X0i , 2.2
Xi T XTi .
Here X0i and XTi stand for the initial and final states of the ith spacecraft inside the formation,
and U1 , . . . , UN , are the controls we are searching with the aim of being optimal in terms of
fuel consumption.
The basis of the FEFF methodology is to use the finite element method in time to
discretize the spacecraft trajectories and to obtain the controls. Essentially, for each spacecraft,
the time interval 0, T which we consider for the reconfiguration is split in Mi subintervals
of the domain called elements. For a given trajectory, its mesh can have elements of different
lengths and of course different satellites can have associated different meshes. This will
depend on the nature of the trajectories of the reconfiguration and the path they follow. Using
this mesh we impose that controls are maneuvers applied at the points where two elements
join the nodes. The finite element methodology is used to formulate the problem and to
obtain, by means of this discretization, a relation between the states of each spacecraft at the
nodes of the elements and the Δv maneuvers applied. We note that in our discretization we
use elements with two nodes located at the ends of each element consecutive elements share
the connecting node. These elements are called linear elements and their associated trun-
cation error is according to the linear approximation taken about the nonlinear orbit when
considering 2.1. Usually, in the finite element methodology, the solution inside each element
is approximated by a polynomial and, for linear elements, this polynomial is of degree
one. Finite elements in time and greater orders have also been considered in more general
formulations of optimal control problems. An interesting presentation can be found in 12.
By means of the procedure FEFF, we reduce the reconfiguration problem to an optimal
control problem with constrains. The functional to be minimized is related to the fuel
consumption and is taken as the sum of the norm of the maneuvers:
&
N &
Mi
J1 ρi,k Δvi,k , 2.3
i1 k0
where || ∗ || denotes the Euclidean norm, vi,k is the maneuver applied at the kth node of
trajectory, and i and ρi,k are weighting parameters that can be used, for instance, to penalize
fuel consumptions of selected spacecraft with the purpose of balancing fuel resources. For
clarity, in 2.3 we consider that ρi,k multiplies the modulus of the maneuver, but in a similar
way we can impose a weight on each of its components.
An important issue in the formulation of the procedure is collision avoidance between
satellites. It enters in the optimization problem as constraints. We consider that each
spacecraft is surrounded by a security sphere that cannot intersect during the reconfiguration
process. Again, the discretization of the time interval made by the finite element methodology
provides an efficient implementation to check these constraints.
3. Adaptive Remeshing Applied to Reconfigurations

The objective of this paper is how to systematically obtain good meshes for the
reconfiguration problem. We note that, for a given mesh, the FEFF methodology computes
the trajectory of the spacecraft in such a way that J1 in 2.3 is minimized. But at the end, the
trajectories of the satellites have been obtained after a discretization process and the error of
the discretized approximated trajectories with respect to the exact solutions of the problem
is not obvious. If we take a small number of elements, we can have a poor model that is not
accurate enough. On the other side, as it is well known in the finite element methodology, the
approximated trajectories converge towards the true solution when the diameter of the mesh
the length of the longest time interval tends to zero.
Of course when we increase the number of elements in the mesh, it also increases
the complexity of the computations, the required CPU time, and the representation of the
solution someway. Also we could end up with ill conditioned problems when the number
of elements is very high due to the presence of very small maneuvers. To overcome these
difficulties, adaptive remeshing is a technique that allows us to work in the other way round.
Fixing an acceptable level of accuracy, and by means of an iterative procedure, one can find
“the coarsest mesh” providing approximate trajectories with the required accuracy.
Another issue we have to deal is related to the minimization of the functional 2.3.
Its derivatives are ill conditioned when one or more delta-v are near zero. Since the objective
is to find these maneuvers as small as possible, one may expect problems if we perform the
computations in a naive way.
We address these two facts in a two-step methodology. First we find an initial
approximation of the solution minimizing an alternative functional. In a natural way we have
chosen the functional:
&
N &
Mi
J2 ρi,k Δvi,k 2 , 3.1
i1 k0
because it is also directly related to fuel consumption and it does not have any ill-conditioning
problems when computing derivatives near zero.
Using this target functional, there are no problems in finding a solution; moreover, the
errors associated to a coarse mesh are not critical for the second step. Let us call FEFF-DV2
the procedure that provides this approximated solution. Starting with FEFF-DV2, we usually
consider a mesh with a small number of elements generally from 5 to 10 and we take all
of them of the same length. We note that since FEFF-DV2 is an optimization problem, we
need an initial seed. For this purpose we consider each spacecraft alone, this is, we solve N
independent problems, where we compute the trajectory which minimizes J2 without taking
into account possible collision risks.
Using the same discretization as in FEFF formulation, these initial seeds can be found
semianalytically by means of solving a linear system 10, The proof. The solution is unique
considering that the elements are all of the same length. Moreover, if the obtained trajectories
do not have collision risks the exclusion spheres do not intersect, they are already the output
of FEFF-DV2 i.e., the approximate solution for the given mesh that minimizes J2 .
In the second step of the procedure, we consider an adaptive remeshing strategy with
two purposes: to control the error due to the finite element methodology and to suppress
all the nodes that can give ill-conditioning problems in the minimization of J1 . Let us call
FEFF-DV1 the formulation that uses the FEFF methodology to optimize the functional J1 once
Initial coarse mesh FEFF-DV2 Estimate error
FEFF-DV Generate new mesh
Estimate error
Pass
(error below the threshold)
Adapted solution
Fail
Generate new mesh
Figure 1: Schema of the procedure of adaptive remeshing.
the nodes that could give ill-conditioned problems have been removed. Then the general
idea of the adaptive methodology follows the scheme displayed in Figure 1. Once we have
the approximation given by FEFF-DV2, we start the iterative second step which involves
FEFF-DV1 and an estimation of the error that is used to generate a new mesh. This iterative
procedure is repeated till the estimated error is below the given threshold requirement.
Finally let us comment more in detail how the adaptive remeshing works. The general
idea is that, given a threshold value e, we want to find a mesh that provides an approximate
solution with error understood as the difference between the solution of the problem and its
approximation inside of an element less than e in some norm.
Adaptive remeshing methods penalize the elements where the error is considered
big, dividing them into smaller elements. On the other hand, if the estimation of the error
is small in an element, this element is made bigger in the next iteration. Since, as we will
see, our estimation of the error is basically related to the value of the delta-v maneuvers to
be implemented; roughly speaking, this method tends to increase the length of the elements
which have associated small delta-v and tends to decrease the length of the elements which
have associated big delta-v’s. As a consequence, it is also suitable to avoid the ill-conditioned
problems that FEFF-DV1 might have.
Essentially, to decide whether the current mesh is good enough or not, we base on a
criterion which compares the modulus of the estimated error, ||e||, on the mesh with the total
gradient of the solution related in our problem with velocities. For this purpose we compute
the following integral by means of adding the results obtained in each element. We compute
T
u v2 dt, 3.2
0
where in each element we numerically propagate the initial condition at the starting point by
means of the dynamical equations, in order to obtain the velocity function v2 inside the kth
element. Then a Simpson quadrature is employed to compute the integral.
To get an estimation of the error inside a given element, we consider the former v2
velocities inside the element vk2 and the velocities vk1 obtained taking the derivative of the
approximate solution given by the finite element method as well inside this element a linear
function in our case. An estimator of the error inside the element is computed by means of
t 1/2
C C
C C k 1
ek Cvk1 − vk2 C vk1 − vk2 · vk1 − vk2 dt , 3.3

tk
where tk and tk 1 are the ends

of the kth element. From these values we have an estimation
of the error of the mesh: ||e|| e12 e22 · · · eM
2
, and we accept the mesh when
e ≤ νu, 3.4
where ν is the acceptability criteria, the threshold parameter control of the adaptive
remeshing procedure that will be discussed and tuned in the examples of Section 4.
In order to compute a new mesh when it is not accepted, we use the Li and Bettess
remeshing strategy see 13. This strategy is based on the idea that the error distribution on
an optimal mesh is uniform
νu
ek D ,
9 3.5
E
M
where ν is again the acceptability criteria, ek is the computed error on element k, M is

the number of elements of the mesh, and the hat distinguishes the parameters of the new
mesh to be generated. The strategy consists on finding the new length of the elements using
the number of elements of the new mesh, M. E According to Li and Bettess, if d denotes
the dimension of the problem and m the maximum degree of the polynomials used in the
interpolation for the approximate solutions inside an element k, then the number of elements
that should have the new mesh is
m d/2/m
&M
E νu−d/m
M ek d/m d/2
. 3.6
k1
Since we work with linear elements in dimension one, we have m 1 and d 1, and the
recommended number of elements of our new mesh is
3/2
&M
E νu−1
M ek 2/3
. 3.7
k1
Once we have the estimation of the number of new elements, we can find the length of them
by means of
1/m d/2
9k νu
h D hk , 3.8
E k
Me
that, in our case, turns out to be
3/2
9k νu
h D hk . 3.9
E k
Me
4. Simulations with Adaptive Remeshing

As it has been mentioned, in the following simulations we have located the formation in
the vicinity L2 in the Sun-Earth system. In particular we choose a halo orbit of 120000 km of
z-amplitude.
We have considered two limiting cases. The first one involves no collision risk. It is
known that in this case the optimal solution for each spacecraft is a bang-bang control and
in this section we see that our methodology converges towards it. We note that this is the
most critical case for our procedure, since the optimal maneuver is not a continuous function
but it consists of two impulsive delta-v: one at departure and another one at the arrival
position. The remaining nodal delta-v must be zero, and consequently this is a case where
the computation of derivatives for J1 is very ill conditioned.
The second limiting case corresponds to reconfigurations with collision risks. In this
case the simple bang-bang controls would cause collision between the spacecraft, and the
FEFF methodology solves the problem tending to low thrust solutions when the diameter of
the mesh tends to zero. This is, the methodology can cope with both impulsive and smooth
function controls selecting the optimal one for each case or part of the trajectory.
With the purpose of calibrating the acceptability parameter ν, in this section we present
some simulations with reconfigurations in different situations involving and not involving
collision risks.
4.1. A Bang-Bang Simulation Considering a Single Spacecraft

When the reconfiguration maneuver is not affected by collision risk for one or more spacecraft
of the formation, these satellites follow independent trajectories i.e., the collision avoidance
constraints in fact will not be active. So we can consider a formation just consisting of a single
spacecraft to exemplify the procedure.
Let us consider in this example a shift of a single spacecraft. The reference frame for
2.2 is aligned with respect to the RTBP one but with origin on the nominal point of the base
halo orbit which has been taken of 120000 km in the z-amplitude. When t 0, this point
on the halo orbit corresponds to the “upper” position, this is, when it crosses the RTBP plane
Y 0 with Z > 0. The initial condition for this example is taken 100 meters far from the base
nominal halo orbit in the X direction, and the goal is to transfer it to a symmetrical position
with respect to the halo orbit in 8 hours. This is to 100 meters in the opposite X direction
0.2
v (cm/s)
0.1
0
0 4 8
Time (hours)
Figure 2: Delta-v obtained in the minimization of the J2 functional 3.1 in a case of no collision risk.
performing a total shift of 200 meters during the transfer maneuver. For this particular case
we know that the optimal solution is a bang-bang control with maneuvers of 0.69 cm/s at
departure and arrival.
Our procedure starts with FEFF-DV2 minimizing the J2 functional 3.1 and obtaining
a trajectory with the delta-v profile of Figure 2. In fact, as we have discussed previously, since
there are no collision risks, this optimal solution corresponds to the initial seed we provide
for FEFF-DV2. Moreover, if we do not take into account the magnitude of the maneuvers for
this particular example, the delta-v profile displayed in Figure 2 is the usual one we find in
similar situations. As expected, since it does not correspond to the optimal solution of the J1
functional 2.3, it is not a bang-bang control.
Using this solution as the initial seed, we start the second part of the procedure corres-
ponding to the iterative part in the schema of Figure 1. It involves FEFF-DV, the estimation of
the error for the current mesh and the generation of the new one. In Figure 3 we can see the
delta-v profile that we obtain after the iterations one and three; this last one is already very
close to the bang-bang control the method converges after 5 iterations.
Of course in a real situation, and specially for small thrusters, the maximum value of
delta-v may be constrained. In Figure 4 we show the delta-v profile for the same simulation
but constraining its maximum value. Values of 0.4 cm/s and 0.3 cm/s have been chosen. We
see how the methodology splits the optimal impulsive bang-bang control of about 0.69 cm/s
in longer arcs at the end points of the trajectory.
4.1.1. Considerations and Calibration of the ν Parameter

Let us consider now the impact of the parameter ν in the performance of the procedure. We
note that this parameter does not only appear in the acceptability criteria 3.4, but it is also
used to obtain the new mesh in 3.9.
Intuitively one can expect that if we take a small value of ν, we could end up with a
mesh with a big number of nodes, which turns into an optimization problem with a very large
number of variables. If we take into account that to a mesh of 100 elements we have associated
an optimization problem of 594 variables, we could end up with unsolvable problems in
0.6 0.6
v (cm/s)
v (cm/s)
0.3 0.3
0 0
0 4 8 0 4 8
Time (hours) Time (hours)
a b
Figure 3: Delta-v profile obtained with the minimization of the J1 functional 2.3 in a case without collision
risk. a we have the result after the first iteration and b we show the result after three iterations.
0.6 0.6
v (cm/s)
v (cm/s)
0.3 0.3
0 0
0 4 8 0 4 8
a b
Figure 4: Delta-v profile obtained with the minimization of the J1 functional 2.3 for a case without
collision risk and constraining the maximum delta-v allowed. a we have imposed a maximum delta-
v of 0.4 cm/s and b a maximum value of 0.3 cm/s.
practice. On the other way round, if we use a big ν, we could end up accepting some meshes
with big errors. In Table 1, we have a summary of the results obtained for different values of
the parameter ν, the number of iterations needed to reach the bang-bang solution, and the
number of elements after the first iteration of the methodology.
We note that when ν is very small, convergence can fail. The case with ν equal to
0.0001 makes the optimal procedure awkward. When ν is 0.001, the final number of elements
is greater than 1 that we know is our final target number although the maneuvers associated
to the central nodes are very small. Moreover, when ν is big, there is no convergence: the final
mesh contains more elements than expected because it passes the acceptability criteria before
converging to the bang-bang control. For this bang-bang case, we can conclude that the best
values for ν are inside the range 0.04, 0.06.
Table 1: Number of iterations necessary to obtain the bang-bang solution of the first example depending
on the parameter ν. We have indicated by “fail” the cases where the procedure does not converge and in
M1 the number of elements in the first iteration.
ν 0.0001 0.001 0.002 0.005 0.01 0.02 0.03 0.04 0.05 0.06 0.07
M1 3008 301 149 61 31 15 11 7 6 4 4
Iter Fail Fail 25 16 14 10 6 4 4 2 Fail
4
80 3
2
1
Z (m)
0
5
−80
80
−80 0
0 Y (m)
−80
X (m) 80
a b
Figure 5: Example of reconfiguration with collision risk: the switch of two pairs of spacecraft of the TPF
formation. a shows the schema of the reconfiguration and b is a solution of the reconfiguration obtained
with the FEFF methodology.
4.2. A Simulation Using the TPF Formation

For this case we consider a configuration based on the Terrestrial Planet Finder TPF model
see 1. We assume that the satellites are initially contained in the local plane Z 0, with
the interferometry baseline aligned on the X axis. The length of the baseline is 150 meters.
We simulate the swap between two pairs of satellites in the baseline: each inner satellite
changes its location with the outer satellite which is closest to it in position inner satellites are
maneuvered to attain outer positions and vice versa as shown in Figure 5. Again we consider
8 hours for the reconfiguration maneuver. The process of swapping positions is affected of
collision risk for any radius of the sphere of influence, and simple bang-bang controls are no
longer valid. We have considered a sphere of radius 10 meters, and the FEFF methodology
obtains solutions of the form displayed in Figure 5. We note that the convergence of the
methodology does not depend on the radius of the sphere. The reconfiguration cost increases
with the radius. The final number of elements also increases with the radius.
A discussion for the parameter ν similar to the one in the previous example is also
valid here: using a small ν, we can end up with a mesh with too many elements. For example,
taking ν 0.0005, in the first iteration we have around 1000 elements and we do not have only
the problem of having very small elements. The optimization problem has 29970 variables
something that it is not desirable at all. Also, if we take a big ν, we end up with a mesh with
very few elements and a big truncation error.
In Table 2 we display a summary of the results obtained for different values of the
parameter ν including the number of iterations till the methodology converges Iter, and the
number of elements in the first and last iterations, M1 and MF . Due to symmetry reasons,
the number of elements in each spacecraft trajectory is the same. We note that now the best
values are inside the range 0.005, 0.05 and that the value ν 0.05 is appropriated for the two
Table 2: Number of iterations and elements involved in the swapping example of TPF depending on the
parameter ν. M1 refers to the number of elements in the first iteration and MF to the final ones.
ν 0.0001 0.001 0.002 0.005 0.01 0.02 0.03 0.04 0.05 0.06 0.07
M1 3504 350 175 70 34 18 12 9 6 6 5
Iter Fail 10 8 8 7 6 4 3 3 3 2
MF 232 202 171 89 45 33 27 15 9 7
Figure 6: Example of reconfiguration with mixed bang-bang and collision risk: the swap of a pair of
spacecraft and a shift.
cases studied. Since the shift case and the swapping case are someway the building blocks
of the reconfiguration maneuver, we could suggest that ν 0.05 is a good value for the
computation of reconfiguration maneuvers by means of adaptive remeshing.
4.3. Mixed Case Simulation: 3 Spacecraft

We consider in this section a case that would demand both a bang-bang and low thrust arc in
the reconfiguration maneuver. The formation consists of 3 spacecraft which are in the same
plane as shown in Figure 6. The reconfiguration is the swap of two spacecraft and the shift of
the third one. If we perform the transfers sequentially in time, the maneuver decouples into
two independent problems like the ones considered in the previous examples bang-bang
plus swap. However, we are going to consider all the transfers in parallel in the same time
interval, this is, with a collision risk of the three satellites in the center of the formation.
Again, we have a similar discussion about the parameter ν and the simulation results
are summarized in Table 3. In this case, the values of ν that are good for our purposes are
inside the range 0.002, 0.05.
4.4. Summary of Considerations about the Value of ν

In the previous sections we have seen that a desirable value for the adaptive remeshing
control parameter ν should be in the range 0.005, 0.05. This range already gives us an idea
Table 3: Number of iterations needed to obtain the mixed solution depending on ν.
ν 0.0001 0.001 0.002 0.005 0.01 0.02 0.03 0.04 0.05 0.06 0.07
M1 3305 331 165 66 33 17 11 8 7 6 5
Iter Fail Fail 36 27 21 16 12 6 5 Fail Fail
Table 4: Mean value of the number of iterations as a function of ν for the 25 test bench reconfigurations
considered in the simulation.
ν 0.005 0.01 0.02 0.03 0.04 0.05 0.055 0.06
Iter. 10.2 8.4 7.1 4.2 3.7 3.2 4.3 5.2
of the value of ν that we should choose when using adaptive remeshing for the computation
of reconfiguration maneuvers by means of FEFF.
We have applied the reconfiguration procedure to a test bench of 25 reconfigurations
which include swaps between spacecraft located at opposite vertices of polygons 6 cases,
swaps in the TPF formation 9 cases, and parallel shifts 10 cases. Different sizes and
number of spacecraft, from 3 to 10, have been considered. Ten of the reconfigurations would
be converging to a bang-bang solution while the other 15 would converge to low-thrust arcs
when the diameter of the mesh tends to zero. We have considered our methodology taking
different values of ν, and we have computed the mean of the number of iterations of the
adaptive process necessary to converge. The obtained results are summarized in Table 4 and
point again to the value of ν 0.05 as a convenient parameter for this kind of proximity
maneuver computations.
5. Conclusions
This paper presents an adaptive remeshing strategy applied to a methodology to find
trajectories for reconfigurations of spacecraft formations. The procedure adapts the mesh
in a systematic and optimal way, and a suitable value for the parameter controlling the
procedure has been found. Moreover, the methodology is robust in all the ranges of possible
reconfiguration cases: from the ones that should result in a bang-bang control to the ones that
should be performed with low thrust arcs.
Appendix
Let us consider the usual restricted three-body problem RTBP in synodic coordinates, where
the unit of mass and length is taken such as the sum of the masses of the primaries and the
distance between the primaries is equal to 1, and the unit of time is taken such as the period
of the primaries is equal to 2π. In this synodic coordinate system, the origin is located on the
center of mass and the x axis is defined by the line of the two primaries, from the smallest
primary to the larger one. The z axis is normal to the rotation plane, in the direction of the
angular momentum of the primaries, and the y axis is chosen orthogonal to the previous ones
in order to have a positively orientated coordinate system.
Using this reference frame, the equations of motion of the RTBP are
∂Ω ∂Ω ∂Ω
ẍ − 2ẏ , ÿ 2ẋ , z̈ , A.1
∂x ∂y ∂z
where Ωx, y, z x2 y2 /2 1 − μ/r1 μ/r2 1 − μμ/2, μ is the mass of the small
primary, and r1 and r2 are the distances from the spacecraft to the big and small primaries
respectively.
Writing A.1 as a system of first order equations, ẋ fx, we have that x
x1 , x2 , x3 , x4 , x5 , x6 is the state vector x, y, z, ẋ, ẏ, ż, and f is given by
f1 x ẋ,
f2 y ẏ,
f3 z ż,
∂Ω
f4 x 2ẏ , A.2
∂x
∂Ω
f5 x −2ẋ ,
∂y
∂Ω
f6 x .
∂z
The reference system we consider in this paper for 2.1 is parallel to the one of the
RTBP but with origin in a halo orbit; that is, it is a time-dependent translation of the synodic
RTBP one given by
Xt xt − xh t, A.3
where xh t is the current state on the chosen halo orbit.

Linearizing then ẋ fx about the halo orbit, we have Ẋ ẋh fxh Dfxh X, and
since xh t is a solution of ẋ fx, we obtain Ẋ Dfxh X, which defines At Dfxh t
in 2.1.
Acknowledgments
This research has been supported by the MICINN-FEDER Grant, MTM2009-06973, and the
CUR-DIUE Grant 2009SGR859.
References
1 The TPF Science Working Group, The Terrestrial Planet Finder: A NASA Origins Program to Search for Ha-
bitable Planets, JPL Publication, 1999.
2 The Science and Technology Team of Darwin and Alcatel Study Team, Darwin. The Infrared Space Interf-
erometer. Concept and Feasibility Study Report, ESA-SCI 12, European Space Agency, 2000.
3 K. C. Gendreau, N. White, S. Owens, W. Cash, A. Shipley, and M. Joy, “The MAXIM X-ray inter-
ferometry mission concept study,” in Proceedings of the 36th Liège International Astrophysics Colloquium,
pp. 11–16, Liège, Belgium, July, 2001.
4 A. N. Parmar, G. Hasinger, M. Arnaud, X. Barcons, D. Barret, and A. Blanchard, “XEUS—the X-ray
evolving universe spectroscopy mission,” in X-Ray and Gamma-Ray Elescopes and Instruments for Astro-
nomy, pp. 304–313, 2003.
5 M. Farrar, M. Thein, and D. C. Folta, “A comparative analysis of control techniques for formation fly-
ing spacecraft in an earth/moon-sun L2-centered lissajous orbit,” AIAA Paper 2008-7358, 2008.
6 A. Badawy and C. R. McInnes, “On-orbit assembly using superquadric potential fields,” Journal of
7 P. K. C. Wang and F. Y. Hadaegh, “Minimum-fuel formation reconfiguration of multiple free-flying
spacecraft,” Journal of the Astronautical Sciences, vol. 47, no. 1-2, pp. 77–102, 1999.
8 C. Y. Xia, P. K. C. Wang, and F. Y. Hadaegh, “Optimal formation reconfiguration of multiple spacecraft
with docking and undocking capability,” Journal of Guidance, Control, and Dynamics, vol. 30, no. 3, pp.
694–702, 2007.
9 L. Garcia-Taberner and J. J. Masdemont, “Maneuvering spacecraft formations using a dynamically
adapted finite element methodology,” Journal of Guidance, Control, and Dynamics, vol. 32, no. 5, pp.
1585–1597, 2009.
10 L. Garcia-Taberner, Proximity maneuvering of libration point orbit formations using adapted finite element
methods, Ph.D. dissertation, Universitat Politècnica de Catalunya, Barcelona, Spain, 2010.
11 L. Garcia-Taberner and J. J. Masdemont, “FEFF methodology for spacecraft formations reconfigura-
tion in the vicinity of libration points,” Acta Astronautica, vol. 67, no. 7-8, pp. 810–817, 2010.
12 M. Vasile, “Finite elements in time: a direct transcription method for optimal control problems,” AIAA
Paper 2010-8275, 2010.
13 L.-Y. Li, P. Bettess, J. W. Bull, T. Bond, and I. Applegarth, “Theoretical formulations for adaptive finite
element computations,” Communications in Numerical Methods in Engineering, vol. 11, no. 10, pp. 857–
868, 1995.
doi:10.1155/2012/967248
Research Article
Closed Relative Trajectories for Formation Flying
with Single-Input Control
Anna Guerman,1 Michael Ovchinnikov,2 Georgi Smirnov,3

and Sergey Trofimov4
1
Centre for Aerospace Science and Technologies, Department of Electromechanical Engineering, University
of Beira Interior, Calçada Fonte do Lameiro, 6201-001 Covilhã, Portugal
2
Orientation and Motion Control Department, Keldysh Institute of Applied Mathematics, Miusskaya pl. 4,
Moscow 125047, Russia
3
Centre of Physics, Department of Mathematics and Applications, School of Sciences, University of Minho,
Campus de Gualtar, 4710-057 Braga, Portugal
4
Department of Control and Applied Mathematics, Moscow Institute of Physics and Technology, Institutskij
per. 9, Dolgoprudny, Moscow 141700, Russia
Correspondence should be addressed to Georgi Smirnov, smirnov@math.uminho.pt
Received 13 July 2011; Revised 11 November 2011; Accepted 15 November 2011
Copyright q 2012 Anna Guerman et al. This is an open access article distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and
We study the problem of formation shape control under the constraints on the thrust direction.
Formations composed of small satellites are usually subject to serious limitations for power
consumption, mass, and volume of the attitude and orbit control system AOCS. If the purpose
of the formation flying mission does not require precise tracking of a given relative trajectory,
AOCS of satellites may be substantially simplified; however, the capacity of AOCS to ensure a
bounded or even periodic relative motion has to be studied first. We consider a formation of two
satellites; the deputy one is equipped with a passive attitude control system that provides one-
axis stabilization and a propulsion system that consists of one or two thrusters oriented along
the stabilized axis. The relative motion of the satellites is modeled by the Schweighart-Sedwick
linear equations taking into account the effect of J2 perturbations. We prove that both in the case
of passive magnetic attitude stabilization and spin stabilization for all initial relative positions and
velocities of satellites there exists a control guaranteeing their periodic relative motion.
1. Introduction
Nowadays, design of formation flying missions is one of the main directions of modern
space system development. Many studies have been carried out, and a number of books on
dynamics of such distributed systems have been published see, e.g., 1, 2.
One of the main problems to be solved in design of a formation flying mission is that of
maintenance of the required spatial configuration of satellites. The straightforward approach
is to correct orbits using one or several thrusters. Usually no constraints are imposed on the
thrust direction. However, for a nano- or picosatellite formation subject to severe restrictions
on mass, volume, and energy resources, the number of thrusters is limited, and the available
control systems rarely provide three-axis orientation. Therefore, the thrust direction cannot
be arbitrary changed.
Consider a two-satellite formation the aim of which is to perform measurements or
observations, at several points of the orbit. Suppose that the deputy satellite is equipped
with a propulsion system with its thrust axis fixed in the body of satellite. The thrust can
be directed in both ways or in only one, depending on the propulsion system employed.
As the simplest example of such a system, one can suggest a cold gas thruster. A number
of simple and lightweight attitude control systems are available that can stabilize motion
of the thrust axis. Thus one can formulate the problem of orbital control assuming the
thrust axis orientation to be known at any moment in time. In control theory, the above
control is referred to as single-input control. The principal question is whether the above
AOCSs suffice to provide the required formation shape at least at some points of the
orbit.
The cases of successively implemented single-input control are known since the
early days of space exploration. One of them occurred by accident as a result of hull
depressurization during one of the first Veneras, Soviet Venus probe missions. The spacecraft
was spinning in a sun-oriented mode, and so the average jet force of the leaking air happened
to be directed towards the Sun, resulting unexpectedly in the proper orbital correction
3.
Development of modern miniature satellites, such as Cubesats, motivates research
on single-input control to simplify satellite control system. Applications of the single-input
control concept to the problem of formation maintenance have been considered for several
missions. For example, the microsatellite Magion-2 launched in 1989 was equipped with a
passive one-axis magnetic attitude control system and a propulsion system with a thrust
vector along the oriented axis. The aim was to keep it at 10 km distance from the chief satellite
4; however, due to the thruster failure, formation maintenance was not possible.
Much research is focused on compensation of the relative drift of satellites caused by
the J2 harmonic of the Earth’s gravitational potential. In 5 this problem is studied assuming
the deputy satellite to be equipped with a passive magnetic attitude control system and two
thrusters installed along the axis of the magnet. The use of solar radiation pressure to solve
this problem is studied in 6 see also 7.
Another approach to the decoupling of the attitude and orbital control in formation
is presented in 8. The authors interpret a formation as a quasirigid body. It is shown that
control of such a formation can be effectively separated into a control torque that maintains
the attitude and control forces that maintain the rigidity of a formation. The respective control
strategy is based on the Lyapunov controller synthesis 9.
In this paper, we analyze the general problem of compensation of J2 perturbations
for the deputy satellite in two-satellite formation. The chief satellite is assumed to move
passively. We study the Schweighart-Sedwick linear equations 10, that is, the modification
of the Hill-Clohessy-Wiltshire equations of relative motion. This modification well describes
the effect of J2 perturbations and has been successfully used to study many problems of
relative dynamics, such as formation keeping and rendezvous see, e.g., 11–14.
We consider two different types of single-input control:
1 bilateral control oriented along a vector fixed in the inertial space the case of spin
stabilization;
2 bilateral and unilateral control oriented along the vector of local geomagnetic field
the case of passive magnetic stabilization.
We prove that for any initial conditions there exists a control that provides a periodic
relative motion of chief and deputy satellites with a period T between 1 and 2 orbital periods.
This means that the maximum distance between satellites does not become very large.
Though the shape of relative trajectory is not controlled, the existence of bounded short-
period relative motion suffices to perform the required measurements in many nano- and
picosatellite formation missions.
Throughout this paper, the set of real numbers is denoted by R and the N-dimensional
space of vectors with components in R by RN . We denote by a, b
the usual scalar product
in RN and by · the Euclidean norm. The transposition of a matrix A is denoted by AT .
2. Existence and Stabilization of Closed Trajectories for

a Single-Input Control System
Consider a linear single-input control system
η̇t Aηt at wtbt, ηt ∈ Rn , wt ∈ R, 2.1
where A is a N × N-matrix, a : R → RN and b : R → RN are given continuous functions,

and wt is a control. The control wt may be subjected to the constraint
wt ≥ 0. 2.2
The set of admissible controls w· is denoted by W and consists of locally integrable
functions. The general solution to 2.1 is given by the Cauchy formula
t
ηt etA η0 et−sA as wsbsds. 2.3
0
We say that system 2.1 has a T -closed trajectory η· satisfying η0 η0 , if and only if there
exists an admissible control wη0 · such that
T

η0 e TA
η0 eT −tA at wη0 tbt dt. 2.4
0
Put
= T F
T −tA
KT e wtbtdt | w· ∈ W . 2.5
0
If wt ∈ R, then KT is a subspace. In the case wt ≥ 0, the set KT is a convex cone. If
KT RN , then for any η0 η0 there exists an admissible control wη0 · satisfying 2.4.
Moreover, for any initial point η1 and any terminal point η2 there exists an admissible control
wη1 ,η2 such that
T

η2 e TA
η1 eT −tA at wη1 ,η2 tbt dt, 2.6
0
that is, the system is controllable.

The established controllability allows one to correct closed trajectories, that is, if there
is a deviation in the initial condition of the closed trajectory, it can be compensated for by an
appropriate choice of control.
To verify the controllability condition KT RN in the case of unconstrained control,
we use the following direct consequence of the Pontryagin Maximum Principle.
Theorem 2.1. Assume that there is no nontrivial solution of the equation
ṗt −AT pt 2.7
satisfying
pt, bt
0, t ∈ 0, T , 2.8
then the equality KT RN holds.
In the case of controls subject to constraint 2.2, the situation is more involved.
Later on we consider only the τ-periodic functions b·. This assumption is satisfied in all
applications considered here and significantly simplifies the study.
The following two propositions are well known to the specialists in the control theory.
However, to make the presentation self-contained, we include their short proofs in the
Appendix.
First of all, note that the periodicity condition bt τ bt implies that the cones
KMτ , M 1, 2, . . ., form a monotonously increasing sequence.
Lemma 2.2. Assume that bt is τ-periodic. Let M be a positive integer. Then the inclusion KMτ ⊂
KM 1τ holds.
The next theorem is also a consequence of the Pontryagin Maximum Principle and
contains sufficient conditions of controllability for system 2.1 when the control satisfies
condition 2.2.
Theorem 2.3. Assume that there is no nontrivial solution to the differential equation
ṗt −AT pt 2.9

satisfying
pt, bt
≥ 0, t ≥ 0. 2.10
G
Then the equality K∞ M KMτ RN holds.
G
Since the sequence of convex cones KMτ is monotonous, the equality M>0 KMτ RN
implies the existence of a positive integer M such that
KMτ RN . 2.11
Indeed, let points ξk , k 1, . . . , N 1, be the vertices of a simplex Ξ containing the origin as an

6
interior point. Then any point ξ ∈ RN can be represented as ξ k λk ξk with λk ≥ 0. For any
k 1, . . . , N 1, there exist a positive integer Mk and an admissible control uk · satisfying
Mk τ
ξk eMτ−tA btuk tdt. 2.12
0
So from Lemma 2.2 we see that any vertex ξk can be represented in the form
Mτ
ξk eMτ−tA btwk tdt, 2.13
0
where M max{Mk | k 1, . . . , N 1} and wk · is an admissible control. This implies the

equality
Mτ &
ξ eMτ−tA bt λk wk tdt, 2.14
0 k
arriving at 2.11.
Let η0 ∈ RN . Condition 2.11 leads to the existence of a control w0 · such that
Mτ Mτ
η0 − eMτA η0 − eMτ−tA atdt eMτ−tA btw0 tdt. 2.15
0 0
Therefore, the control w0 · corresponds to a closed trajectory of 2.1 satisfying η0 η0 .
The above results permit one also to compensate for the errors caused by the model or
measurements not requiring considerable computational efforts. Under condition 2.11 it is
easy to develop an algorithm that reaches the point η0 even if the initial point η0 is different
from η0 . Note that this algorithm does not require solving the integral equation 2.6.
Consider a simplex Σ containing η0 in its interior. Let {η1 , . . . , ηN 1} be the vertices of Σ.

Condition 2.11 implies the existence of admissible controls wk ·, k 1, . . . , N 1, satisfying
the equalities
Mτ Mτ
η0 − eMτA ηk − eMτ−tA atdt eMτ−tA btwk tdt, k 1, . . . , N 1. 2.16
0 0
If η0 ∈ Σ, there exist nonnegative numbers λk , k 1, . . . , N 1, such that
&
N 1 &
N 1
η0 λk ηk , λk 1, 2.17
k1 k1
and so the control
&
N 1
wt λk wk t 2.18
k1
drives system 2.1 to the point η0 . Thus, if the controls wk ·, k 1, . . . , N 1, are known, it
suffices to find nonnegative numbers λk , k 1, . . . , N 1, satisfying 2.17 in order to reach
the point η0 from η0 .
3. Equations of Relative Motion with Single-Input Control

To take into account the influence of the J2 -harmonic on relative motion of two satellites with
close near-circular orbits, the following modification of the Hill-Clohessy-Wiltshire equations
has been introduced by Schweighart and Sedwick 10:
ẍ 2ncż wtex t,

ÿ q y 2lq cos qt φ wtey t,
2
3.1

z̈ − 2ncẋ − 5c2 − 2 n2 z wtez t.
The linearization is done with respect to the circular reference orbit with the mean motion n.
Here x, y, and z are coordinates in the respective orbital reference frame Oxyz. The axes are
chosen in the following way: Oz indicates the radial direction outwards from the Earth, Ox is
directed along the velocity of the point O, and y is normal to the orbital plane. The coefficients
c, q, l, and φ are properly defined constants see the appendix, Proof of Lemma 3.1.
The direction of the control acceleration wt is defined by the vector function
T
et ex t, ey t, ez t . 3.2
Using the notations
T
η x, y, z, ẋ, ẏ, ż ,
T
at 0, 0, 0, 0, 2lq cos qt φ , 0 ,
T
bt 0, 0, 0, ex t, ey t, ez t ,
⎛ ⎞
0 0 0 1 0 0 3.3
⎜0 0 0 1 0 ⎟
⎜ 0 ⎟
⎜0 0 0 0 1 ⎟
⎜ 0 ⎟
A⎜ ⎟,
⎜0 0 0 0 0 −2nc⎟
⎜ ⎟
⎝0 −q2 ⎠
20 2 0 0 0
0 0 5c − 2 n 2nc 0 0
we obtain a system of type 2.1. System 2.7 that describes the evolution of the vector p
p1 , p2 , p3 , p4 , p5 , p6 T takes the form
ṗ1 0,
ṗ2 q2 p5 ,

ṗ3 − 5c2 − 2 n2 p6 ,
3.4
ṗ4 −p1 − 2ncp6 ,
ṗ5 −p2 ,
ṗ6 −p3 2ncp4 .
Its general solution is given by
p1 t p10 ,

p2 t A2 cos qt φ2 ,
2 D
5c − 2 n 5c2 − 2 0
p3 t p30 − A6 √ sin 2 − c2 nt φ6 2nc p t,
2 − c2 2 − c2 1
p0 2c D 5c2 − 2
p4 t 3 − A6 √ sin 2 − c2 nt φ6 p0 t, 3.5
2nc 2 − c2 2 − c2 1
A2
p5 t − sin qt φ2 ,
q
D 2cp10
p6 t A6 cos 2 − c2 nt φ6 − ,
2 − c2 n
where p10 , p30 , A2 , A6 , φ2 , and φ6 are constants. Conditions 2.8 and 2.9 are equivalent to the
conditions
p4 tex t p5 tey t p6 tez t 0, t ∈ 0, T , 3.6
p4 tex t p5 tey t p6 tez t ≥ 0, t ∈ 0, T , 3.7

respectively. According to Theorem 2.3, to prove the controllability of the Schweighart-

Sedwick system with single-input control it suffices to show that there are no nontrivial
functions p1 t, . . . , p6 t satisfying 3.6 or 3.7.
Denote the radius and the inclination of the reference circular orbit by rref and iref ,
respectively. Assume that the chief satellite moves passively in an orbit with inclination i1 .
The orbit inclination of the deputy satellite is denoted by i2 . Set
D D
ω0 nc, ω1 q, ω2 2 − c2 − c n, ω3 2 − c2 c n. 3.8
The following lemma proved in the Appendix is crucial for the analysis of the Schweighart-
Sedwick system controllability.
arccos−1/3, then ωj /
Lemma 3.1. If 2iref / 0, j 0, 1, 2, 3, and ω2 < ω0 < ω1 < ω3 .
Below we assume that the main condition of this lemma is satisfied and consider two
systems with single-input control relevant for practical applications.
4. Bilateral Control Oriented along the Geomagnetic Field

Consider first a formation with the deputy satellite equipped with a passive magnetic attitude
control system PMACS and has two thrusters installed along its axis of orientation i.e., axis
of permanent magnet included in PMACS in opposite directions. We also assume that at any
moment in time this axis coincides with the direction of geomagnetic field described by the
direct dipole model:
cos θt sin i2

ex t
,
1 3 sin θtsin i2
2 2
cos i2
ey t
, 4.1
1 3 sin2 θtsin2 i2
−2 sin θt sin i2

ez t
.
1 3 sin2 θtsin2 i2
The argument of latitude is given by θt nct.

Under some nonrestrictive conditions the system is controllable in any time interval
0, T ; for example, one can take T 2π/nc.
Theorem 4.1. Let T > 0. If sin 2i2 / 0, then there exists a T -closed trajectory of system 2.1.
Moreover, an error in the initial conditions can be compensated for.
The proof of this theorem can be found in the appendix.

5. Bilateral Control Oriented along a Fixed Vector in the Inertial space

Spin-stabilized satellites represent another interesting possibility for orbit correction by
single-input control. Once rapidly rotated about an axis, the spacecraft keeps spinning around
this direction in the inertial space in the absence of perturbing torques.
Assume that the deputy satellite possesses a spherically symmetrical mass distribu-
tion, is spin stabilized, and has two thrusters oriented in opposite directions along its spin
axis fixed in the inertial space. Suppose that λ is the angle between this axis and the vector
pointing to the vernal equinox direction, and ε is the inclination of the plane containing
these vectors with respect to the Earth’s equator. Then in the Earth-centered inertial reference
frame the spin axis direction has the components cos λ, sin λ cos ε, sin λ sin εT . In the Oxyz
reference frame the expressions are
ex t σz sin θt − σx cos θt,

ey −σy , 5.1
ez t −σx sin θt − σz cos θt.
Here the vector σ σx , σy , σz T defines the direction of spin axis in the ascending node of
the orbit via the inclination i2 and the right ascension Ω2 :
σx cos Ω2 cos i2 sin λ cos ε − sin Ω2 cos i2 cos λ sin i2 sin λ sin ε,
σy − cos Ω2 sin i2 sin λ cos ε sin Ω2 sin i2 cos λ cos i2 sin λ sin ε, 5.2
σz cos Ω2 cos λ sin Ω2 sin λ cos ε.
We set θt nct. As in the case of the satellite oriented along the local geomagnetic field,
under some nonrestrictive conditions the system is controllable in any time interval 0, T ,
for example, for T 2π/nc.
Theorem 5.1. Let T > 0. If σx2 σz2 / 0 and σy / 0, then there exist a T -closed trajectory of system
2.1. Moreover, an error in the initial conditions can be compensated for.
See the appendix for the proof.
6. Unilateral Control Oriented along the Geomagnetic Field

Now assume that the control wt has to satisfy the nonnegativity condition 2.2. Set τ
2π/nc. In this case we have the following result.
Theorem 6.1. If sin 2i2 / 0, there is a positive integer M > 0 and a Mτ-closed trajectory of system
2.1. Moreover, an error in the initial conditions can be compensated for.
The theorem is proved in the Appendix.

Note that a similar result can be proved for the case of the satellite oriented along a
fixed vector in the inertial space. However, this result is of quite limited practical importance.
Indeed, while in the case of magnetic orientation M 2 see the numerical example in the
next section, in the case of the satellite oriented along a fixed vector in the inertial space, the
150
100
50
0
z m
−50
−100
−150
−200
300
200
100 600
y 0 400 500
300
m −100 100 200
0
x m
Figure 1: Closed trajectory of linearized system LT: bilateral control with magnetic ACS.
value of M is very large, and so is the distance between the chief and deputy satellites. In this
case the linearized equations cease to describe adequately the system dynamics and so the
generated periodic trajectories are of merely academic interest.
7. Numerical Results
The aim of the following numerical simulations is to verify the analytical results listed above
and to compare the trajectories of initial and linearized systems in the presence of the control.
On solving the integral equation 2.4 numerically, we substitute the obtained control into
the Gauss variational equations for the deputy satellite and propagate them in time. For
the passively flying chief satellite the propagation can be done directly. Then, subtracting
one motion from another, we convert the result to the Oxyz reference frame. Only the J2
perturbing effect is taken into account. Indeed, for time intervals of several orbital periods
the influence of atmospheric drag and solar radiation pressure on relative motion of identical
satellites in close orbits is negligible 3 to 4 orders smaller than J2 perturbations.
Integral equation 2.4 has many solutions; we use the minimal one in the sense of L2 -
norm. This criterion can be interpreted as that of minimal energy consumption for low-thrust
constant-power engines see, e.g., 15.
Below, we compare the trajectories of the linearized Schweighart-Sedwick system
LT and the trajectories obtained by integration of the nonlinear equations of motion NT.
Figure 1 shows a T -closed LT with the numerically obtained bilateral single-input control
oriented along the geomagnetic field. This trajectory has a length T τ 2πnc−1 and
corresponds to the following initial conditions: x0 70.71 m; y0 70.71 m; z0 35.36 m;
ẋ0 76.25 mm/s; ẏ0 76.32 mm/s; ż0 −38.07 mm/s. The radius of the circular reference
orbit is rref 7000 km; the inclination of the chief satellite i1 is the same as the reference
inclination iref 35 deg. The projections of LT on xy and xz planes are demonstrated in
Figures 2 and 3. As we see, the shape of trajectories is rather complex. Modelling errors of LT
and the corresponding control are shown in Figures 4 and 5, respectively.
Figure 4 shows that the difference between the LT and NT obtained with the same
control is not significant. The difference appears because the in-plane drift is not completely
eliminated. It is caused by the errors of linearization in the Schweighart-Sedwick model.
In the case of free flight, these errors can be compensated for by a proper choice of initial
conditions, which should be done numerically see 10. We obtain a similar situation with
250
200
150
100
y m
50
−50
−100
0 100 200 300 400 500 600
x m
Figure 2: Projection on xy plane.
150
100
50
0
z m
−50
−100
−150
−200
0 100 200 300 400 500 600
x m
Figure 3: Projection on xz plane.
12
10
6
Error m
0
−2
−4
0 1000 2000 3000 4000 5000 6000
Time s
x
y
z
Figure 4: Coordinate-wise modelling error.

×10−3
1. 5
Control (m/s2 )
0. 5
−0. 5
−1
−1. 5
0 1000 2000 3000 4000 5000 6000
Time (s)
Figure 5: Control.
Table 1: Results of simplex experiment.
Number of vertex 1 2 3 4 5 6 7
x0 , m 70.71 −141.42 70.71 70.71 70.71 70.71 70.71
y0 , m 70.71 70.71 −141.42 70.71 70.71 70.71 70.71
z0 , m 35.36 35.36 35.36 −70.71 35.36 35.36 35.36
ẋ0 , mm/s 76.25 76.25 76.25 76.25 −152.51 76.25 76.25
ẏ0 , mm/s 76.32 76.32 76.32 76.32 76.32 −152.64 76.32
ż0 , mm/s −38.07 −38.07 −38.07 −38.07 −38.07 −38.07 76.15
T/τ 2 2 2 2 2 2 2
w1 , m/s 0.16 0.16 0.14 0.18 0.08 0.15 0.16
w22 , 10−4 m2 /s3 6.83 6.84 5.35 7.93 1.64 5.59 6.84
the controlled flight, but now the control has to be corrected; for example, it can be used as
the first approximation in an iteration procedure such as Newton’s method applied to the
Gauss system. A Newton-type method suitable to solve control problems with nonnegativity
constraints can be found in 16. The control problem for nonlinear system is to be described
in a future paper.
Now proceed with the case of unilateral single-input control oriented along the
geomagnetic field. The construction described in Section 2 is fulfilled numerically. We show
that it is possible to construct a 2τ-closed trajectory for all vertices of a simplex containing the
origin in its interior. Therefore a 2τ-closed trajectory exists for any initial point. The results of
the “simplex” experiment are summarized in Table 1. The last two rows of this table contain
the values of L1 -norm
2τ
w1 |wt|dt 7.1
0
and of squared L2 -norm
2τ
w22 |wt|2 dt, 7.2
0
400
300
200
100
z m
0
−100
−200
−300
−400
−500
400
200
0
0 500 1000 1500
−200
−1500 −1000 −500
y m
x m
Figure 6: Closed LT: unilateral control with magnetic ACS.
300
250
200
150
100
y m
50
0
−50
−100
−150
−200
−1500 −1000 −500 0 500 1000 1500
x m
Figure 7: Projection on xy plane.
of the controls wt providing closed trajectories for the simplex vertices 1, . . . , 7. 2τ-closed
LT along with their projections on xy and xz planes, the coordinate-wise errors, and the
corresponding nonnegative LT-control are shown in Figures 6, 7, 8, 9, 10. Since the time
interval becomes twice as long, the error of linearization results in larger modelling errors
due to considerable along-track drift.
For the bilateral control oriented along a fixed vector in the inertial space, the results
are qualitatively similar to the case of bilateral magnetic control see Figures 11, 12, 13. We
use the same initial conditions, and the fixed vector is defined by the following angles: λ
45 deg, ε 23.45 deg. This choice of ε may correspond to stabilization of the spacecraft axis
in the Sun direction.
8. Conclusions
We consider the problem of formation maintenance under constraints on the thrust vector
directions. The formation consists of two satellites; the deputy satellite is equipped with one
or two thrusters oriented along a given axis. We assume that the orientation of this axis is
kept by an available passive ACS and is known at any instant of time. A possibility to obtain a
periodic relative motion of the chief and deputy satellites is demonstrated for several types of
400
300
200
100
z m
0
−100
−200
−300
−400
−500
−1500 −1000 −500 0 500 1000 1500
x m
Figure 8: Projection on xz plane.
40
30
20
10
Error m
−10
−20
−30
0 2000 4000 6000 8000 10000 12000
Time s
x
y
z
Figure 9: Coordinate-wise modeling error.
×10−4
8
7
6
Control (m/s2 )
5
4
3
2
1
0
−1
0 2000 4000 6000 8000 10000 12000
Time (s)
Figure 10: Control.

200
150
100
50
z m
0
−50
−100
−150
−200
200
100 800
0 600
−100 200
400
y m −200 0
x m
Figure 11: Closed LT: bilateral control along axis fixed in absolute space.
×10−4
6
4
Control (m/s2 )
2
0
−2
−4
−6
−8
0 1000 2000 3000 4000 5000 6000
Time (s)
Figure 12: Control.
4
Error m
−2
−4
−6
0 1000 2000 3000 4000 5000 6000
Time s
x
y
z
Figure 13: Coordinate-wise modelling error.

single-input control. In each case sufficient controllability conditions are deduced. In general,
these conditions can be formulated as follows: the vector of control direction should have
nonzero components both in the orbital plane and along the normal to the orbit. For the
unilateral control oriented along the geomagnetic field, the existence of a closed trajectory
of relative motion with double period is established for arbitrary initial conditions. A single-
input control numerically obtained for the system of Schweighart-Sedwick equations suffices
to guarantee almost closed trajectories. We also prove that the inaccuracy caused by the errors
of the Schweighart-Sedwick model can be corrected.
Appendix
Proof of Lemma 2.2. Let z ∈ KMτ . Then an admissible control u· exists such that
Mτ
z eMτ−tA btutdt. A.1
0
Set
=
0, s ∈ 0, τ,
ws A.2
us − τ, s ∈ τ, M 1τ.
Then we have
Mτ
z eMτ−tA btwτ tdt
−τ
Mτ τ
eMτ τ−sA bs − τwsds A.3
0
M 1τ
eM 1τ−sA bswsds.
0
Thus we get KMτ ⊂ KM 1τ .
Proof of Theorem 2.3. Suppose that K∞ / RN . From Lemma 2.2 we see that K∞ is a convex
/ 0 satisfying x, p∞
≥ 0, for all x ∈ K∞ . Therefore, we have
cone. So there exists a vector p∞
x, p∞
≥ 0, for all x ∈ KMτ and any positive integer M. Consider the functions

exp AT Mτ − t p∞
pM t C C , M 1, 2, . . . . A.4
Cexp AT Mτ p∞ C
From the Pontryagin maximum principle we have pM t, bt

0, t ∈ 0, Mτ, if wt ∈ R,
and pM t, bt
≥ 0, t ∈ 0, Mτ, if wt ≥ 0. Consider the sequence pM 0. Without loss of
generality it converges to a vector p0 satisfying p0 1. Thus we have

lim pM t p0 t exp −AT t p0 , t ∈ 0, Mτ. A.5
M→∞
Obviously the solution p0 · to 2.7 is nontrivial and satisfies 2.8 if wt ∈ R, and 2.9 if
wt ≥ 0, a contradiction.
Proof of Lemma 3.1. By definition,

3nJ2 R2⊕ cos i1 − cos i2 cot i1 sin i2 cos ΔΩ0 − cos i2
q nc 2
cos i2 −
2
,
2rref sin2 ΔΩ0 cot i1 sin i2 − cos i2 cos ΔΩ0 2
A.6
y0 √ 3J2 R2⊕
ΔΩ0 , y0 y0, c 1 s, s 2
1 3 cos 2iref ,
rref sin iref 8rref
where R⊕ is the Earth’s radius, J2 ≈ 10−3 is the second zonal harmonic. Since the orbits of
satellites are close, the difference i2 − i1 is small. Taking into account sin 2i1 /
0 and sin 2i2 /
0,
we have
3nJ2 R2⊕
q ≈ nc 2
cos2 i2 . A.7
2rref
arccos−1/3, one can see that c /

Since 2iref / 1. At the same time |c − 1| 1. Thus, all the
frequencies ωj , j 0, 1, 2, 3, are nonzero and pairwise different: ω2 < ω0 < ω1 < ω3 .
Proof of Theorem 4.1. Indeed, in this case condition 3.6 reads
&
3

κt g: t cos ω0 t gk cos ωk t hk sin ωk t ≡ 0. A.8
k0
From Lemma 3.1 we see that all the frequencies are pairwise different, and therefore the
coefficients of the quasipolynomial κt equal zero. Since
5c2 − 2 0
g: p sin i2 ,
2 − c2 1
p0
g0 3 sin i2 ,
2nc
4cp10
h0 sin i2 ,
2 − c2 n
A2
g1 − cos φ2 cos i2 ,
q
A2
h1 − sin φ2 cos i2 ,
q

c
g 2 A6 1 − √ sin i2 sin φ6 ,
2 − c2

c
h2 A6 1 − √ sin i2 cos φ6 ,
2 − c2

c
g3 −A6 1 √ sin i2 sin φ6 ,
2 − c2

c
h3 −A6 1 √ sin i2 cos φ6
2 − c2
A.9
/ 0, we obtain pt ≡ 0. Hence we have KT RN , T > 0. This implies

from the condition sin 2i2
the existence of a T -closed trajectory for any initial point.
Proof of Theorem 5.1. The proof is almost identical to that of the previous theorem. Indeed,
condition 3.6 of Theorem 2.1 takes the form
&
3

: sin ω0 t
κt g: t cos ω0 t ht gk cos ωk t hk sin ωk t ≡ 0. A.10
k0
As in the previous proof, one can apply Lemma 3.1 and see that all the frequencies are
pairwise different. Consequently the coefficients of the quasipolynomial κt equal zero. Since
5c2 − 2 0
g: − p σx ,
2 − c2 1
: 5c − 2 p0 σz ,
2
h
2 − c2 1
p0 2cp10
g0 − 3 σx σz ,
2nc 2 − c2 n
p0 2cp10
h0 3 σz σx ,
2nc 2 − c2 n
A2
g1 σy sin φ2 ,
q
A.11
A2
h1 σy cos φ2 ,
q

1 c
g2 −A6 √ σz cos φ6 − σx sin φ6 ,
2 2 − c2

1 c
h2 A6 √ σx cos φ6 σz sin φ6 ,
2 2−c 2

1 c
g3 −A6 −√ σz cos φ6 σx sin φ6 ,
2 2−c 2

1 c
h3 A6 −√ σz sin φ6 − σx cos φ6 ,
2 2−c 2
from the conditions σx2 σz2 / 0, we obtain pt ≡ 0. Thus we have KT RN , T > 0. This
0, σy /
implies the existence of a T -closed trajectory for any initial point.
G
Proof of Theorem 6.1. Assume that M KMτ / RN . Then, by Theorem 2.3, there exists a
nontrivial solution to 2.7 satisfying
p4 tex t p5 tey t p6 tez t ≥ 0, t ∈ 0, ∞. A.12
Condition A.12 takes the form
&
3

κt g: t cos ω0 t gk cos ωk t hk sin ωk t ≥ 0, t ∈ 0, ∞, A.13
k0
where the coefficients are those defined in the proof of Theorem 4.1. Now we show that the
coefficients of the quasipolynomial κt are equal to zero. Indeed, we have

κt 1
0≤ g: cos ω0 t O , t −→ ∞. A.14
t t
So g: 0, and we obtain
&
3

κt gk cos ωk t hk sin ωk t ≥ 0. A.15
k0
0, j 0, . . . , 3. Multiplying κt by
From Lemma 3.1 we have ω2 < ω0 < ω1 < ω3 and ωj /
1 ± cos ωj t, j 0, 1, 2, 3, we get
T
1 gj
0 ≤ lim κt 1 ± cos ωj t dt ± . A.16
τ →∞T 0 2
Multiplying κt by 1 ± sin ωj t, j 0, 1, 2, 3, we obtain
T
1 hj
0 ≤ lim κt 1 ± sin ωj t dt ± . A.17
τ →∞T 0 2
Therefore all of the coefficients of κt are equal to zero. As in the proof of Theorem 4.1, we
have pt ≡ 0, a contradiction.
Acknowledgments
The authors are grateful to Arun Misra for the bibliographical support and to the
reviewers for their comments and valuable suggestions. This research is supported by the
Portuguese Foundation for Science and Technologies FCT, the Portuguese Operational
Programme for Competitiveness Factors COMPETE, the Portuguese Strategic Reference
Framework QREN, the European Regional Development Fund FEDER, and by the
Russian Foundation for Basic Research RFBR, Grant 09-01-00431.
References
1 K. T. Alfriend, S. R. Vadali, P. Gurfil, J. P. How, and L. S. Breger, Spacecraft Formation Flying. Dynamics,
Control and Navigation, Elsevier Astrodynamics Series, Elsevier, 2009.
2 P. A. Capo-Lugo and P. M. Bainum, Orbital Mechanics and Formation Flying: A Digital Control
Perspective, Woodhead Publishing, 2011.
3 A. Platonov, “On motion synthesis in ballistics and mechatronics,” in Applied Celestial Mechanics and
Motion Control, T. Eneev, M. Ovchinnikov, and A. Golikov, Eds., pp. 127–222, KIAM, Moscow, Russia,
2010.
4 Magion 2. NASA Website, NSSDC Master Catalog Search, 2011, http://nssdc.gsfc.nasa.gov/nmc/
masterCatalog.do?sc1989-080B.
5 M. Ovchinnikov, G. V. Smirnov, and I. Zaramenskikh, “Orbital corrections by a single-input impulsive
control applied along the geomagnetic field,” Acta Astronautica, vol. 65, no. 11-12, pp. 1826–1830, 2009.
6 M. Ovchinnikov and I. E. Zaramenskikh, “Eliminating of two satellites relative drift caused by the
earth oblateness,” Cosmic Research, vol. 49, no. 1, pp. 65–71, 2011.
7 G. V. Smirnov, M. Ovchinnikov, and A. Guerman, “Use of solar radiation pressure to maintain a
spatial satellite formation,” Acta Astronautica, vol. 61, no. 7-8, pp. 724–728, 2007.
8 C. Blake and A. Misra, “Dynamics and control of satellite formations using a quasi-rigid body
formulation,” in Proceedings of the AAS/AIAA Space Flight Mechanics Meeting, Galveston, Tex, USA,
2008.
9 C. Blake and A. Misra, “Constrained reorientation of satellite formations using a quasi-rigid body
formulation,” in Proceedings of the 5th IAF Workshop on Constellations and Formation Flying, Evpatoria,
Ukraine, 2008.
10 S. A. Schweighart and R. J. Sedwick, “High-fidelity linearized J2 model for satellite formation flight,”
11 R. Bevilacqua, M. Romano, and F. Curti, “Decoupled-natural-dynamics model for the relative motion
of two spacecraft without and with J2 perturbation,” Nonlinear Dynamics and Systems Theory, vol. 10,
no. 1, pp. 11–20, 2010.
12 R. Bevilacqua, J. S. Hall, and M. Romano, “Multiple spacecraft rendezvous maneuvers by differential
drag and low thrust engines,” Celestial Mechanics & Dynamical Astronomy, vol. 106, no. 1, pp. 69–88,
2010.
13 T. Reid and A. Misra, “Formation flying of satellites in the presence of atmospheric drag,” Journal of
Aerospace Engineering, Sciences and Applications, vol. 3, pp. 64–91, 2011.
14 S. A. Gilani and P. L. Palmer, “Analysis of fidelities of linearized orbital models using least squares,”
in Proceedings of the IEEE Aerospace Conference, Big Sky, Mont, USA, 2011.
15 J. P. Marec, Optimal Space Trajectories, Elsevier, 1979.
16 S. Dias and G. Smirnov, “On the Newton method for set-valued maps,” Nonlinear Analysis, vol. 75,
pp. 1219–1230, 2012.
doi:10.1155/2012/926158
Research Article
Modified Chebyshev-Picard Iteration Methods for
Station-Keeping of Translunar Halo Orbits
Xiaoli Bai and John L. Junkins

Department of Aerospace Engineering, Texas A&M University, TAMU 3141, College Station,
TX 77843-3141, USA
Correspondence should be addressed to Xiaoli Bai, xiaolibai@tamu.edu
Received 15 July 2011; Revised 7 November 2011; Accepted 14 November 2011
Copyright q 2012 X. Bai and J. L. Junkins. This is an open access article distributed under the
The halo orbits around the Earth-Moon L2 libration point provide a great candidate orbit for a lunar
communication satellite, where the satellite remains above the horizon on the far side of the Moon
being visible from the Earth at all times. Such orbits are generally unstable, and station-keeping
strategies are required to control the satellite to remain close to the reference orbit. A recently
developed Modified Chebyshev-Picard Iteration method is used to compute corrective maneuvers
at discrete time intervals for station-keeping of halo orbit satellite, and several key parameters
affecting the mission performance are analyzed through numerical simulations. Compared with
previously published results, the presented method provides a computationally efficient station-
keeping approach which has a simple control structure that does not require weight turning and,
most importantly, does not need state transition matrix or gradient information computation. The
performance of the presented approach is shown to be comparable with published methods.
1. Introduction
For the spatial circular restricted three-body problem CR3BP, where the two large bodies
move in planar circular orbits about their center of mass and a third body of negligible
mass moves under only the 1/r 2 gravitational influence of the large bodies, there exist five
stationary points in the rotating reference frame. These points, called the Lagrangian points or
libration points, have unique roles for many scientific missions because they lie in the plane of
the primary bodies’ motion and have fixed positions with respect to these two bodies, similar
to the geostationary orbits although a little more complicated. Among the five points, three
of them are collinear with the other two bodies, and the motion near these points is unstable;
the other two libration points with the primaries form equilateral triangles in the plane of
motion of the two large bodies and the motions near these two points are neutrally stable.
There has been a lot of interest in finding and controlling the periodic orbits near the
collinear libration points. Farquhar originally proposed using the orbits near the L2 point of
the Earth-Moon system for communication with the far side of the Moon 1. By appropriate
station-keeping and with a modest cost, the satellite is visible from the Earth all the time.
However, such orbits are not generally periodic. Later, Farquhar and Kamel found that if
the amplitude of the in-plane motion is large enough so that the nonlinear effects become
significant, purely periodic three dimensional orbits, or halo orbits, exist that are permanently
visible from the Earth 2. The interest in the Earth-Moon halo orbits later shifted to the Earth-
Sun system following the end of the Apollo missions and the launch of International Sun-
Earth Explore-3 ISEE3 which is the first libration-point satellite 3.
The techniques for station-keeping of halo orbits can be classified as 1 using
continuous control, or 2 generating impulses at discrete time intervals. Breakwell et al.
studied station-keeping for translunar communication station using a continuous feedback
control to minimize a cost function, which is a weighted combination of position deviation
from the nominal orbit and the control acceleration based on a linearization about the
reference orbit 4. Xin et al. developed a suboptimal closed-form continuous feedback
controller by approximately solving a Hamiltonian-Jacobian-Bellman equation 5. Kulkarni
et al. applied the H∞ approach for station-keeping of the halo orbit around the L1 of the
Earth-Sun system 6. Cielaszyk and Wie modeled the nonlinearities of the halo orbit problem
as persistent disturbances and applied disturbance accommodation with a linear quadratic
regulator for the control 7. Although using continuous control derived from optimal control
theory usually generates orbits closer to their reference orbits than using impulses, discrete
maneuvers have the advantages of less complexity and risk and higher precision orbit
estimation can be obtained during coasting arcs. Dunham and Roberts described the station-
keeping strategies for three Earth-Sun libration point missions, all of which used discrete
controls 8. Two station-keeping techniques for the Earth-Moon libration point orbits were
studied by Gómez et al. 9. One is what they called the target point strategy where a cost
function, which is a weighted summation of the position and velocity deviations from several
future points and the fuel cost, is minimized to find corrective maneuvers. Another approach
they studied is based on Floquet methods which combine invariant manifold theory and
Floquet modes to only eliminate the unstable component of the error. Howell and Pernicka
also studied using the target point strategy for the Earth-Sun system in a later paper 10. The
optimal spacing time between impulsively applied controls has been studied by Renault and
Scheeres 11, and the optimal time to update control law with continuous control has been
recently studied by Gustafson and Scheeres 12.
A newly developed numerical computation approach, Modified Chebyshev-Picard
Iteration MCPI method, is used in this paper for station-keeping to maintain a halo
orbit in the Earth-Moon system. This orbit provides an excellent parking orbit for a lunar
communication satellite. Details about MCPI methods can be found in the papers by Bai
and Junkins 13, 14 and Bai’s dissertation 15. Fusing Chebyshev polynomials with the
classical Picard iteration method, MCPI methods iteratively refine an orthogonal function
approximation of the entire state trajectory. MCPI methods can solve both initial value
problems IVPs and two-point boundary value problems BVPs by constraining the coeffi-
cients of the Chebyshev polynomials. A unique characteristic of MCPI is that the Chebyshev
coefficients are constrained linearly without approximation on each Picard iteration. As a
consequence, shooting techniques to impose boundary conditions are not required. Although
perhaps the most striking feature about MCPI methods is their naturally parallel structure
because computation of the integrand along each path iteration can be rigorously distributed
over many parallel cores with negligible cross communication needed, MCPI methods are
computationally efficient even prior to parallelization according to the results reported by
Bai and Junkins, 13, 14 where MCPI methods have been compared with several classical
methods in solving IVPs and BVPs. In several examples, it has been demonstrated that both
greater efficiency and accuracy can be obtained, compared to the pseudospectral method, for
example. The MCPI method can be applied to either the impulsive or the continuous control
case; in this paper, we will provide only the case of impulsive control.
This paper is structured as follows. We first present the numerical approach to generate
the reference halo orbit. We use the third order analytical formulations derived by Richardson
16 to provide a starting guess for the halo orbit, and then use a differential-correction
approach to numerically find the accurate initial conditions to generate the periodic halo
orbit. The reference orbit is integrated using MCPI method, and the Chebyshev coefficients
that precisely represent the nominal orbit are saved for future reference. We then use MCPI
methods to generate impulses at fixed discrete time intervals for corrective maneuvers.
Several important criteria such as fuel cost and deviations from the reference orbit are
analyzed through simulations and we compare our results with previous published results.
Conclusions and future directions are presented at the end.
2. The Reference Orbit

2.1. Equations of Motion
As is well known, a good nominal reference orbit reduces the fuel cost for unnecessary
maneuvers, so the dynamical model used to generate the reference orbit ideally should
include the gravitational perturbation forces from the Sun and other planets as well as
solar radiation and other forces. Farquhar and Kamel presented fairly detailed equations of
motion where the effects of the Sun’s gravitation force, solar pressure, and the Moon’s orbital
eccentricity were considered 2. Gómez et al. 9 used JPL DE403 to include the gravitational
influences of the Sun, Moon, and all nine planets. To illustrate our novel approach, we
have chosen to only consider the CR3BP idealization of the Earth-Moon system. Because
the station-keeping technique we present later is a numerical approach to solve for corrective
maneuvers and is essentially an open loop control, using a more detailed dynamical model
will not change the structure of the algorithm but may moderately affect the performance of
the presented technique. For real mission operations, detailed dynamical models should be
used to generate the reference orbit as well as the corrective maneuvers.
Most of the development of the equations of motion can be found in the book by
Schaub and Junkins 17. One difference here is that we use the L2 Lagrangian point as the
origin. Figure 1 shows how the rotating reference frame E : {9er , e9θ , e93 } is defined. m1 , m2 ,
and m represent the Earth, Moon, and spacecraft, respectively. Note the vectors denoting the
location of the Earth, the Moon, and the center of mass are defined such that r1 r1 e9r , which
results in r1 , r2 , and rCM all being negative quantities since the vectors extend from the origin
along the negative e9r axis. The two central bodies rotate in circles about their center of mass
at a constant angular velocity, defined by
Gm1 m2
ω2 3
, 2.1
r12
ξ r
1
ξ
2
ri
eθ
m1
m2
er
r2 L2
r CM
r1
r 12
Figure 1: System Diagram.
where r12 is the distance between m1 and m2 . Thus, the E frame in use rotates at a constant
velocity ω with respect to the inertial frame, as given by the following equation:
ω ω9e3 . 2.2
The inertial position vector of the spacecraft is defined as follows:
ri rx − rCM 9er ry e9θ rz e93 , 2.3
where rx , ry , and rz are the components of the r vector. The inertial derivative of this equation
can be taken using the transport theorem 17 and 2.2 to produce the inertial velocity vector:

ṙi ṙx e9r ṙy e9θ ṙz e93 ω × rx − rCM 9er ry e9θ rz e93
2.4
ṙx − ry ω e9r ṙy rx − rCM ω e9θ ṙz e93 .
Finally, the above equation can be differentiated once more to produce the inertial accelera-
tion vector:

r̈i r̈x − 2ṙy ω − rx − rCM ω2 e9r r̈y 2ṙx ω − ry ω2 e9θ r̈z e93 . 2.5
The massless force due to gravity from the two massive bodies is defined in E frame compo-
nents as

m1 m2
Fx −G rx − r1 3 rx − r2 ,
ξ13 ξ2

m 1 m2
Fy −G 3 ry , 2.6
ξ13 ξ2

m 1 m2
Fz −G 3 rz ,
ξ13 ξ2
with ξi defined as follows:

ξi rx − ri 2 ry2 rz2 . 2.7
When combined with 2.5, the equations of motion can be written as

m1 m2
r̈x − 2ṙy ω − rx − rCM ω G 2
rx − r1 3 rx − r2 0,
ξ13 ξ2

m 1 m2
r̈y 2ṙx ω − ry ω G
2
3 ry 0, 2.8
ξ13 ξ2

m 1 m2
r̈z G 3 rz 0.
ξ13 ξ2
The equations of motion are nondimensionalized by first using a new time variable, τ, de-
fined as
τ ωt, 2.9
and then using the distance between the Earth and L2 point |r2 |, which leads to the new
distance variables as
rx ry rz r1 r2
x , y , z , x1 , x2 ,
|r2 | |r2 | |r2 | |r2 | |r2 |
2.10
rCM ξi r12
xCM , ρi , x12 ,
|r2 | |r2 | |r2 |
and lastly, the masses are eliminated in favor of a nondimensionalized mass ratio μ, defined
as
m2
μ . 2.11
m1 m2
Now, rearrange 2.8 and divide by the distance r2 and also by the factor ω2 , yielding:

◦◦ ◦ G m1 m2
x 2y x − xCM − 2 x − x1 3 x − x2 ,
ω ξ13 ξ2

◦◦ ◦ G m1 m2
y −2x y − 3 y, 2.12
ω2 ξ13 ξ2

◦◦ G m1 m2
z − 2 3 z.
ω ξ13 ξ2
Replacing m2 by μm1 m2 and m1 by 1 − μm1 m2 using 2.11, and introducing ρi , the

equations of motion become

◦◦ ◦ Gm1 m2 1−μ μ
x 2y x − xCM − x − x1 3 x − x2 ,
ω2 r23 ρ13 ρ2

◦◦ ◦ Gm1 m2 1 − μ μ
y −2x y − 3 y, 2.13
ω2 r23 ρ13 ρ2

◦◦ Gm1 m2 1 − μ μ
z − 3 z.
ω2 r23 ρ13 ρ2
Combining 2.13 with 2.1 yields the final, nondimensionalized equations of motion:

◦◦ ◦ 1−μ μ
x 2y x − xCM − 3
x12 x − x1 x − x2 ,
ρ13 ρ23

◦◦ ◦ 1−μ μ
y −2x y − x12
3
y, 2.14
ρ13 ρ23

◦◦ 1−μ μ
z −x12
3
z.
ρ13 ρ23
Using 2.15, the nondimensionalized distance x, which is the distance from the center of
mass to the L2 point, is calculated:
1−μ μ
x− 2 − 2 0. 2.15
μ x x−1 μ
The other values, r1 , r2 , and rCM , are found using the center of mass equation:
m1 r1 − rCM m2 r2 − rCM m1 m2 r12 . 2.16
Table 1 provides the values of the three parameters used in this model.
2.2. Finding Initial Conditions

A third-order closed-form solution for the equations of motion is given in Richardson’s paper
16. We use Ay 45000 km which is defined as the amplitude of the linearized motion along
the y direction in Richardson’s paper 16. This is the same magnitude used in the paper by
Breakwell et al. 4, at which position a satellite is visible from the Earth all the time. The
approximate initial conditions obtained from the third order formula quickly diverge when
propagated using the full equations of motion due to the inherent instability of the system.
A differential corrections method 18 was used to find the initial conditions that lead to
Table 1: Parameters used.

Parameter Value
Earth mass 5.976 × 1024 kg
Moon mass 7.3477 × 1022 kg
r12 0.3844 × 106 km
Table 2: Approximate initial conditions.
Variable Nondimensional value

xt0 −0.390895010335809
zt0 0.353556629315019
vy t0 1.554577497503360
Tp 3.336429964438981
a bounded periodic orbit. In order to find a halo orbit one important property of halo orbits
was exploited: halo orbits display symmetry about the x-z plane. This means that at the point
where the orbit crosses that plane y 0, the x and z components of velocity are both zero.
As such, choosing a point on the x- z plane as the initial conditions reduces the variable
set from three position components x, y, z, three velocity components vx , vy , vz , and the
orbit period T to two position components x, z, one velocity component vy , and the
orbit period T . Additionally, after a single orbit, or after half an orbit, y, vx , and vy will all
be zero again. Table 2 gives the initial conditions we found, where the physical time of the
period is Tp ≈ 14.5 days. Note that the solution accuracy in the numerical correction process
to find these initial conditions has been set as 10−15 .
2.3. Using MCPI Method to Save the Reference Orbit

A nice feature about MCPI methods is that the achieved solutions are represented to high
precision in the form of orthogonal Chebyshev polynomials, thus if the coefficients of the
states are saved, the state values at arbitrary time can be obtained straightforward by
numerically computing the orthogonal polynomials and then a simple inner product matrix
multiplication. We integrate the reference orbit using MCPI method for one orbit and save
the coefficients of the position and velocity for future station-keeping reference. An important
parameter to choose here is the order of the polynomials to use. Generally, higher order
approximation leads to higher accuracy solutions but requires larger storage space to save
all the coefficients. We use closure error after one period of the reference orbit, defined in
2.17, as the accuracy criterion to find the optimal order to use:

Ep S Tp − ST0 , 2.17
where STp is the state values position or velocity after one complete orbit and ST0
is the state values position or velocity at the initial time. Figure 2a shows how the
nondimensional closure errors change with the order, and Figure 2b shows the dimensional
results. Both figures demonstrate the spectral accuracy that the MCPI method achieved,
Nondimensional closure error versus order Closure error versus order

102 105
100
10−2 100
S(Tp )− S(T0 )
S(Tp )− S(T0 )
−4
10
10−6 10−5
−8
10
10−10 10−10
10−12
10−14 10−15
10 20 30 40 50 60 70 80 10 20 30 40 50 60 70 80
Polynomial order: N Polynomial order: N
Position Position (km)

Velocity Velocity (km/s)
a Nondimensional b Dimensional
Figure 2: Closure error versus polynomial order.
where the closure errors of both position and velocity decrease exponentially as the order
increases. Thirteen significant digits can be achieved with N ≥ 60, but this is higher
precision than required for Earth-Moon halo orbits. In the simulations we present in the
next section, we use polynomials of order 30, which still leads to a position accuracy about
0.1 km and velocity accuracy about 1 mm/s as proved by Figure 2b. As we will discuss
in the next section, this obtained accuracy is one magnitude smaller than the range of typical
measurement noise. Of course this order can be tuned according to the accuracy requirement,
and we simply illustrate the methodology. The reference orbit in three dimension is as well
as the projection in the three orthogonal reference planes is shown in Figure 3.
3. Simulation Studies
We use MCPI method for station-keeping of the halo orbit generated in the last section.
The matrix-vector form of MCPI methods for solving this special type of two-point BVPs,
where the positions at the two boundaries are constrained, is shown in Figure 4. The matrices
shown in the formulas, Cx and CαB , are constant once the order of approximation polynomials
are specified, and Θxif is an N 1 × 1 vector depending on the order and the two-point
boundary conditions. Detailed derivations of the approach can be found in the paper by Bai
and Junkins 14. Procedures to use the formula and definitions of the matrices and vectors
are briefly discussed in the appendix of the present paper for completeness.
For the current problem, at the time when a maneuver is possible, the current position
Xti xi , yi , zi and velocity Vti vx , vy , vz are measured. The reference position at the
target time Xtf xf , yf , zf is computed from the coefficients of the reference orbit through
a matrix multiplication and the time can be arbitrary. Then MCPI method is used to solve this
two-point BVP using the formula in Figure 4, and the desired velocity Vr ti vxr , vyr , vzr
at the maneuver time is obtained. The corrective maneuver is computed from

ΔV ti Δvx ti , Δvy ti , Δvz ti Vr ti − Vti . 3.1
×104 ×104
2 4
1
3
z (km)
0
2
−1
1
−2
y (km)
−3 0
−1
4
2 −2
y
(k 0 −3
m −2
) 0 −4
−4 −10000
−20000 −6 −4 −2 0 2 4
×104 m) ×104
x (k x (km)
a 3D b x-y plane
×104 ×10 4
2
2
1 1
0 0
z (km)
z (km)
−1 −1
−2
−2
−3
−3
−4
−4 −3 −2 −1 0 1 2 −4 −3 −2 −1 0 1 2 3 4
×104 y (km) ×104
x (km)
c x-z plane d y-z plane
Figure 3: Reference Orbit.
As a numerical approach to solve BVPs, MCPI method can be easily adjusted if the dynamical
model used for the real mission is more complicated than the model used in the preliminary
design. MCPI method does not require neither gradient information nor the computation of
state transition matrices, and a prior knowledge can be wisely used to provide a good initial
guess to start the interaction, all of which make the method computational very attractive.
The position and velocity of the satellite when it is inserted to the halo orbit will be
away from the ideal initial conditions of the reference orbit. The measurement of the position
and velocity from Earth will not be perfect. And the implementation of the impulses will
also have “execution errors”. We include these three kinds of errors in the simulation. All the
errors are assumed to be Gaussian distribution with a mean of zero. The standard deviations
of these errors are chosen to be consistent with those used by Gómez et al. 9 and are listed
as follows
i tracking errors: σx σy σz 1 km; σvx σvy σvz 1 cm/s,
ii insertion errors: σx σy σz 1 km; σvx σvy σvz 1 cm/s,
iii maneuver errors: σΔvx 0.05; σΔvy 0.05; σΔvz 0.02.

Constant matrix initialization
Starting guess xi (τ)

→
− T
X old xi τ0 , xi τ1 , xi τ2 , . . ., xi τN
Force evaluation
→
−g g τ0 , xi τ0 , g τ1 , xi τ1 , g τ2 , xi τ2 , . . ., g τN , xi τN
T
State update
→
− →
−
X old = X new
eold = enew → −
X new ω2 2 Cx CαB f Cx xif
→
− T
X new xi 1
τ0 , xi 1
τ1 , xi 1
τ2 , . . ., xi 1
τN
Correction calculation
→
− →−
enew X new − X old max
Stopping criterion check
No enew < δ? Yes

and Exit
eold < δ?
Figure 4: Vector-matrix form of MCPI method for solving second order bvps.
Table 3: Baseline case results.

Variable Averaged Results
Max ΔV 27.894 cm/s
Min ΔV 1.178 cm/s
Total ΔV 1606.5 cm/s
Mean deviation distance from the reference 3.98 km
Max deviation distance from the reference 17.59 km
Maneuver numbers 182
Iterations of MCPI 3.05
We look at station-keeping of the halo orbit for 26 revolutions, which corresponds to 377 days,
a little more than one year. 100 Monte Carlo simulations are run and the results in Table 3 are
the averaged numbers from the 100 trials. Here we have used 1/ 7Tp as the spacing between
two impulsive maneuvers which corresponds to a physical time about 2 days. Later, we will
study how this parameter affects the results. Significantly, it might be surprising to notice
that MCPI methods only took about three iterations to converge to such accurate solutions.
This is because we have used the coefficients of the reference orbit to find the corresponding
state values and then use these values to start the MCPI iteration. Thus MCPI method starts
from an ideal trajectory then takes only three Picard iterations to converge to the neighboring
solutions that restore the halo orbit to within the tolerance adopted.
Considering most of the thrusters used to generate impulses will have some limit on
the minimum Δv it can provide; we added one parameter Δvmin 2 cm/s as a threshold
for the control: if the solved impulses are less than the threshold, no correction maneuver
will be made. Table 4 shows the results for this setting. As expected, the deviations from the
reference become larger and the maximum Δv become larger. The good thing is that both
Table 4: Δvmin 2 cm/s.
Variable Averaged results

Max ΔV 29.692 cm/s
Min ΔV 0 cm/s
Total ΔV 1523.5 cm/s: about 1476.5 cm/s per year
the overall Δv and maneuver numbers are reduced, which in part is because that the small
maneuvers that might be in the same magnitude of the measurement noise are avoided.
We also increased this threshold to Δvmin 3 cm/s. Unfortunately and not surprisingly,
the deviations from the reference become larger and more Δv is required. We notice when
we set Δvmin 2 cm/s, we have a similar parameter setting as those studied by Gómez
et al. 9, for the case where they set the tracking and minimum maneuver intervals as two
days. Compared with their results, our total impulses 1476.5 cm/s are larger than theirs
1111.8 cm/s, and we have more impulsive maneuvers 168 than theirs 80; however, our
maximum tracking deviations 17.38 km are much smaller than their results 60.9 km and
our maximum impulse 29.692 cm/s is also much less than theirs 73.554 cm/s. The reason
we have more fuel required than theirs can result from several reasons. First, our reference
halo orbit is different from their orbit. Second, they have used two targeting points and have
minimized a cost function which is a weighted summation of both position deviations and
fuel cost. As they commented in the paper, the weights have been tuned, which very likely
emphasized a large penalty on the fuel cost. Instead, we have not optimized anything in our
approach. This situation is similar as the one reported by Renault and Scheeres 11, where
the LQR control is more fuel efficient than targeting the stable manifold of the equilibrium
point. Additionally, as shown in Figure 4, using the MCPI method only involves evaluating
the differential equations on the discrete nodes and doing vector-matrix multiplication and
summation. Of special interest for the current case, only three iterations are required for
the method to converge. However, the approach by Gómez et al. 9 requires weight tuning
beforehand and also a complicated state transition matrix derivation and then computation
during the mission when a more realistic dynamical model is used. Thus, we believe our
method is computationally very attractive, due to the absence of required tuning, and
extremely simple and easy to implement. Figure 5a shows the history of the magnitude
of the impulses from one Monte Carlo simulation. Due to measurement noise and simulated
control errors, the computed solution of the two-point BVP will of course contain errors that
propagate into an error at the frequent maneuver time. These errors along 26 revolutions
377 days are shown in Figure 5a. Note the numbers shown in Tables from 3 to 5 are
the averaged numbers from 100 Monte Carlo simulations. For the 168 maneuvers along
the same orbit of Figures 5a and 5b, Figure 5c shows the number of Picard iterations
required to achieve convergence. We conclude, 92.2% of the cases converge in three iterations
and 100% converged in four or fewer iterations, a remarkable algorithm! Another important
parameter is the time interval to solve for and then execute the maneuver. Usually there exists
a minimum maneuver time because of the necessity for orbit determination process and the
mission operation capability. On the other hand, if the maneuver intervals become too large,
the deviations from the reference can become too large to be easily corrected. We have studied
Distance from the reference

V from one Monte Carlo simulation from one Monte Carlo simulation
11
−3
10
10
9
8
V (km/s)
R (km)
7
6
5
4
3
2
1
20 40 60 80 100 120 140 160 180 20 40 60 80 100 120 140 160 180
Impulse time (1/7Tp ) Impulse time (1/7Tp )
Deviation
Mean deviation
a Impulse history b Deviation
Number of iterations
from one Monte Carlo simulation
4
3.9
3.8
3.7
Iterations
3.6
3.5
3.4
3.3
3.2
3.1
3
20 40 60 80 100 120 140 160 180
Impulse time (1/7Tp )
Iterations
Mean iterations
c Iteration numbers
Figure 5: Results from One Monte Carlo Simulation.
the performance while changing the interval times from a half day to a little more than three
and a half days. The results below show how the position and velocity deviations, number
of maneuvers, total ΔV , and maximum ΔV change with respect to the interval time, see
Figures 6a to 6d. As expected, Figures 6a and 6b show that the deviations increase
and the number of maneuvers decrease as time interval between corrections becomes larger.
It is interesting to see that there exists a broad window of one and half to three days for
time intervals between corrections, that gives near-minimum fuel cost, as evident in Figure
6c. This is similar as a situation reported by Renault and Scheeres 11, where the optimal
spacing for both LQR approach and targeting equisetum method exists for the problems they
studied. From Figure 6d, the fact that the maximum impulse is minimum for 1 to 1.5 days
implies that separation is significant.
Table 5: Δvmin 3 cm/s.
Variable Averaged results

Max ΔV 33.86 cm/s
Min ΔV 0 cm/s
Total ΔV 1574.1 cm/s
4. Approaches to Reduce the Fuel Cost

In this section, we study how to reduce the fuel cost using the presented method
while maintaining its advantages such as no requirement for any gradient information,
computational efficiency, and the final results in an orthogonal polynomial form.
The above-method results in exact satisfaction of terminal boundary conditions on
each subinterval if there are no control errors and measurement errors. As a result, the posi-
tion errors are much smaller than those in the literature 9, whereas the control required is
about 24% higher. To gain some qualitative insight to the tradeoff between allowing increased
terminal errors and the associated cost, the current method can be used in a heuristic fashion.
We reasoned that applying less than 100% of the control impulse computed would result in
increased terminal errors and smaller control, and in fact, we found that using 80% of the
computed impulse in the current problem results in a control cost only 4% higher than those
in reference 9, while our terminal errors in fact still remain significantly smaller than those
in the references. We emphasize that the relationships between the amount of the control to
apply and the performance such as the total fuel cost, the maximum position deviation,
and the mean position deviation are not linear. In Figure 7, we show how the total fuel
cost, the maximum deviation, and the mean deviation change with respect to the introduced
coefficient. The simulation uses the same conditions we studied for the results reported in
Table 4. Notice all these three performance variables have been averaged from the 100 Monte
Carlo simulations. Especially, we found, by applying 80% of the fuel required for solving the
exact two-point boundary value problem, the total fuel cost is 1160 cm/s, the mean position
deviation from the reference orbit is 4.84 km, and the maximum position deviation is 22.2 km.
Thus with a fuel cost only 4% larger than the reported results 1111.8 cm/s 9, our position
deviations from the reference orbit are much smaller than the reported results, for which the
mean and maximum position deviations are 61 km and 74 km, respectively.
Another approach to reduce the fuel cost is to adopt a hybrid propulsion system
which uses low thrust during the flight and impulses at some specified points. The hybrid
propulsion concept has been studied by Bai et al. for an Earth to Apophis mission design
19. Since the presented methods can also solve optimal control problems for continuous
system 14, utilizing this hybrid propulsion system is straightforward. We could also reduce
the fuel cost by varying the time interval used for the station-keeping instead of using the
current constant time interval. However, this will increase the computation requirements,
since all the matrices used in the current approach are constant and are precomputed, which
have to be recomputed if the time interval changes. We will also explore to efficiently utilize
the dynamic information of some specified orbits to reduce the fuel cost in the future study.
Averaged number of maneuver versus spacing

Averaged deviation versus spacing
800
180
160 700
Number of maneuver
140
Deviation (km)
600
120
500
100
80 400
60 300
40
200
20
100
0.5 1 1.5 2 2.5 3 3.5 0.5 1 1.5 2 2.5 3 3.5 4
Maneuver spacing (days) Maneuver spacing (days)
Mean deviation
Max deviation
a Deviations. b Number of maneuvers
Averaged total V versus spacing Averaged max V versus spacing

11000 180
10000 160
9000
140
8000
V (cm/s)
V (cm/s)
7000 120
6000 100
5000 80
4000
60
3000
2000 40
1000 20
0.5 1 1.5 2 2.5 3 3.5 4 0.5 1 1.5 2 2.5 3 3.5 4
Maneuver spacing (days) Maneuver spacing (days)
c Total impulses d Max impulses
Figure 6: Effects of the spacing of maneuvers averaged from 100 Monte Carlo simulation.
5. Conclusions
Modified Chebyshev-Picard Iteration method is used for station-keeping of a halo orbit
around the L2 libration point of the Earth-Moon system. Compared with other station-
keeping techniques which also generate impulses at discrete times, the presented method
is much simpler, since it does not need to compute state transition matrix or gradient
information, which can be complicated for the full dynamical model and might be impossible
to implement for the on-board system. Previous studies 13, 14 have shown that the MCPI
algorithm works well for highly perturbed orbits; it is anticipated that little change will be
required in the present algorithm upon including solar and other non-CR3BP effects. There is
also no tuning process e.g., for the weight matrices which is required for other approaches.
Although the fuel cost from this approach, for the examples presented herein, might be a
little higher than other optimal control results, the simple structure of the method as well as
its salient computation efficiency makes this technique very attractive. The method can be
used for station-keeping other orbits with the only change being the use of the appropriate
Averaged total V versus coe cient of control Averaged max deviation versus coe cient of control
1650 60
1600 55
1550 50
1500
Deviation (km)
45
V (cm/s)
1450
40
1400
35
1350
1300 30
1250 25
1200 20
1150 15
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Coefficient of control Coefficient of control
a Averaged total impulses b Averaged maximum deviation
Averaged mean deviation versus coe cient of control

8
7.5
7
Deviation (km)
6.5
6
5.5
5
4.5
4
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Coefficient of control
c Averaged mean deviation
Figure 7: Performance versus coefficient of control.
equations of motion. We believe this method can take the same role as the classical Battin’s
form of the Lambert algorithm for solving orbit transfer problems, with the differences that
this method is a numerical approach and can be used for rather general nonlinear systems,
including perturbations. We are investigating extending the current approach which solves
two-point boundary value problems to solve optimal control problems which include several
target points as well as hybrid propulsion mode with both impulsive and continuous thrust.
Appendix
MCPI Method for Solving Second Order BVPS
Assume the second order dynamic system is described as
d2 xt > ?
ft, xt, t ∈ t0 , tf , A.1
dt2
and the boundary conditions are xt t0 x0 and xt tf xf . The first step of MCPI
methods is to transform the generic independent variable t to a new variable τ, which is
defined on the valid range, the closed interval −1, 1, of Chebyshev polynomials
t f t0 tf − t 0
t ω1 ω2 τ, ω1 , ω2 . A.2
2 2
Introducing this time transformation of A.2 into A.1, it is rewritten as
d2 xτ
ω22 fω1 ω2 τ, xτ ≡ gτ, xτ. A.3
dτ 2
The position update equation is
k 1 ω2 Cx CαB f Cx Θxif ,
X 2
A.4
where X xτ0 , xτ1 , xτ2 , . . . , xτN T is the vector representing the position trajectory
evaluated at all the N 1 Chebyshev-Gauss-Lobatto CGL nodes, which are computed
through

jπ
τj cos , j 0, 1, 2, . . . , N. A.5
N
The components of Cx ≡ T W and CαB SB T V are defined as below, and f

fτ0 , fτ1 , fτ2 , . . . , fτN T . Θxif is defined by the boundary condition as Θxif x0
xf , xf − x0 /2, 0, 0, . . . , 0T ∈ RN 1 . Notice although the second order formula in A.4
only solves for the position, the velocity solution can be computed conveniently by taking
the derivative of the position, which has been obtained in a form of Chebyshev polynomi-
als:
⎡ ⎤
T0 τ0 T1 τ0 · · · TN τ0
⎢ ⎥
⎢ T0 τ1 T1 τ1 · · · TN τ1 ⎥
T ⎢
⎢ .. .. .. ..
⎥,
⎥ A.6
⎣ . . . . ⎦
T0 τN T1 τN · · · TN τN
and the two diagonal matrices W and V are defined as

1
W diag , 1, 1, . . . , 1, 1 ,
2
A.7
1 2 2 2 1
V diag , , ,..., , .
N N N N N
The definition of SB as shown below is more complicated than others
SB k 1, k − 1 1/4/kk − 1, k 2, 3, −2, . . . , N − 2,

SB k 1, k 1 −1/2/ k2 − 1 , k 2, 3, −2, . . . , N − 2,
SB k 1, k 3 1/4/k/k 1, k 2, 3, −2, . . . , N − 2,

SB N, N −1/2/ N − 12 − 1 ,
SB N, N − 2 1/4/N − 1/N − 2,

SB N 1, N 1 −1/2/ N 2 − 1 ,
SB N 1, N − 1 1/4/N/N − 1,

SB 1, k 1 −3 1 −1k / k2 − 4 / k2 − 1 , k 4, 5, . . . , N − 2,

SB 2, k 1 −3/2 1 − −1k / k2 − 4 / k2 − 1 , k 4, 5, . . . , N − 2,
1
SB 1, 1 − ,
4 A.8
SB 1, 2 0,
7
SB 1, 3 ,
24
SB 1, 4 0,
SB 2, 1 0,
1
SB 2, 2 − ,
24
SB 2, 3 −0,
1
SB 2, 4 ,
20
SB 1, k 1 1 −1k /4k − 5/ k2 − 1 /k − 2, k N − 1,

SB 2, k 1 1 − −1k /8k − 5/ k2 − 1 /k − 2, k N − 1,

SB 1, k 1 1 −1k /4k − 5/ k2 − 1 /k − 2, k N,

SB 2, k 1 1 − −1k /8k − 5/ k2 − 1 /k − 2, k N,
where Si, j represents the element in the ith row and jth column.
Acknowledgment
This work is supported by the Air Force Office of Scientific Research Contract number
FA9550-11-1-0279.
References
1 R. W. Farquhar, “Station-keeping in the vicinity of collinear libration points with an application to a
lunar communications problem,” in Proceedings of the Space Flight Mechanics Specialist Symposium, AAS
Science and Technology Series, pp. 519–535, Denver, Colo, USA, July 1966.
2 R. W. Farquhar and A. A. Kamel, “Quasi-periodic orbits about the translunar libration point,” Celestial
3 R. W. Farquhar, D. P. Muhonen, C. Newman, and H. Heuberger, “Trajectories and orbital maneuvers
for the first libration-point satellite,” Journal of Guidance, Control and Dynamicss, vol. 3, no. 6, pp. 549–
554, 1980.
4 J. V. Breakwell, A. A. Kamel, and M. J. Ratner, “Station-keeping for a translunar communication sta-
tion,” Celestial Mechanics, vol. 10, pp. 357–373, 1974.
5 M. Xin, M. W. Dancer, S. Balakrishnan, and H. J. Pernicka, “Stationkeeping of an L2 Libration point
satellite with θ − D technique,” in Proceedings of the American Control Conference, Boston, Mass, USA,
2004.
6 J. E. Kulkarni, M. E. Campbell, and G. E. Dullerud, “Stabilization of spacecraft flight in halo orbits: an
H∞ approach,” IEEE Transactions on Control Systems Technology, vol. 14, no. 3, pp. 572–578, 2006.
7 D. Cielaszyk and B. Wie, “New approach to halo orbit determination and control,” Journal of Guidance,
8 D. W. Dunham and C. E. Roberts, “Stationkeeping techniques for libration-point satellites,” Journal of
the Astronautical Sciences, vol. 49, no. 1, pp. 127–144, 2001.
9 G. Gomez, K. Howell, K. Howell, C. Simo, and J. Masdemont, “Station-keeping strategies for trans-
lunar libration point orbits,” in Proceedings of the AAS/AIAA Space Mechanics Meeting, Monterey, Calif,
USA, 1998.
10 K. C. Howell and H. J. Pernicka, “Stationkeeping method for libration point trajectories,” Journal of
11 C. A. Renault and D. J. Scheeres, “Statistical analysis of control maneuvers in unstable orbital environ-
ments,” Journal of Guidance, Control, and Dynamics, vol. 26, no. 5, pp. 758–769, 2003.
12 E. D. Gustafson and D. J. Scheeres, “Optimal timing of control law updates for unstable systems with
continuous control,” in Proceedings of the American Control Conference, Seattle, Wash, USA, June 2008.
13 X. Bai and J. L. Junkins, “Modified Chebyshev-Picard iteration methods for solution of initial value
problems,” Advances in the Astronautical Sciences, vol. 139, pp. 345–362, 2011.
14 X. Bai and J. L. Junkins, “Modified Chebyshev-Picard iteration methods for solution of boundary
value problems,” Advances in the Astronautical Sciences, vol. 140, pp. 381–400, 2011.
15 X. Bai, Modified Chebyshev-Picard iteration methods for solution of initial value and boundary value problems,
Ph.D. dissertation, Texas A&M University, College Station, Tex, USA, 2010.
16 D. L. Richardson, “Analytic construction of periodic orbits about the collinear points,” Celestial Mech-
anics, vol. 22, no. 3, pp. 241–253, 1980.
17 H. Schaub and J. L. Junkins, Analytical Mechanics of Space Systems, American Institute of Aeronautics
and Astronautics, Reston, Va, USA, 1st edition, 2003.
18 J. V. Breakwell and J. V. Brown, “The “Halo” family of 3-dimensional periodic orbits in the Earth-
Moon restricted 3-body problem,” Celestial Mechanics, vol. 20, no. 4, pp. 389–404, 1979.
19 X. Bai, J. D. Turner, and J. L. Junkins, “Optimal thrust design of a mission to apophis based on a homo-
topy method,” in Proceedings of the AAS/AIAA Spaceflight Mechanics Meeting, Savannah, Ga, USA, 2009.
doi:10.1155/2012/734280
Research Article
Assessment of the Ionospheric and Tropospheric
Effects in Location Errors of Data Collection
Platforms in Equatorial Region during High
and Low Solar Activity Periods
Áurea Aparecida da Silva,1 Wilson Yamaguti,1

Hélio Koiti Kuga,2 and Cláudia Celeste Celestino3
1
Space Systems Division (DSE), National Institute for Space Research (INPE),
2
Space Mechanics and Control Division (DMC), National Institute for Space Research (INPE),
3
Centro de Engenharia, Modelagem e Ciências Sociais Aplicadas (CECS),
Universidade Federal do ABC (UFABC), 09210-170 Santo André, SP, Brazil
Correspondence should be addressed to Áurea Aparecida da Silva, aurea.silva@dss.inpe.br
Received 15 August 2011; Accepted 3 November 2011
Copyright q 2012 Áurea Aparecida da Silva et al. This is an open access article distributed under
The geographical locations of data collection platforms DCP in the Brazilian Environmental Data
Collection System are obtained by processing Doppler shift measurements between satellites and
DCP. When the signals travel from a DCP to a satellite crossing the terrestrial atmosphere, they
are affected by the atmosphere layers, which generate a delay in the signal propagation, and
cause errors in its final location coordinates computation. The signal propagation delay due to
the atmospheric effects consists, essentially, of the ionospheric and tropospheric effects. This work
provides an assessment of ionospheric effects using IRI and IONEX models and tropospheric delay
compensation using climatic data provided by National Climatic Data Center. Two selected DCPs
were used in this study in conjunction with SCD-2 satellite during high and low solar activity
periods. Results show that the ionospheric effects on transmission delays are significant about
hundreds of meters in equatorial region and should be considered to reduce DCP location errors,
mainly in high solar activity periods, while in those due to tropospheric effects the zenith errors
are about threemeters. Therefore it is shown that the platform location errors can be reduced when
the ionospheric and tropospheric effects are properly considered.
1. Introduction
The Brazilian Environmental Data Collection System is currently composed by SCD-1 and
SCD-2 satellites, hundreds of data collection platforms DCPs, receiving ground stations
located at Cuiabá and Alcântara, and Data Collection Mission Center located at Natal,
Brazil.
In this system, a satellite works as a message relay transponder. A DCP to receiving
ground station communication link is established when a simultaneous DCP to satellite
uplink and satellite to receiving station downlink occurs simultaneously. UHF frequency
bands are used for uplink and S band for the downlink. The uplink Doppler shifts are
measured and time-stamped at a receiving ground station while the downlink Doppler shift
is compensated for by the receiving equipments. The DCPs messages transmitted by satellites
and received by Cuiabá or Alcântara ground station are sent to the Data Collection Mission
Center, now located at INPE Brazilian National Institute for Space Research Northeast
Regional Center in Natal city, Brazil, where the data are processed, stored, and distributed
to users. The users receive data at most 30 minutes after the satellite pass.
The DCP geographical location can be determined using the uplink around
401,635 MHz Doppler measurements, the statistical least squares method, and knowledge
of the satellite orbit. As the S band 2.26752 GHz downlink is compensated by the ground
station receiver, it is not considered in this study.
The signal transmitted from a DCP to a satellite is affected during its travelling
across the atmosphere. Among other factors the signal path is influenced by the chemical
elements that make up the extensive Earth’s atmosphere resulting in propagation delays. As
a consequence, errors are present in the final coordinates provided to system users.
As such the ionosphere and the troposphere in vertical direction can be an important
source of errors in the platforms location. The errors due to ionosphere can vary from tens to
hundreds of meters depending on frequency, while in the troposphere the zenith errors are
generally between two and three meters 1–3.
In Celestino et al. 3 the simulated and real effects of ionosphere and troposphere
in DCPs geographical location in the Brazilian Environmental Data Collection System were
evaluated. In this work, in order to evaluate the effects associated with the ionosphere, the
values of total electron content TEC were obtained from the standard IRI International
Reference Ionosphere model. The troposphere values, dry and wet components, were
obtained from data of CPTEC-INPE Weather Forecast and Climate Studies Center. The
study did not consider the effects of solar activity periods, but only the time of a day. Results
of the analysis indicated that correction of the ionosphere and troposphere effects can, on the
average, reduce the location errors to the scale between 10 to 250 meters.
In the absence of the selective availability signal corruption which was turned off
on May 1st, 2000 the largest source of error in GPS positioning and navigation, using
L1 frequency receiver, has been the ionosphere. The ionospheric effects were investigated
by Camargo et al. 4 in the determination of point positioning and relative positioning
using single frequency data. The model expressed by a Fourier series and the parameters
were estimated from data collected at the active stations of RBMC Brazilian Network for
Continuous Monitoring of GPS satellites. Experiments were carried out in the equatorial
region, using data collected from dual frequency receivers. In order to validate the model, the
estimated values were compared with “ground truth”. Results show, for point positioning
and relative positioning, a reduction of error better than 80% and 50%, respectively, compared
to the processing without the ionosphere model. These results give an indication that more
research must be done to provide support to L1 GPS users in the equatorial region.
Hence, the present work is divided into two phases: the first is an analysis of the
DCP location errors generated by the delay of the signal propagated in the ionosphere.
Therefore, the ionosphere is considered in two different periods according to their intensity
periods of high and low solar activity. The ionosphere representation through TEC values
Total Electron Content was obtained from IRI model—International Reference Ionosphere
5—and the one derived from GPS measurements in IONEX version—IONosphere map
EXchange 6. These analyses show the relevance of the errors due to the ionospheric delay in
different periods and with use of different models. Doppler shift measurements are simulated
for ideal cases in order to measure the influence of the ionosphere at the location. A Doppler
shift measurement corresponds to the difference between a received signal frequency at the
satellite and the nominal platform transmitter frequency. This shift is caused by the relative
velocity between the satellite and platform. In this work, the DCP no. 32590 15.55293◦ S and
56.06875◦ W was selected for geographical location studies due to its location in Cuiabá,
Brazil, in the equatorial region, and data availability. The period of data covers the high 2001
and the low solar activity 2009.
The second phase of the work consisted of the analysis of the best model to express
the ionosphere in the equatorial region and the use of the selected model for DCP location
with Doppler measurements. This work shows that the IONEX file models the ionosphere
with a better accuracy, especially in periods of high solar activity, increasing sufficiently the
density of the atmosphere in low latitudes, that is, in the equatorial anomaly region. This
motivated the selection of the IONEX model for the second phase of the work. In addition to
the correction of the ionosphere, the effects of the troposphere were considered by the use of
climate data provided from National Climatic Data Center—NOAA Satellite and Information
Service.
Real Doppler shift measurements are also used for DCP no. 109 location located
in French Guyana, at 5.1860◦ S and 52.687◦ W and SCD-2 satellite, taking into account the
following conditions: time of the satellite pass from 15:00 to 22:00 UT Universal Time,
periods of high solar activity in 2001, and of low solar activity in 2008.
2. Characteristics of the Ionosphere and Troposphere

The ionosphere is roughly located between 50 to 1000 kilometers above the terrestrial surface,
being the electromagnetic radiation the largest agent of the ionization process. Furthermore,
the Sun inserts great amount of free electrons that contributes for the composition of
ionosphere. The solar wind and the solar electromagnetic radiation can have drastic change
during a solar storm, which implies changes in the conditions of the magnetic field and the
terrestrial ionosphere.
The location error is associated with the ionosphere which is directly proportional to
total electron content TEC contained along the signal trajectory in the ionosphere, and it
is inversely proportional to the square of signal frequency. TEC and, consequently, the error
due the ionosphere vary in time and space, and they are affected by several variables, such
as: solar cycle, local time of the day, season, geographical location, geomagnetic activity, and
others 4. Thus, it is very important to know the behavior of the ionosphere, taking into
account their dependence on the location and time.
Besides the ionosphere temporal variation over the years, it presents variation of the
electron density in cycles of about 11 years. During these peaks, there is a higher incidence
of sun radiation solar activity with increase of ionization in terrestrial atmosphere layers.
TEC values are proportional to the increase of solar activity, that is, the increase of sunspots.
In periods of maximum activity, TEC can reach values about twice as larger than in periods
350
300
250
200
150
100
50
0
2000 2002 2004 2006 2008 2010 2012
Time (years)
Sunspot number RI
Daily SM predictions
Monthly CM predictions
Smoothed
Figure 1: Sunspot numbers of cycle 23 and beginning of cycle 24. The daily yellow, monthly
blue, and monthly smoothed red sunspot numbers since 2000, together with predictions for
12 months ahead: SM red dots: classical prediction method, based on an interpolation of Wald-
meier’s standard curves; CM red dashes: combined method proposed by K. Denkmayr. Source:
http://sidc.oma.be/html/wolfjmms.html—Access in October 2011.
of low solar activity. Currently, the Sun is starting the cycle 24 that should have its maximum
period between 2012 and 2014. The previous cycle, 23, had its maximum between 2000 and
2002, when there was a considerable increase of the sunspot numbers and, consequently, the
number of electrons in the ionospheric layer.
Figure 1 shows the sunspot numbers of solar cycle 23 and the beginning of cycle 24,
in a period from 2000 to the first semester of 2011 and a prediction of the new cycle for 12
months ahead.
Figure 2 shows three different geographical regions of the ionosphere on terrestrial
globe known as high-latitude region, medium-latitude region, and equatorial region. Figure 3
shows that in the equatorial region, where a great part of South America and Africa region
are located, the equatorial anomaly occurs, increasing the TEC values, mainly in the most
active period of Sun.
TEC is given along the direction between the transmitting DCP and the satellite. TEC
quantities in vertical direction VTEC are given by
TEC VTEC sec Z, 2.1
where sec Z is the mapping function, and Z is the signal path zenithal angle in relation to the
plan of a mean reference altitude, estimated by means of so-called Bent Ionospheric Model.
For IRI model an altitude of 350 km was used, whereas for IONEX, 450 km was used. The
180◦W 150◦W 120◦W 90◦W 60◦W 30◦W 0◦ 30◦E 60◦E 90◦E 120◦E 150◦E 180◦E
90◦ N 90◦N
High latitude
60◦ N 60◦N
Medium latitude
30◦ N 30◦N
Equatorial
0◦ region 0◦
30◦ S 30◦S
Medium latitude
60◦ S 60◦S
High latitude
90◦ S 90◦S
180◦W 150◦W 120◦W 90◦W 60◦W 30◦W 0◦ 30◦E 60◦E 90◦E 120◦E 150◦E 180◦E
Figure 2: Geographical regions of the ionosphere.
ionosphere signal delay in DCP/satellite direction is calculated, according to Aksnes et al.

7 by
40.3 VTEC sec Z

RI , 2.2
f2
where f is the DCP frequency transmission. Considering VTEC constant for short times, the
temporal variation of the signal delay due to the ionosphere is given by
k1 VTEC cos γ sin γ

ṘI − 3/2 γ̇, 2.3
f 2 1 − k2 cos2 γ
where VTEC is the total electron content in vertical direction, γ is the satellite elevation
angle with respect to DCP, and γ̇ is the satellite elevation angle rate. Depending on the
mean reference altitude h, one has for h 350 km, then k1 36.21 and k2 0.8985 and
for h 450 km, then k1 35.16 and k2 0.8723.
The last equation can be applied to the ionosphere correction based on Doppler shift
measurements. The deviation due to the ionosphere is sensitive to the values of VTEC, which
can be obtained from IGS International GNSS Service with free access to any users in IONEX
format 6. Such data correspond to vertical TECU that means vertical TEC unitary with the
following relationship: a unit of TEC 1 TECU corresponds to 1016 electrons per square meter
8. Figures 3a, 3b, and 3c show a set of vertical TECU global maps on October 22, 2001.
The figures are presented in different times, 9:00, 13:00, and 17:00 UT and show the equatorial
anomaly evolution along low-latitude regions from 30◦ S to 30◦ N.
Tropospheric effect depends on atmosphere density and the satellite elevation
angle. This effect can be observed from the terrestrial surface up to about 50 km 9.
The representation of the troposphere deviation depends on the atmospheric pressure,
◦ ◦ ◦ ◦
−180 −120 −60 0 60◦ 120◦ 180◦
88◦ 88◦
58◦ 58◦
Geographic latitude
29◦ 29◦
0◦ 0◦
−29◦ −29◦
−58◦ −58◦
(a) Vertical TECU for 22/10/2001 0900 UT

−88◦ −88◦
◦ ◦ ◦ ◦ ◦ ◦ ◦
−180 −120 −60 0 60 120 180
Geographic longitude
a
◦ ◦ ◦
−180 −120 −60 0◦ 60◦ 120◦ 180◦
88◦ 88◦
58◦ 58◦
Geographic latitude
29◦ 29◦
0◦ 0◦
−29◦ −29◦
−58◦ −58◦
(b) Vertical TECU for 22/10/2001 1300 UT

−88◦ −88◦
◦ ◦ ◦ ◦ ◦ ◦ ◦
−180 −120 −60 0 60 120 180
b
−180◦ −120◦ −60◦ 0◦ 60◦ 120◦ 180◦
88◦ 88◦
58◦ 58◦
Geographic latitude
29◦ 29◦
0◦ 0◦
−29◦ −29◦
−58◦ −58◦
(c) Vertical TECU for 22/10/2001 1700 UT

−88◦ −88◦
◦ ◦ ◦ ◦ ◦ ◦ ◦
−180 −120 −60 0 60 120 180
0 38 75 113 150
c
Figure 3: Vertical TECU from IONEX model for October 22, 2001. Source: Produced by Leica
Geosystems/GNSS QC software.
atmospheric temperature, and water vapor pressure. According to the IERS International
Earth Rotation Service model the tropospheric signal delay is given by 10
A B
RT . 2.4
sin γ B/A B/ sin γ 0.01
Thus, the temporary variation of the signal delay due to the troposphere is given by
+ 2 ,
cos γ B − A B · sin γ 0.01
ṘT A B2 > ?2 γ̇, 2.5
sin γ sin γ 0.01 · A B B
where A 0.002357Po 0.000141eo , B 1.084 × 10−8 Po To K 4.734 × 10−8 Po2 /To 2/3 −
1/K, K 1.163 − 0.00968 cos 2φ − 0.00104 To 0.00001435 Po , γ satellite elevation
angle, γ̇ satellite elevation angle rate, Po atmospheric pressure in 10−1 kPa, equivalent to
millibars, To atmospheric temperature in Kelvin, eo water vapor pressure in 10−1 kPa,
equivalent to millibars, and φ DCP geodetic latitude.
Values of Po , To , and eo are monthly averages obtained for different geographical
latitudes and longitudes from National Climatic Data Center—NOAA Satellite and
Information Service 11.
3. Results
The geographical location is computed by the in-house developed GEOLOC geographical
location program in FORTRAN 12, which processes the uplink measured Doppler shift
suffered by the signal transmitted from the DCPs, together with a statistical least squares
method and satellite orbit ephemeris 2, 3.
Given the complex dependence of the ionosphere with the solar cycle, local time of the
day, season, geographical location, geomagnetic activity, and others, it is important to make
an analysis of different models that describe VTECU values for the geographical location,
date, and time for the DCP to be located. First we introduce the ionosphere effect using either
IRI or IONEX model and through simulation one verifies its impact on location computation.
Second we use real data to check the consistency of the findings in the former simulation.
3.1. Ideal Doppler Data for DCP no. 32590 with Ionosphere Influence
Two different models of the ionosphere were analyzed: IRI International Reference
Ionosphere 5 and IONEX IONosphere map EXchange 6.
The chosen periods for the location were in October 2001 and October 2009, periods
that are characterized by high and low solar activity, respectively. The month chosen was
October, because it is within the time that VTECU has its maximum values spring equinox,
mainly in 2001. Furthermore, it was possible to find a set of SCD-2 satellite pass data in a
period of the day that VTECU data are most intense 18-19 UT for that month, in 2001 and
2009. This plays an important role for the location with real data.
In the first set of results VTECU values on October 19–22, 2001 and 2009 are shown
for geographical location of DCP no. 32590 15.55293◦ S and 56.06875◦ W. Doppler shift
Vertical TECU for 19/10/2001 1900 UT

−180◦ −120◦ −60◦ 0◦
88◦ 88◦
58◦ 58◦
Geographic latitude
29◦ 29◦
0◦ 0◦
−29◦ −29◦
−58◦ −58◦
−88◦ −88◦
◦
−180◦ −120◦
−60 ◦
0
a
−180◦ −120◦ −60◦ 0◦
◦
88 88◦
58◦ 58◦
Geographic latitude
29◦ 29◦
0◦ 0◦
−29◦ −29◦
−58◦ −58◦
−88◦ −88◦
−180◦ −120◦ −60◦ 0◦
0 38 75 113 150
b
Figure 4: VTECU maps on October 19 and 22, 2001 at 19 UT showing the equatorial anomaly crests in the
region close to Brazil for maximum values of VTECU geographic equator in solid line and the magnetic
equator in dashed line. Source: Produced by Leica Geosystems/GNSS QC software.
measurements are simulated from this DCP real location, and the ionospheric effect is added
to the model that generates such data. Thus, it is possible to verify in the simulation how
far the influence of ionosphere in the location error is. VTECU values obtained from IRI and
IONEX models are used.
Figure 4 shows a sequence of VTECU maps of the region in latitude between 88◦ South
and 88◦ North and longitudes between −180◦ and 0◦ , where American continent is located.
This figure shows maps for two days October 19 and 22, 2001, a period that is characterized
by high solar activity, and it can be seen the equatorial anomaly effect 13, when VTECU

−180◦ −120◦ −60◦ 0◦
88◦ 88◦
58◦ 58◦
Geographic latitude
29◦ 29◦
0◦ 0◦
−29◦ −29◦
−58◦ −58◦
−88◦ −88◦
◦
−180◦ −120◦
−60 ◦
0
a
−180◦ −120◦ −60◦ 0◦
◦
88 88◦
58◦ 58◦
Geographic latitude
29◦ 29◦
0◦ 0◦
−29◦ −29◦
−58◦ −58◦
−88◦ −88◦
−180◦ −120◦ −60◦ 0◦
0 38 75 113 150
b
Figure 5: VTECU maps on October 19 and 22, 2009 at 18 UT for minimum values of VTECU in low
solar activity period geographic equator in solid line and the magnetic equator in dashed line. Source:
Produced by Leica Geosystems/GNSS QC software.
values are maximum in Brazilian region, around 18-19 UT universal time or 15 h local time.
Through the diagram of colors it is possible to verify that in this period VTECU reaches
values close to 150 × 1016 electrons/m2 in the equatorial anomaly region red. On the other
hand, in Figure 5, which represents the period of October 19 and 22, 2009, there is not a
considerable difference in VTECU values for the same time, 18-19 UT. The diagram of colors
is maintained between the dark and the light blue, which represents a value about 10 to 40 ×
1016 electrons/m2 . This fact is given in low solar activity.
Vertical TECU for 19/10/2001—DCP #32590 Vertical TECU for 22/10/2001—DCP #32590
140 140
120 120
100 100
VTECU
VTECU
80 80
60 60
40 40
20 20
0 0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
IONEX IONEX
IRI IRI
SCD-2 SCD-2
a b
Figure 6: VTECU values on October 19 and 22, 2001 in a period of one day at geographical location of DCP
no. 32590 IONEX blue, IRI pink, and triangles green represent the time of SCD-2 satellite passes.
Vertical TECU for 19/10/2009—DCP #32590 Vertical TECU for 22/10/2009—DCP #32590
140 140
120 120
100 100
VTECU
VTECU
80 80
60 60
40 40
20 20
0 0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
IONEX IONEX
IRI IRI
SCD-2 SCD-2
a b
Figure 7: VTECU values on October 19 and 22, 2009 in a period of one day at geographical location of DCP
no. 32590 IONEX blue, IRI pink, and triangles green represent the time of SCD-2 satellite passes.
Figures 6 and 7 show VTECU values in geographical location where DCP no. 32590
is located, for the same periods mentioned above, 2001 and 2009, using IRI pink line and
IONEX blue line. The triangle symbols green in the figures represent the time of SCD-
2 satellite passes, which span a period between 15 and 22 UT. VTECU is showed with an
interval of two hours and in a full period of a day. In Figure 6 it is verified a discrepancy
between VTECU values of IRI and IONEX, mainly from 14 to 24 UT, and such values differ
by up to 3 times in the comparison between the models; that is, IONEX represents better the
total electron content in high solar activity periods. This difference results in an increase of
location error in up to 50% when the ionospheric delays are added to the simulated Doppler
data. Figure 7 has the same representation as the previous one, however, in 2009 low solar
activity. In this case, it appears that both IRI and IONEX express the total electron content
in a similar way. Thus, the increase in DCP location error is consistent to any chosen model
for obtaining VTECU values. It depicts the importance of the ionospheric correction in high
Location of the DCP no.32590—2001, October ionosphere effect evaluation

1
0.8
Error (km)
0.6
0.4
0.2
0
10/19/01 12:00
10/19/01 18:00
10/20/01 0:00
10/20/01 6:00
10/20/01 12:00
10/20/01 18:00
10/21/01 0:00
10/21/01 6:00
10/21/01 12:00
10/21/01 18:00
10/22/01 0:00
10/22/01 6:00
10/22/01 12:00
10/22/01 18:00
10/23/01 0:00
Time
Ionosphere: IONEX
Ionosphere: IRI
Without correction
Figure 8: Ionospheric effect evaluation in DCP no. 32590 location error on October 19–22, 2001 location
errors without ionospheric effect blue, with the addition of the ionospheric delay using IRI model pink,
and with the addition of the ionospheric delay using IONEX model green.
solar activity periods, because according to Figure 6, VTECU values jump from about 40 to
more than 120 × 1016 electrons/m2 at the end of the day.
With the data of VTECU from different models IONEX e IRI, in high and low solar
activity periods, geographical location error for DCP no. 32590 due to ionosphere delay is
generated.
A first analysis of these results shows that the location error due to ionospheric effect
cannot be neglected because the error can increase to around 50% and the model which must
be used depends mainly on the intensity of solar activity. In fact, in low solar activity, both
models yield VTECU values with similar values. However, in high solar activity, IONEX has
yielded the values in a best way, mainly in the region that is called equatorial anomaly, close
to magnetic equator, which can be easily observed in Figure 4.
Figures 8 and 9 show DCP no. 32590 location errors in 2001 and 2009, respectively,
including the ionospheric effect and making use of VTECU values from both models. These
results are compared with the ideal location, that is, without the inclusion of ionospheric
effect.
3.2. Real Doppler Data for DCP no. 109 with Ionospheric and Tropospheric
Effects Correction
This second phase of the study aimed at applying the ionosphere correction in high and
low solar activity periods making use of IONEX model, which was previously studied, as
well as the tropospheric delay correction with climatic data obtained from National Climatic
Data Center—NOAA Satellite and Information Service 11. Due to the availability of real
Location of the DCP no.32590—2009, October ionosphere effect evaluation

1
0.8
Error (km)
0.6
0.4
0.2
0
10/19/09 12:00
10/19/09 18:00
10/20/09 0:00
10/20/09 6:00
10/20/09 12:00
10/20/09 18:00
10/21/09 0:00
10/21/09 6:00
10/21/09 12:00
10/21/09 18:00
10/22/09 0:00
10/22/09 6:00
10/22/09 12:00
10/22/09 18:00
10/23/09 0:00
Time
Ionosphere: IONEX
Ionosphere: IRI
Without correction
Figure 9: Ionospheric effect evaluation in DCP no. 32590 location error on October 19–22, 2009 location
errors without ionospheric effect blue, with the addition of the ionospheric delay using IRI model pink
and with the addition of the ionospheric delay using IONEX model green.
Doppler shift data, it has been used the DCP no. 109, located in French Guyana 5.1860◦ S and
52.687◦ W, on October 06–15, 2001 for high solar activity and on October 01–14, 2008 for low
solar activity.
Figure 10 shows DCP no. 109 location error without ionospheric and tropospheric
corrections blue, with the ionospheric correction using IONEX green and with both
corrections, ionospheric and tropospheric pink for the high solar activity period. Figure 11
makes use of the same color code and shows DCP no. 109 location errors for the low solar
activity period.
As in the simulated results, it is observed in real cases of DCP location that both
ionospheric and tropospheric effect should be inserted in the signal delay correction to
minimize the location error. It also observed that the ionospheric delay is very representative,
especially in maximum solar activity period, resulting in location error around 1 km, while
the tropospheric delay takes a smaller error scale, about some dozen of meters. In low
solar activity period, the ionosphere has much less influence on location error while the
troposphere is in the same order as mentioned above.
4. Conclusions and Recommendation

Due to high solar radiation in the equatorial region and Earth electric and magnetic field, the
electron density in the ionosphere layers suffers sensitive consequences mainly in Brazilian
regions or any areas close to magnetic equator. After some tests it was observed that, in such
regions, it is mandatory the use of an ionosphere correction model that takes into account
this effect called equatorial anomaly, depending on the solar activity intensity during the
period in which the location is accomplished. Then, a detailed study shows VTECU values
Error (km) Error (km)
0
0.5
1
1.5
2
2.5
3
3.5
0
0.5
1
1.5
2
2.5
3
3.5
10/1/08 0:00 10/6/01 12:00

10/7/01 0:00 10/6/01 19:52
10/2/08 0:00 10/6/01 21:35
10/1/08 19:05
10/2/08 18:17 10/7/01 12:00
10/3/08 0:00 10/8/01 0:00
10/2/08 21:54 10/7/01 20:49
10/4/08 0:00 10/8/01 12:00
10/3/08 17:30
Without correction
Without correction
Ionosphere (IONEX)
Ionosphere (IONEX)
10/9/01 0:00 10/8/01 20:04
10/5/08 0:00
10/4/08 20:18 10/9/01 12:00
10/6/08 0:00 10/10/01 0:00 10/9/01 19:22

10/5/08 19:30
Ionosphere + troposphere (NOAA)

10/10/01 12:00
Ionosphere + troposphere (NOAA)

10/7/08 0:00
10/11/01 0:00 10/10/01 18:33
10/8/08 0:00
10/7/08 19:45
Time
10/11/01 12:00
Time
10/9/08 0:00 10/12/01 0:00 10/11/01 19:35
10/8/08 18:57
10/12/01 12:00
10/10/08 0:00
and tropospheric effects correction
and tropospheric effects correction

10/13/01 0:00 10/12/01 18:46
10/11/08 0:00 10/10/08 17:22 10/13/01 12:00
10/12/08 0:00 10/14/01 0:00 10/13/01 18:02

Location of the DCP no.109—2001, October ionospheric
10/12/08 17:38
Location of the DCP no.109—2008, October ionospheric

10/14/01 12:00
10/13/08 0:00
10/15/01 0:00
10/14/08 0:00 10/13/08 16:49 10/15/01 12:00 10/15/01 18:16
10/14/08 16:01
10/15/08 0:00 10/16/01 0:00
Figure 11: Evaluation of DCP no. 109 error location on October 01–14, 2008 low solar activity.
Figure 10: Evaluation of DCP no. 109 error location on October 06–15, 2001 higher solar activity.
13
for each DCP location in order to have its location errors minimized as much as possible. A
comparison was made between two mentioned models IRI and IONEX taking into account
the local time of the day, the year of high or low solar activity, and the region where the
ionospheric effect can cause high location errors.
Through VTECU analysis, mainly in the equatorial region, it is observed the influence
on DCPs location, whereas better the ionospheric model better the platform location.
According to the results, it is verified that the ionospheric data provided by IONEX are,
generally, better than IRI data, especially in high solar activity periods. In this case, VTECU
values from IONEX model reach values 2-3 times higher than IRI model, resulting in an
ionospheric correction difference of up to 1 km.
When IRI and IONEX model data are used in high solar activity periods, the VTECU
values may be quite different and consequently DCPs location. In low solar activity periods
both models have shown similar performance.
Simulated ideal Doppler shift measurements were used for the first case. It is shown
that the geographical location has its error increased due to the ionospheric effect, and such
effect cannot be neglected when the platform location is made with real data. For this first
case, IONEX model allows an improvement of about 1 km in location 2001 around 18:00 UT
compared to IRI model.
After analyzing VTECU values and its influence on DCP no. 32590 location for the
simulated ideal case, it was chosen IONEX data to model the ionosphere. In addition to
the ionospheric correction, the second phase of the work considered the tropospheric effect
whose data depend largely on climatic conditions in the regions where DCPs are. Such data
can be found easily by the users through the National Climatic Data Center 11. The work is
based on real Doppler data of DCP no. 109, taking into account high and low solar activity,
in 2001 and 2008, respectively. The analysis is made under the following conditions: first, no
correction was applied; then only the ionospheric correction is applied to the location process;
finally ionospheric and tropospheric delays are corrected according to the conditions listed
above.
Results show that DCPs location error is minimized when the ionospheric effect is
considered, mainly in high solar activity period, when VTECU values are about 3 times larger
than low solar activity period. The tropospheric delays correction also decreases location
error, but in smaller proportions.
Finally, it is important to note that both delays, ionospheric and tropospheric, should
be corrected to obtain a better platform location, mainly when mobile platforms were used.
The choice of ionospheric model depends largely on the intensity of solar activity, because
the level of these activities influences the ionosphere density, due to high VTECU values.
References
1 J. A. Klobuchar, “Ionospheric effects on GPS,” in Global Positioning System: Theory and Applications,
B. W. Parkinson and J. J. Spilker, Eds., vol. 1, pp. 485–515, American Institute of Aeronautics and
Astronautics, Cambridge, UK, 1996.
2 C. C. Celestino, C. T. De Sousa, H. Koiti Kuga, and W. Yamaguti, “Errors due to the tropospheric
and ionospheric effects on the geographic location of data collection platforms,” Revista Brasileira de
Geofisica, vol. 26, no. 4, pp. 427–436, 2008.
3 C. C. Celestino, C. T. Sousa, W. Yamaguti, and H. K. Kuga, “Evaluation of tropospheric and
ionospheric effects on the geographic localization of data collection platforms,” Mathematical Problems
in Engineering, vol. 2007, Article ID 32514, 11 pages, 2007.
4 P. D. O. Camargo, J. F. G. Monico, and L. D. D. Ferreira, “Application of ionospheric corrections in the

equatorial region for L1 GPS users,” Earth, Planets and Space, vol. 52, no. 11, pp. 1083–1089, 2000.
5 D. Bilitza, “International Reference Ionospheric Model–IRI,” http://modelweb.gsfc.nasa.gov
/models/iri.html.
6 “CDDIS – NASA’s Archive of Space Geodesy Data,” ftp://cddis.gsfc.nasa.gov/pub/gps/products/
ionex/.
7 K. Aksnes, P. H. Andersen, and E. Haugen, “A precise multipass method for satellite Doppler
positioning,” Celestial Mechanics, vol. 44, no. 4, pp. 317–338, 1988.
8 S. Schaer and W. Gurtner, “IONEX: the IONosphere map eXchange format version 1,” in Proceedings
of the IGS AC Workshop, Darmstadt, Germany, February 1998.
9 J. F. G. Monico, Posicionamento pelo GNSS: Descrição, Fundamentos e Aplicações, UNESP, São Paulo,
Brazil, 2000.
10 D. D. McCarthy and G. Petit, “Tropospheric Model. International Earth Rotation and Reference
System Service IERS,” IERS Technical Note, no. 32, pp. 99–103, 2003.
11 “National Climatic Data Center – NOAA Satellite and Information Service,” http://www7.ncdc.
noaa.gov/IPS/mcdw/mcdw.html.
12 C. T. de Sousa, Geolocation of transmitters by satellites using Doppler shifts in near real time, Ph.D. thesis,
Space Engineering and Technology, Space Mechanics and Control Division, INPE - Instituto Nacional
de Pesquisas Espaciais, São José dos Campos, Brazil, 2000.
13 A. DasGupta, A. Paul, and A. Das, “Ionospheric total electron content TEC studies with GPS in the
equatorial region,” Indian Journal of Radio & Space Physics, vol. 36, pp. 278–292, 2007.
doi:10.1155/2012/137672
Research Article
Study of Stability of Rotational Motion of
Spacecraft with Canonical Variables
William Reis Silva,1 Maria Cecı́lia F. P. S. Zanardi,1

Regina Elaine Santos Cabette,2, 3
and Jorge Kennety Silva Formiga3, 4
1
Group of Orbital Dynamics and Planetology, São Paulo State University (UNESP),
Guaratinguetá 12516-410, SP, Brazil
2
Department of Mathematics, São Paulo Salesian University (UNISAL), Lorena 12600-100,
SP, Brazil
3
Space Mechanic and Control Division, National Institute for Space Research (INPE),
São José dos Campos 12227-010, SP, Brazil
4
Department of Aircraft Maintenance and Aeronautical Manufacturing, Faculty of
Technology (FATEC), São José dos Campos 12247-004, SP, Brazil
Correspondence should be addressed to William Reis Silva, reis.william@gmail.com
Received 16 July 2011; Revised 10 October 2011; Accepted 23 October 2011
Copyright q 2012 William Reis Silva et al. This is an open access article distributed under the
This work aims to analyze the stability of the rotational motion of artificial satellites in circular orbit
with the influence of gravity gradient torque, using the Andoyer variables. The used method in this
paper to analyze stability is the Kovalev-Savchenko theorem. This method requires the reduction
of the Hamiltonian in its normal form up to fourth order by means of canonical transformations
around equilibrium points. The coefficients of the normal Hamiltonian are indispensable in the
study of nonlinear stability of its equilibrium points according to the three established conditions
in the theorem. Some physical and orbital data of real satellites were used in the numerical
simulations. In comparison with previous work, the results show a greater number of equilibrium
points and an optimization in the algorithm to determine the normal form and stability analysis.
The results of this paper can directly contribute in maintaining the attitude of artificial satellites.
1. Introduction
Stability analysis of the rotational motion of a satellite taking into account the influence of
external torques is very important in maintaining the attitude to ensure the success of a space
mission.
Recently, some studies on the subject have been developed, and they motivated the
development of this paper. In 1, 2 is presented a study on the stability of the rotational
motion of artificial satellites in an elliptic orbit, considering disturbance due to gravity
gradient torque, using canonical formulations and the Andoyer canonical variables. These
studies use the procedure presented in 3 to determine a normal form of the Hamiltonian up
to 4th order.
In 4 is developed a numerical-analytical method for normalization of Hamiltonian
systems with 2 and 3 degrees of freedom. The normal form is obtained using the Lie-Hori
method 5. The stability analysis of the system is made by the Kovalev-Savchenko theorem
6. The most important role of this work is the results obtained analytically for the generating
function of 3rd order, necessary for determining the coefficients of the normal Hamiltonian
of 4th order.
Thus, the objective of this work is to optimize the stability analysis developed in
2 to determine the equilibrium points, the normal form for dynamical systems with two
degrees of freedom, and the applications of the Kovalev-Savchenko theorem 6. It will be
done by applying the expressions obtained in 4 for the coefficients of the normal 4th-order
Hamiltonian.
In this paper equilibrium points and/or regions of stability are established when
parcels associated with gravity gradient torque acting on the satellite are included in the
equations of rotational motion.
The Andoyer variables are used to describe the rotational motion of the satellite
in order to facilitate the application of methods of stability of Hamiltonian systems.
The Andoyer canonical variables 7 are represented by generalized moments L1 , L2 , L3
and by generalized coordinates
1 ,
2 ,
3 that are outlined in Figure 1. The angular
variables
1 ,
2 ,
3 are angles related to the satellite system Oxyz with axes parallel to the
spacecraft’s principal axes of inertia and equatorial system OXYZ with axes parallel to the
axis of the Earth’s equatorial system. Variables metrics L1 , L2 , L3 are defined as follows: L2
is the magnitude of the angular momentum of rotation L2 , L1 is the projection of L2 on the
z-axis of principal axis system of inertia L1 L2 cos J2 , where J2 is the angle between the
z-satellite axis and L2 , and L3 is the projection of L2 on the Z-equatorial axis L3 L2 cos I2 ,
where I2 is the angle between Z-equatorial axis and L2 .
The nonlinear stability of equilibrium points of the rotational motion is analyzed here
by the Kovalev-Savchenko theorem 6, which requires the normalized Hamiltonian up to
terms of fourth order around the equilibrium points.
The equilibrium points are found from the equations of motion. With the application
of the Kovalev-Savchenko theorem, it is possible to verify if they remain stable under the
influence of terms of higher order of the normal Hamiltonian.
In this paper numerical simulations were made for two hypothetical groups of
artificial satellites, considering them in a circular orbit and with symmetric shape in relation
to their physical and geometric characteristics. The satellites are classified as medium and
small sized; they have orbital data and physical characteristics similar to real satellites.
2. Equations of Motion
The Andoyer variables, defined above, are used to characterize the rotational motion of a
satellite around its center of mass 7, and the Delaunay variables describe the translational
motion of the center of mass of the satellite around the Earth 8.
√
The Delaunay variables L, G, H, l, g, h are defined as 8 L M μa, G
√
L 1 − e2 , H G cos I, l is the mean anomaly, g is the argument of perigee, h is the longitude
Plane perpendicular to the

zm principal axis of inertia
L3 y Plane perpendicular
to the angular
L2 ym
m om entum vector
z
x
L1 I2 J2
J2 Equatorial plane
l1
O Y
l2
l3 I2
xm
Figure 1: The Andoyer canonical variables.
of the ascending node, M is the mass of the satellite, μ is the Earth gravitational parameter, I
is the inclination of the orbit, a is the semimajor axis, and e is the orbital eccentricity.
Here it is assumed that the satellites are in a circular orbit, which differs from the
study presented in 2 which considers satellites in eccentric orbits. This consideration was
adopted to simplify the Hamiltonian of the problem, which is extensive 1, and to facilitate
the stability analysis of the equilibrium points.
Thus, assuming that the satellites have well-defined circular orbit, the main focus of
the work is only aimed at stability analysis of the rotational movement of the satellites in
study.
In this case the Hamiltonian of the problem is expressed in terms of the Andoyer and
Delaunay variables 9 as follows:

F L1 , L2 , L3 ,
2 ,
3 , L, G, H, l, g, h Fo L, L1 , L2 F1 L1 , L2 , L3 ,
2 ,
3 , L, G, H, l, g, h , 2.1
where Fo is the unperturbed Hamiltonian and F1 is the term of the Hamiltonian associated
with the disturbance due to the gravity gradient torque, both are described respectively by
10

μ2 M 3 1 1 1 1 1 1 1 1 1 1 2
Fo − − − L1 2 L2 2 − L2 − L1 2 cos 2
1 ,
2L2 2 C 2A 2B 2 A B 4 B A
6
μ M 2C − A − B
4
A−B
F1 H
1 m , L n H
2 m , L
n ,
L6 2 4
2.2
where m 2, 3 and n 1, 2, 3; A, B, and C are the principal moments of inertia of the satellite
on x-axis, y-axis, and z-axis, respectively; H1 and H2 are functions of the variables
m , Ln ,
where
2 and
3 appear in the arguments of cosines. The complete analytical expression is
presented in 1 for eccentric orbits. In this paper F1 will be simplified considering the satellite
in a circular orbit; its complete analytical expression is given in the Appendix.
Thus, the equations of motion associated with the Hamiltonian 2.1 are given by
d
i ∂F dLi ∂F
, − i 1, 2, 3. 2.3
dt ∂Li dt ∂
i
These equations are used to find the possible equilibrium points of the rotational
movement.
In this study, it is also considered that the satellite has two of its principal moments
of inertia equal, B A. With this relationship, the variable
1 will not be present in the
Hamiltonian, reducing the dynamic system to two degrees of freedom, a necessary condition
for applying the stability theorem chosen for analysis of equilibrium points.
3. Determining the Normal Form of the Hamiltonian of

the Rotational Motion
3.1. Normal Form of the Hamiltonian of 2nd Order and
Linear Stability Analysis
Because the variable
1 is a cyclic coordinate, the Hamiltonian 2.1 is written as
FL2 , L3 ,
2 ,
3 and the equations of motion 2.3 can be written in vector form as 1
ẇ JFw , 3.1
where w is the state vector and Fw is the matrix of partial derivatives with respect to their
respective variables given by
⎛ ⎞
∂F
⎜ ∂
2 ⎟
⎜ ⎟
⎛ ⎞ ⎜ ∂F ⎟

2 ⎜ ⎟
⎜ ⎟
⎜L 2 ⎟ ⎜ ∂L2 ⎟
w ⎜ ⎟
⎝
3 ⎠, F w ⎜
⎜ ∂F ⎟
⎟ 3.2
⎜ ⎟
⎜ ⎟
L3 ⎜ ∂
3 ⎟
⎜ ⎟
⎝ ∂F ⎠
∂L3
and J a symplectic matrix given by
⎛ ⎞
0 1 0 0
⎜−1 0 0 0⎟
J ⎜
⎝0
⎟. 3.3
0 0 1⎠
0 0 −1 0
Now, it is necessary to linearize the system around the equilibrium point. However,
it is convenient to make a translation of the coordinates in the Hamiltonian 2.1 so that the
origin coincides with the equilibrium point under study; it means that
L1 L1e ,

2
2e q1 ,
L2 L2e p1 , 3.4

3
3e q2 ,
L3 L3e p2 ,
where L1e ,
2e , L2e ,
3e , and L3e are the coordinates of equilibrium points.
With this translation we can expand the Hamiltonian in Taylor series around the
new origin that is nothing more than expand the Hamiltonian in the neighborhoods of the
equilibrium point when q1 p1 q2 p2 0. Then,
&
∞
F
m , Ln Fk
m , Ln F2
m , Ln F3
m , Ln F4
m , Ln · · · , 3.5
k2
where m 2, 3, n 1, 2, 3, and Fk is the Hamiltonian expanded up to terms of order k, with

k 2, 3, 4, . . . .
The Hessian P of the problem is calculated from the Hamiltonian expanded up to 2nd
order around the equilibrium point:
⎛ ⎞
∂2 F ∂2 F ∂2 F ∂2 F
⎜ ∂q2 ∂q1 ∂p1 ∂q1 ∂q2 ∂q1 ∂p2 ⎟
⎜ ⎟
⎜ 21 ⎟
⎜ ∂F ∂2 F ∂2 F ∂2 F ⎟
⎜ ⎟
⎜ ∂p1 ∂q1 ∂p12 ∂p1 ∂q2 ∂p1 ∂p2 ⎟
⎜ ⎟
P ⎜ 2 ⎟. 3.6
⎜ ∂F 2
∂F 2
∂F ∂F ⎟
2
⎜ ⎟
⎜ ∂q2 ∂q1 ∂q2 ∂p1 ∂q22 ∂q2 ∂p2 ⎟
⎜ ⎟
⎜ 2 ⎟
⎝ ∂F ∂2 F ∂2 F ∂2 F ⎠
∂p2 ∂q1 ∂p2 ∂p1 ∂p2 ∂q2 ∂p22
Thus, the system of linearized equations can be written as
Ẇ JP W, 3.7
where W is the state vector in the new variables
⎛ ⎞
q1
⎜ p1 ⎟
W⎜ ⎟
⎝q2 ⎠. 3.8
p2
So, to find the eigenvalues of the reduced linear system, we should diagonalize the
matrix JP present in 3.7 by means of the determinant detλI −JP 0, where I is an identity
matrix and λ are the eigenvalues of the matrix JP . If these eigenvalues are pure imaginary
λj ±iωj , the normal form of 2nd-order Hamiltonian can be expressed as 2
2 ω
& j
H2∗ qj , pj qj 2 pj 2 . 3.9
j1
2
3.2. Extension of the Normal Form of the Hamiltonian to Higher Orders

Extending the process of normalization of the Hamiltonian to higher orders H3 , H 4 , . . ., the
Hamiltonian normal in terms of variables qi and pi can be expressed by
& 2
ωi 2
H qj , pj qj pj 2 H3∗ qj , pj H4∗ qj , pj · · · . 3.10
j1
2
To obtain the Hamiltonian normal H3∗ , H4∗ , . . ., it is necessary to use the method of Lie-
Hori. According to 11, it is desirable that the normal Hamiltonian is written in complex
variables before using this method.
This transformation of variables is given by
1 1
qj √ xj iyj , pj √ ixj yj 3.11
2 2
with inverse given by
1 1
xj √ qj − ipj , yj √ −iqj pj . 3.12
2 2
Thus, the normal Hamiltonian of 2nd order is given by
&2
H2∗ xj , yj iωj xj yj 3.13
j1
and normal Hamiltonian until the 4th order in complex variables is represented by
&2

H ∗ xj , yj iωj xj yj H3∗ xj , yj H4∗ xj , yj · · · , 3.14
j1
where H3∗ x, y and H4∗ x, y can be expressed briefly as
&
H3∗ x, y h3,α,β xα yβ , 3.15
|α| |β|3
&
H4∗ x, y h4,α,β xα yβ . 3.16
|α| |β|4
In 3.15 and 3.16, is always assumed it α, β > 0. For 3.15, |α| |β| 3 means that
the sum of the exponents of the new variables must obey the order of the Hamiltonian H3∗ ,
and, for 3.16, |α| |β| 4 means that the sum of the exponents of the new variables must
obey the order of the Hamiltonian H4∗ . These equations are presented entirely in 1, 4.
Thus, we have the Hamiltonian written in complex variables. It is important to note
that H2 is already normalized, and now we can apply the method of Lie-Hori to find the
normal form to higher orders H3∗ , H4∗ .
3.2.1. Series Method of Lie-Hori

The normal form is obtained from the expansion of the Hamiltonian in terms of the Lie series
given by 12, 13
1 1
H new H ∗ {H ∗ , G} {{H ∗ , G}, G} {{{H ∗ , G}, G}, G} · · · , 3.17
2! 3!
where H new is normal Hamiltonian obtained by canonical transformations around the

equilibrium point, G Gxi , yi is the generating function developed as a power series, where
each degree of Gn is a sum of homogeneous polynomials of degree n, H ∗ H ∗ x, y is the
original Hamiltonian, and {H ∗ , G} is the Poisson brackets of order r s − 2, where r and s
represent the order of the polynomials H ∗ and G, respectively.
The Poisson bracket is defined as
∂H ∗ ∂G ∂H ∗ ∂G
{H ∗ , G} − . 3.18
∂x ∂y ∂y ∂x
The new Hamiltonian H new after the new canonical transformation in the neighbor-
hoods of identity can be expressed as 2
H new H2new H3new H4new · · · . 3.19

So the new Hamiltonian, ordered by degree up to 4th order, can be expressed following
the next form 11, 13:
2nd order: H2new H2∗ , 3.20

; <
3rd order: H3new H3∗ H2∗ , G3 , 3.21
; < 1 ;; ∗ < < ; <
4th order: H4new H4∗ H3∗ , G3 H2 , G3 , G3 H2∗ , G4 . 3.22
2!
As H2new is already given in its normal form, the next step is to calculate the generating
function which leads to the normal form up to 4th-order terms. This procedure allows to
obtain the minimum number of nonresonant monomials defined in 12; thus, the generating
function Gn is chosen to eliminate the complex variables in the total Hamiltonian 3.5 in
which the monomials xj and yj have different exponents of the desired degree, leaving
only the monomials that carry on the resonance of the intrinsic Hamiltonian systems. The
following is a procedure to obtain the normal Hamiltonian to 4th order 1, 4.
I It is known that Hamiltonian systems are naturally resonant in even orders, 2nd,
4th, and so forth 10, 12, so to find the normal Hamiltonian until the 4th order, the 3rd
disturbance H3new must first be removed from H new , it means
H3new 0. 3.23
Assuming H3∗ is briefly defined in 3.15 to eliminate terms of 3rd order, we consider
the generating function G3 x, y expressed in abbreviated form as
&
G3 x, y g3,α,β xα yβ . 3.24
|α| |β|3
Then it is necessary to find G3 x, y of the third-order homological equation 3.21,

where
; ∗ <
H2 , G3 −H3∗ . 3.25
After some algebraic manipulations, we find G3 x, y given by 4
& h3,α,β
G3 x, y −H I xα y β . 3.26
|α| |β|3
β − α, W
Note that this equation is well defined if β − α, W

/ 0, where β − α, W
represents
an inner product between β − α β1 − α1 , β2 − α2 , . . . , βn − αn and W λ1 , λ2 , . . . , λn
iω1 , iω2 , . . . , iωn which are pure imaginary eigenvalues obtained from the quadratic part of
normal Hamiltonian. The complete analytical representation for β − α, W
was presented in
4.
For the resonance condition it is not accepted that β − α, W

/
0, which is always true
for monomials of third-order Hamiltonian when considering only the natural resonance of
the Hamiltonian.
II This process is repeated for the polynomial of degree 4 finding G4 x, y that will
eliminate some of monomials H4∗ in 3.16. The remaining monomials in H4new are called
resonant monomials 3. Then 3.22 can be reduced as 1
; < ; < 1 ;; ∗ < <

G4 , H2∗ H4new H4∗ H3∗ , G3 H2 , G3 , G3 , 3.27
2!
; <
G4 , H2∗ H4new F4. 3.28
Equation 3.28 is called homological equation of 4th order, and it can be rewritten as
; <
G4 , H2∗ F4 − H4new ,
; < 3.29
G4 , H2∗ R4 .
Assume that R4 x, y and G4 x, y will be expressed in short form as
&
R4 x, y r4,α,β xα yβ , 3.30
|α| |β|4
&
G4 x, y g4,α,β xα yβ . 3.31
|α| |β|4
After some algebraic manipulations, we find G4 x, y given by 4
& r4,α,β
G4 x, y H I xα y β , 3.32
|α| |β|4
α − β, W
where α − β, W
/
0 represents an inner product as mentioned above.
Thus, using 3.34 in 3.29 it is possible to find H4new after some algebraic
manipulations.
This procedure can be repeated for any order that is required in order to find a normal
form of the new Hamiltonian H new . In this paper it is considered the 4th order of the H new .
3.3. Normal Form of the Hamiltonian of 4th Order and

Nonlinear Stability Analysis
As we get the normal form of 2nd-order Hamiltonian 3.9, we can apply the method
described previously to find the Hamiltonian normal until the 4th order using the
Hamiltonian in complex variables 3.14 represented by 1, 4

H ∗ xj , yj iω1 x1 y1 iω2 x2 y2 H3∗ x1 , y1 , x2 , y2 H4∗ x1 , y1 , x2 , y2 · · · , 3.33
where H3∗ x1 , y1 , x2 , y2 and H4∗ x1 , y1 , x2 , y2 are obtained from expressions
&
H3 ∗ x1 , y1 , x2 , y2 h3,α1 ,β1 ,α2 ,β2 x1 α1 y1 β1 x2 α2 y2 β2 ,
|α| |β|3
& 3.34
∗
H4 x1 , y1 , x2 , y2 h4,α1 ,β1 ,α2 ,β2 x1 α1 y1 β1 x2 α2 y2 β2 .
|α| |β|4
Using the method of Lie-Hori presented before, it is possible to determine the

generating function G3 x1 , y1 , x2 , y2 in order to satisfy the homological equation 3.25,
where G3 x1 , y1 , x2 , y2 was determined analytically in 4. This generating function is
responsible for the elimination of the 3rd-order terms of the Hamiltonian 3.33.
The generating function G3 x1 , y1 , x2 , y2 is used to find the terms of 4th order, F4 ,
and perform the separation described in 3.28, obtaining H4 and G4 x1 , y1 , x2 , y2 . The
coefficients of the normal form H4new can be calculated using the terms obtained in H4 and
the eigenvalues of H2 .
Thus, the normal Hamiltonian in complex variables separated by degree can be
expressed as follows4:
2nd order: H2new iω1 x1 y1 iω2 x2 y2 , 3.35
3 rd order: H3new 0, 3.36

2 2
4th order: H4new δ11 x1 y1 δ12 x1 y1 x2 y2 δ22 x2 y2 , 3.37
where ωj j 1, 2 is the imaginary part of eigenvalues associated with the matrix defined by
the product of a matrix symplectic of order 4 with the Hessian of the Hamiltonian expanded
in Taylor series up to 2nd order around the equilibrium point; δij are real coefficients obtained
from the combination of hk,α,β xα yβ k 3, 4.
However, the Kovalev-Savchenko theorem requires that the normal Hamiltonian is in
real variables, and with the applications of the transformation of variables 3.12 the normal
form Hamiltonian can been expressed as follows:
ω1 2 ω
2

2nd order: H2new q1 p1 2 q2 2 p2 2 , 3.38
2 2
3 rd order: H3new 0, 3.39

4th order: H4new δ11 q1 4 p1 4 2q1 2 p1 2 δ12 q1 2 q2 2 q1 2 p2 2 p1 2 q2 2 p1 2 p2 2
3.40
δ22 q2 4 p2 4 2q2 2 p2 2 ,
where δij are the coefficients of the Hamiltonian normal 4th order which are expressed
analytically in 4.
4. Kovalev-Savchenko Theorem
To use the stability theorem, the normal Hamiltonian of the problem is necessary as it was
discussed above.
Considering the normal Hamiltonian Ho , an analytics function of coordinates qν and
generalized moments pν to a fixed point P is expressed by
&
2
ωνo &2 δo
ν,υ
Ho Rν Rν Rυ O5 , 4.1
ν1
2 ν,υ1
4
where O5 represents higher-order terms; ωνo is the imaginary part of eigenvalues associated
with the matrix defined by the product of a 4th-order matrix symplectic with the Hessian
of the Hamiltonian expanded in Taylor series up to 2nd order around the equilibrium point;
o
δν,υ depend on the eigenvalues ωνo and the coefficients of the Hamiltonian expanded in Taylor
series of 3rd and 4th order around the equilibrium point and they are presented analytically
in Formiga 4, and
Rm qm 2 pm 2 with m 1, 2. 4.2
The stability analysis is performed here by the Kovalev-Savchenko theorem 6 which
ensures that the motion is Lyapunov stable if the following conditions are satisfied.
i The eigenvalues of the reduced linear system are pure imaginary ±iωo1 and ±iωo2 .
ii The condition
k1 ω1o k2 ω2o
/0 4.3
is valid for all k1 and k2 integers satisfying the inequality
|k1 | |k2 | ≤ 4. 4.4
iii The determinant Do must satisfy the inequality

Do − δ11
o
ω2o 2 − 2δ12
o
ω1o ω2o δ22
o
ω1o 2 / 0, 4.5
o
where δν,υ are the coefficients of the normal 4th-order Hamiltonian.
This theorem says that a Hamiltonian reduced in its normal form up to 4th order, in
the absence of the resonance condition of the eigenvalues associated and if the condition 4.5
is satisfied, it is guaranteed the existence of tori invariant in a neighborhood small enough of
equilibrium position 14–16.
Figure 2: Flowchart representative of the stability analysis of equilibrium points.
5. Computational Algorithm for the Normal Form of the Hamiltonian

of Rotational Motion and Analysis of Stability
In order to synthetize the process to analyze the stability of equilibrium points, the logical
sequence of the algorithm is now presented.
5.1. Stability Analysis of Equilibrium Points

See Figure 2.
Table 1: Quantitative summary of the classification of equilibrium points.
Satellite Stable points Unstable points

58
MP 2 Failed 1st condition Failed 2nd condition Failed 3rd condition
58 0 0
43
PP-1 7 Failed 1st condition Failed 2nd condition Failed 3rd condition
43 0 0
40
PP-2 10 Failed 1st condition Failed 2nd condition Failed 3rd condition
40 0 0
6. Numerical Simulations
As already mentioned, we consider two types of satellites: medium sized MP, which
has similar orbital characteristics with the American satellite PEGASUS 17, and small
sized PP, which has similar orbital characteristics of the Brazilian data collection satellite
SCD-1 and SCD-2 18. All the numerical simulations were developed using the software
MATHEMATICA.
Table 1 shows a quantitative summary of the equilibrium points found and their
stability, according to the criteria of Kovalev-Savchenko theorem 6. It can be observed that,
when the first condition is not satisfied, the eigenvalues associated with the matrix JP are
real or are not pure imaginary. This equilibrium point is not linearly stable.
There were found totals of 60 equilibrium points for the MP satellite, 50 equilibrium
points to the PP-1 satellite and 50 equilibrium points for the PP-2 satellite. Tables 2, 3 and 4,
show two equilibrium points found in the simulations, one being Lyapunov stable and the
other unstable, to satellites MP, PP1 and PP2, respectively.
For the Lyapunov stable equilibrium points of Tables 2, 3, and 4 respectively, Table 5
shows the values of the angles I2 , J2 , the rotation speed ω, and the rotation period T .
These values characterize the nonexistence of singularities in these points; it means that
the angles I 2 and J2 are not null or close to zero. By Figure 1, when these angles I2 and J2
are null or close to zero, the Andoyer variables l1 , l2 , and l3 are indeterminate, because it is
difficult to determine the intersection between the involved planes in the definitions of these
variables. This analysis was performed for all equilibrium points found in the simulations.
Table 6 shows the same conditions for the unstable points of Tables 2, 3, and 4.
7. Conclusion
In this paper we presented a semianalytical stability of the rotational motion of artificial
satellites, considering the influence of gravity gradient torque for symmetric satellite
in a circular orbit. Applications in numerical simulations, performed with the software
MATHEMATICA, were made to two types of satellites: medium MP and small PP.
Initially the points of equilibrium were determined using the physical, orbital, and
attitude characteristics of each satellite. Then the algorithm for stability analysis was applied
and it was obtained 2 stable equilibrium points for the satellite MP, 7 stable points for the
satellite PP-1, and 10 stable points for the satellite PP-2.
Table 2: Satellite MP: A B 3, 9499 ∗ 10−1 kg·km2 , C 1, 0307 ∗ 10−1 kg·km2 .
Equilibrium points Lyapunov stable Unstable

L1 kg · km2 /s 2, 50725 ∗ 10−4 −1, 34202 ∗ 10−11
L2 kg · km2 /s 3, 85388 ∗ 10−4 −4, 19645 ∗ 10−10
L3 kg · km2 /s −2, 28562 ∗ 10−4 −3, 53991 ∗ 10−10
l2 rad −1, 6 ∗ 10−7 0,0934286
l3 rad 0,5235 0,41354
Table 3: Satellite PP-1: A B 9.20 ∗ 10−6 kg·km2 , C 13 ∗ 10−6 kg·km2 .

L1 kg · km2 /s 2, 69365 ∗ 10−7 7, 13725 ∗ 10−11
L2 kg · km2 /s 2, 07921 ∗ 10−5 2, 14416 ∗ 10−10
L3 kg · km2 /s 1, 05633 ∗ 10−5 1, 07843 ∗ 10−10
l2 rad 0,066560 −0,090962
l3 rad 0,4 0,07
Table 4: PP-2: A B 28, 7 ∗ 10−4 kg·km2 , C 36, 9 ∗ 10−4 kg·km2 .

L1 kg · km2 /s 6, 25509 ∗ 10−5 2, 48081 ∗ 10−11
L2 kg · km2 /s 7, 24797 ∗ 10−4 2, 77796 ∗ 10−11
L3 kg · km2 /s 2, 17564 ∗ 10−4 −2, 54792 ∗ 10−11
l2 rad 0,098581 −1,68042
l3 rad 1,42527 1,05253
For the satellite MP it was gotten only two equilibrium points because this satellite
has similar characteristics to the satellite PEGASUS, which is tumbling 17. For satellites PP-
1 and PP-2 were obtained many other equilibrium points, but most of them were discarded,
because they lead to the Andoyer variables, a condition of uniqueness it means that the
angles I2 and J2 are null or close to zero, and the Andoyer variables l1 , l2 , and l3 are
indeterminate.
It can be observed that a larger number of stable equilibrium points were determined
in comparison with the results of 1, 2, which show the stability analysis of the rotational
motion with the gravity gradient torque, but with the satellite in an elliptic orbit.
An optimization was done in the algorithm of determining the normal form of the
Hamiltonian and the stability analysis, using expressions obtained by 4 for the coefficients
of the normal 4th-order Hamiltonian. The introduction of these expressions enabled a more
effective stability analysis for equilibrium points in comparison with the results of 1, 2. It is
possible to say that the numerical simulations have become less laborious allowing analysis
of data in more numbers .
This paper presents results that can directly contribute in maintaining the attitude
of artificial satellites. Once the regions of stability are known for the rotational motion, a
smaller number of maneuvers to maintain the desired attitude can be accomplished. In this
case, a fuel economy can be generated to the satellite with propulsion system, increasing the
satellite’s lifetime.
Table 5: Analysis of possible singularities of the Anloyer variables for satellites MP, PP1 of and PP2.
Lyapunov stable equilibrium point of Table 2, 3 and 4, respectively.

Satellite I2 rad J2 rad ω rad/s T s Situation
MP 2,20566 0,862451 0,00373909 1680,4 Accepted
PP-1 1,03788 1,55784 1,59939 3,92848 Accepted
PP-2 1,26592 1,48439 0,196422 31,9882 Accepted
Table 6: Analysis of possible singularities of the Anloyer variables for satellites MP, PP1 and PP2.
Unstable equilibrium point of Table 2, 3, and 4, respectively.

Satellite I2 rad J2 rad ω rad/s T s Situation
−9
MP 0,56693 1,53881 −4, 071 ∗ 10 −1, 543 ∗ 109 Accepted
−5
PP-1 1,04377 1,23145 1, 64936 ∗ 10 380948,0 Accepted
PP-2 2,73176 0,46676 7, 52836 ∗ 10−9 8, 346 ∗ 108 Accepted
Appendix
The disturbance Hamiltonian F1 , due to the gravity gradient torque, for a satellite in a circular
orbit with two of its principal moments of inertia equal, B A, is given by 19
= = 2
μ4 M 7 C−A 3 L1 1
F1 −
L6 M 2 L2 2
2 2 2 2
1 3 H L3 H L3
× − 1 −3
2 8 G L2 G L2
3
− sin2I sin2I2 cosh −
3
8

3
− sin2 Isin2 I2 cos2h − 2
3
8
2
3 H
− 1−3 sin2I2 sin2J2 cos
2
16 G
2
3 H
1−3 sin2 I2 sin2 J2 cos2
2
16 G

3 L3 L3
1− 1 2 sin2I sin2J2 cosh −
3 −
2
16 L2 L2

3 L3 L3
1 1−2 sin2I sin2J2 cosh −
3
2
16 L2 L2

3 L3
1− sin2 I sinI2 sin2J2 cos2h − 2
3
2
16 L2

3 L3
− 1− sin I sinI2 sin2J2 cos2h −
3 2
2
2
16 L2

3 L3
− 1− sin2I sinI2 sin2 J2 cosh −
3 2
2
16 L2

3 L3
1 sin2I sinI2 sin 2 J2 cosh −
3 − 2
2
16 L2

3 L3 2 2
− 1− sin Isin2 J2 cos2h − 2
3 2
2
32 L2

3 L3 2 2
− 1 sin Isin J2 cos2h − 2
3 − 2
2
2
32 L2
2 = 2
3 L1 1 3 2 L3
− sin I 1 − 3
2 L2 2 8 L2
> ?
× cos 2l 2g

3 H
− 1− sinI sinI2
8 G
> ?
× cos 2l 2g − h −
3

3 H 2 2
− 1 sin I2
16 G
> ?
× cos 2l 2g 2h − 2
3
F
3 H 2 2 > ?
− 1− sin I2 cos 2l 2g − 2h − 2
3
16 G
9 > ?
sin2 I sin2I2 sin2J2 cos 2l 2g
2
32
9 > ?
sin2 I sin2I2 sin2J2 cos 2l 2g −
2
32

3 H L3 L3
− 1 1− 1 2 sinI sin2J2
16 G L2 L2
> ?
× cos 2l 2g h −
3
2

3 H L3 L3
− 1 1 1−2 sinI sin2J2
16 G L2 L2
> ?
× cos 2l 2g h −
3 −
2

3 H L3 L3
1− 1− 1 2 sinI sin2J2
16 G L2 L2
> ?
× cos 2l 2g − h −
3 −
2

3 H L3 L3
1− 1 1−2 sinI sin2J2
16 G L2 L2
> ?
× cos 2l 2g − h −
3
2

3 H 2 L3
1 1− sinI2 sin2J2
32 G L2
> ?
× cos 2l 2g 2h − 2
3 −
2

3 H 2 L3
− 1 1 sinI2 sin2J2
32 G L2
> ?
× cos 2l 2g 2h − 2
3
2

3 H 2 L3
1− 1− sinI2 sin2J2
32 G L2
> ?
× cos 2l 2g − 2h − 2
3
2

3 H 2 L3
− 1− 1 sinI2 sin2J2
32 G L2
> ?
× cos 2l 2g − 2h − 2
3 −
2
9
− sin2 Isin2 I2
32
> ?
× sin2 J2 cos 2l 2g 2
2
9 > ?
− sin2 Isin2 I2 sin2 J2 cos 2l 2g − 2
2
32

3 H L3
1 1− sinI sinI2 sin2 J2
16 G L2
> ?
× cos 2l 2g h −
3 2
2

3 H L3
− 1 1 sinI sinI2 sin2 J2
16 G L2
> ?
× cos 2l 2g h −
3 − 2
2

3 H L3
− 1− 1− sinI sinI2 sin2 J2
16 G L2
> ?
× cos 2l 2g − h −
3 − 2
2

3 H L3
1− 1 sinI sinI2 sin2 J2
16 G L2
> ?
× cos 2l 2g − h −
3 2
2

3 H 2 L3 2 2
− 1− 1− sin J2
64 G L2
> ?
× cos 2l 2g 2h − 2
3 2
2

3 H 2 L3 2 2
1− 1 sin J2
64 G L2
> ?
× cos 2l 2g 2h − 2
3 − 2
2

3 H 2 L3 2 2
1 1− sin J2
64 G L2
> ?
× cos 2l 2g − 2h − 2
3 − 2
2

3 H 2 L3 2 2
1 1 sin J2
64 G L2

> ?
× cos 2l 2g − 2h − 2
3 2
2 ,
A.1
where: A B and C are the principal moments of inertia on x-axis, y-axis and z-axis
respectively, M is the mass of the satellite, μ is the gravitational constant, I is the inclination of
the orbit, I2 is the inclination of the plane of angular momentum with the plane of the equator,
and J2 is the inclination of the principal plane with the plane of the angular momentum.
In A.1 the generalized moments L1 , L2 , L3 are implicit in some terms, using the
definitions of generalized moments present in the introduction and trigonometric properties,
they can be explicit replacing.

L3 2
sinI2 1− ,
L2 2
L3 2
sin2 I2 1 − 2 ,
L2

2L3 L3 2
sin2I 2 1 − 2,
L2 L2
A.2
L1 2
sinJ2 1 − 2 ,
L2
L1 2
sin2 J2 1 − 2 ,
L2

2L1 L1 2
sin2J2 1 − 2.
L2 L2
Acknowledgment
This present work was supported by CAPES.
References
1 R. E. S. Cabette, Estabilidade do movimento rotacional de satélites artificiais, Dissertation, National Institute
for Space Research, São Paulo, Brazil, 2006.
2 R. V. de Moraes, R. E. S. Cabette, M. C. Zanardi, T. J. Stuchi, and J. K. Formiga, “Attitude stability of

artificial satellites subject to gravity gradient torque,” Celestial Mechanics & Dynamical Astronomy, vol.
104, no. 4, pp. 337–353, 2009.
3 A. L. F. Machuy, Cálculo efetivo da forma normal parcial para o problema de Hill, Dissertação de mestrado-
instituto de matemática, UFRJ—Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil, 2001.
4 J. K. Formiga, Formas normais no estudo da estabilidade para L4 no problema fotogravitacional, Dissertation,
National Institute for Space Research, São José dos Campos, Brazil, 2009.
5 G. Hori, “Theory of general perturbations for non-canonical system,” Astronomical Society of Japan,
vol. 23, no. 4, pp. 567–587, 1971.
6 A. M. Kovalov and A. Ia. Savchenko, “Stability of uniform rotations of a rigid body about a principal
axis,” Prikladnaia Matematika i Mekhanika, vol. 39, no. 4, pp. 650–660, 1975.
7 H. Kinoshita, “First order perturbation of the two finite body problem,” Publications of the Astronomical
Society of Japan, vol. 24, pp. 423–439, 1972.
8 C. D. Murray and S. F. Dermott, Solar system dynamics, Cambridge University Press, Cambridge, 1999.
9 M. C. Zanardi, Movimento rotacional e translacional acoplado de satélites artificiais, Dissertação de
mestrado, Instituto de Tecnologia Aeronáutica, São José dos Campos, São Paulo, Brazil, 1983.
10 M. C. Zanardi, “Study of the terms of coupling between rotational and translational motions,” Celestial
11 T. J. Stuchi, “KAM tori in the center manifold of the 3-D Hill problem,” in Advanced in Space, O. C.
Winter and A. F. B. Prado, Eds., vol. 2, pp. 112–127, INPE, São Paulo, Brazil, 2002.
12 G. Hori, “Theory of general perturbations with unspecified canonical variables,” Publications of the
13 S. Ferraz-Mello, Canonical Perturbation Theories-Degenerate Systems and Resonance, Springer, New York,
NY, USA, 2007.
14 V. I. Arnold, Geometrical Methods in the Theory of Ordinary Di erential Equations, vol. 250 of Grundlehren
der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Science], Springer, New York,
NY, USA, 1983.
15 V. I. Arnold, Equações Diferenciais Ordinárias, Mir, Moscow, Russia, 1985.
16 V. I. Arnold, “Métodos matemáticos da mecânica clássica,” Editora Mir Moscou, p. 479, 1987.
17 J. U. Crenshaw and P. M. Fitzpatrick, “Gravity effects on the rotational motion of a uniaxial artificial
satellite,” AIAAJ, vol. 6, article 2140, 1968.
18 H. K. Kuga, V. Orlando, and R. V. F. Lopes, “Flight dynamics operations during leap for the INPE’s
second environmental data collecting satellite SCD 2,” Journal of the Brazilian Society of Mechanical
Sciences, vol. 21, pp. 339–344, 1999.
19 W. R. Silva, Estudo da estabilidade do movimento rotacional de satélites artificiais com variáveis canônicas,
Qualificação de mestrado, UNESP—Universidade Estadual Paulista, São Paulo, Brazil, 2011.
doi:10.1155/2012/267875
Research Article
Analysis of Filtering Methods for Satellite
Autonomous Orbit Determination Using
Celestial and Geomagnetic Measurement
Xiaolin Ning,1, 2, 3 Xin Ma,1, 2, 3 Cong Peng,1, 2, 3

Wei Quan,1, 2, 3 and Jiancheng Fang1, 2, 3
1
School of Instrumentation Science & Opto-electronics Engineering, Bei Hang University (BUAA),
Beijing 100191, China
2
Science and Technology on Inertial Laboratory, Beijing 100191, China
3
Fundamental Science on Novel Inertial Instrument & Navigation System Technology Laboratory,
Beijing 100191, China
Correspondence should be addressed to Xiaolin Ning, ningxiaolin@buaa.edu.cn
Received 15 July 2011; Revised 30 September 2011; Accepted 5 October 2011
Copyright q 2012 Xiaolin Ning et al. This is an open access article distributed under the Creative
Satellite autonomous orbit determination OD is a complex process using filtering method to
integrate observation and orbit dynamic equations effectively and estimate the position and
velocity of a satellite. Therefore, the filtering method plays an important role in autonomous orbit
determination accuracy and time consumption. Extended Kalman filter EKF, unscented Kalman
filter UKF, and unscented particle filter UPF are three widely used filtering methods in satellite
autonomous OD, owing to the nonlinearity of satellite orbit dynamic model. The performance of
the system based on these three methods is analyzed under different conditions. Simulations show
that, under the same condition, the UPF provides the highest OD accuracy but requires the highest
computation burden. Conclusions drawn by this study are useful in the design and analysis of
autonomous orbit determination system of satellites.
1. Introduction
Orbit determination OD of satellite plays a significant role in satellite missions, aiming at
estimating the ephemeris of a satellite at a chosen epoch accurately. To date, the conventional
OD system is dominated by measurements based on 1 ground tracking approaches 1
such as range, range rate, and angle, and 2 Global Position System measurement 2, 3. The
orbit determination technologies have shown fair performance on various space missions.
However, its high cost, lack of robustness to loss of contact, space segment degradation, and
OD system
Sensor subsystem Model subsystem Filter subsystem
Star Dynamic orbit Time update

vector model
Star
Star sensor
Earth Measurement Measurement Position and

vector model update velocity
Earth sensor
S
N
Magnetic Measurement Estimator

vector
Magnetometer
Earth
Figure 1: The process of orbit determination.
other factors promote the application of autonomous OD system, which is less costly and less
vulnerable in hostile environment 4.
In general, orbit determination is the process of estimating the satellite’s state variables
position and velocity by comparing in statistical sense the difference between the
measurement data and the estimated data. Orbit determination system, as shown in Figure 1,
usually includes sensor subsystem, model subsystem, and filter subsystem. Sensor subsystem
contains sensing instruments, such as star sensor, earth sensor, and magnetometer, in order
to measure and process the original measurements which are functions of state variables.
Model system generates estimated data including state model and measurement model. In
the filter subsystem, the optimal algorithms filtering methods process both data from sensor
subsystem and from model subsystem and then estimate state variables.
Owing to the nonlinear dynamic model of satellite orbit motion, the filtering method
applied in OD system should be appropriate for nonlinear system 5, 6. Extended Kalman
filter EKF, unscented Kalman filter UKF, and unscented particle filter UPF are three
main methods used in satellite OD system. The EKF is based on the analytical Taylor series
expansion of the nonlinear systems and measurement equations. It works on the principle
that the state distribution is approximated by a Gaussian random variable. However, the
Taylor series approximations in EKF introduce large errors due to the neglected nonlinearities
7. The UKF uses the true nonlinear model and a set of sigma sample points produced by
the unscented transformation to capture the mean and covariance of state, but the UKF has
the limitation that it does not apply to general non-Gaussian distribution 8, 9. The particle
filter PF is a computer-based method for implementing a recursive Bayesian filter by Monte
Carlo simulations. The performance of the PF largely depends on the choices of importance
sampling density and resampling scheme 10, 11. Among many improved PF methods, UPF
is a hybrid of the UKF and the particle filter which uses the UKF to get better importance
sampling density 12, 13. It combines the merits of unscented transformation and particle
filtering and avoids their limitations.
A variety of autonomous orbit determination methods have been proposed and
explored, including a magnetometer-based OD method 14, 15, a celestial OD method
16, 17, a landmark OD method 18, 19, and an X-ray pulsar OD method 20, 21. The first
two methods can be used in low earth orbit LEO satellite autonomous orbit determination
system. Thus, in this paper, these two OD methods are selected for analysis.
This paper is divided into five sections. After this introduction, the basic descriptions
of three filtering methods in autonomous OD system are given in Section 2. Then, the state
model and measurement models in OD model subsystem are described in detail in Section 3.
In Section 4, simulations are shown for analyzing and comparing three filtering methods.
Finally, conclusions are drawn in Section 5.
2. Filtering Methods
The best known algorithms to solve the problem of autonomous satellite orbit determination
are the EKF, UKF, and UPF. In this section, we shall present the theories of the three filter
algorithms. These algorithms will be incorporated into the filtering framework based on the
dynamic state-space model as follows:
xk fxk−1 , k − 1 wk ,
2.1
zk hxk , k vk ,
where xk−1 denotes the state of the system at time k − 1, zk denotes the observations at step k,
wk denotes the process noise, and vk denotes the measurement noise. The mappings f and h
represent the process and measurement models. Ewk wTj Qk , Evk vTj Rk , for all k, j, and
Qk is the process noise covariance at step k, Rk is the measurement noise covariance at step k.
2.1. Extended Kalman Filter

A Kalman filter that linearizes about the current mean and covariance is referred to as an
extended Kalman filter or EKF. The EKF is the minimum mean-square-error estimator based
on the Taylor series expansion of the nonlinear functions. For example,
∂fxk
fxk f x9k|k−1 | xk − x9k|k−1 · · · . 2.2
∂xk xk 9xk|k−1
Using only the linear expansion terms, it is easy to derive the update equations for the mean
and covariance of the Gaussian approximation to the distribution of the states 12.
The equations for the extended Kalman filter fall into two groups: time update
equations and measurement update equations. The specific equations for the time and
measurement updates are presented below as shown in 2.3∼2.8 22.
1 Time Update
Predicted state estimate:
x9k|k−1 f9xk−1 , k − 1. 2.3
Predicted estimate covariance:
P−k Φk Pk−1 ΦTk Qk−1 . 2.4

The time update equations project the state, xk , and covariance, Pk , estimates from the
previous time step k − 1 to the current time step k, Φk is the state transition matrix at step k,
which is defined to be the following Jacobians:
∂f
Φk | . 2.5
∂xk xk 9xk|k−1
2 Measurement Update

Near-Optimal Kalman gain:
−1
Kk P−1 −1 T −1
k Hk Hk Pk Hk Rk
T
. 2.6
Updated state estimate:

x9k x9k|k−1 Kk zk − h x9k|k−1 , k . 2.7
Updated estimate covariance:
Pk I − Kk Hk P−k , 2.8
where Kk is known as the Kalman gain. The measurement update equations correct the state
and covariance estimates with the measurement zk . Hk is the observation matrix at step k,
which is defined to be the following Jacobians:
∂h
Hk | . 2.9
∂xk xk 9xk|k−1
The major drawback of EKF is that it only uses the first order terms in the Taylor series
expansion. Sometimes it may introduce large estimation errors in a nonlinear system and lead
to poor representations of the nonlinear functions and probability distributions of interest. As
a result, this filter can diverge 23.
2.2. Unscented Kalman Filter

The unscented Kalman filter UKF 8, 24 uses the unscented transformation to capture
the mean and covariance estimates with a minimal set of sample points. The UKF process
is identical to the standard EKF process with the prediction-estimation recursive loop. The
exception is that the UKF uses the sigma points and the nonlinear equations to compute the
predicted states and measurements and the associated covariance matrices. If the dimension
of state is n × 1, the 2n 1 sigma point and their weight are computed by 9
τ
X0,k x9k , W0 ,
n τ

√ 1
Xi,k x9k n τ Pk | k , Wi , i 1, 2, . . . , n, 2.10
2n τ

i
√ 1
Xi n,k x9k − n τ Pk | k , Wi n ,
i 2n τ
D
where τ ∈ R, Pk | ki is the ith column of the matrix square root. The UKF process can
be described as follows.
1 Time 0, initialize the UKF with x90 and P0 as follows:
x90 Ex0 ,
+ , 2.11
P0 E x0 − x90 x0 − x90 T .
2 Time k, define 2n 1 sigma points from
> ?
Xk−1 X0,k Xi,k Xi n,k , i 1, 2, . . . , n. 2.12
The equations for the UKF fall into two groups the same as EKF: time update equations
and measurement update equations. The specific equations for the time and measurement
updates are presented below.
1 Time Update
Xk|k−1 fXk−1 , k − 1,
&
2n
x9−k Wi Xi,k|k−1 ,
i0
&
2n
> ? > ?T
P−k Wi Xi,k|k−1 − x9−k · Xi,k|k−1 − x9−k Qk , 2.13
i0

Zk|k−1 h Xk|k−1 , k ,
&
2n
z9−k Wi Zi,k|k−1 .
i0
2 Measurement Update
&
2n
> ?> ?T
Pz9k z9k Wi Zi,k|k−1 − z9−k Zi,k|k−1 − z9−k Rk ,
i0
&
2n
> ?> ?T
Px9k z9k Wi Xi,k|k−1 − x9−k Zi,k|k−1 − z9−k ,
i0 2.14
Kk Px9k z9k P−1
z9k z9k ,

x9k x9−k Kk zk − z9−k ,
Pk P−k − Kk Pz9k z9k KTk .

2.3. Unscented Particle Filter

The unscented particle filter UPF is a hybrid of the UKF and the particle filter which uses
the UKF to get better importance sampling density. A pseudo-code description of UPF is as
follows 11–13.
1 Initialization: Time 0.
Generate N samples xi0 , i 1, 2, . . . , N from the prior px0 , and set the importance
weight w0i of each sample 1/N:
+ , T 1
x9i0 E xi0 , Pi0 E xi0 − x9i0 xi0 − x9i0 , w0i . 2.15
N
2 Time k.
I a Update the particles with the UKF:
i calculate sigma points from {9xik−1 , Pik−1 } using 2.12,

ii propagate particle into future by 2.13,
iii incorporate new observation to update the measurement by 2.14 and obtain
{9xik , Pik }.
b Sample a new particle x9ik and make x9ik ∼ qxik | xik−1 , zk N9xik , Pik .
II Compute the importance weight wik and normalize the importance weights wki :

p zk | x9ik p x9ik | xik−1
wik i
wk−1 · ,
q x9ik | xik−1 , zk−1
2.16
wik
wki 6N ,
i1 wik
where pzk | x9ik is likelihood probability distribution, which is given by measurement model
zk hxk , k vk , p9xik | xik−1 is the forward transition probability distribution, which is given
by process model xk fxk−1 , k − 1 wk , q9xik | xik−1 , zk−1 is the proposal distribution 12.
III Resampling step:
The basic idea of resampling is to eliminate particles with small weights and to con-
centrate on particles with large weights. Multiply/suppress particles {9xik , Pik } with high/low
importance weights wki , respectively, to obtain N random particles {:xik , P: i }.
k
IV Output step:

The overall state estimation and covariance are
&
N
x9k wki x:ik ,
i1
2.17
&
N &
N T
Pk :i
wki P w i
:
x i
− 9
x k :
x i
− 9
x k .
k k k k
i1 i1
3. System Models
3.1. State Model
The state model dynamical model of the celestial OD system for a near-Earth satellite based
on the orbital dynamics in the Earth-Centered Inertial ECI frame J2000.0 is
dx dy dz
vx , vy , vz ,
dt dt dt
2
dvx μx Re 7.5z2
− 3 · 1 − J2 − 1.5 ΔFx ,
dt r r r2
2
dvy μy Re 7.5z2 3.1
− 3 · 1 − J2 − 1.5 ΔFy ,
dt r r r2
2
dvz μz Re 7.5z2
− 3 · 1 − J2 − 4.5 ΔFz ,
dt r r r2

r x2 y2 z2 .
Equation 3.1 can be written in a general state equation as
Ẋt fXt, t wt, 3.2
where X x y z vx vy vz T is the state vector. x, y, z, vx , vy , vz are satellite positions

and velocities of the three axes, respectively, μ is the gravitational constant of earth, J2 is the
second zonal coefficient and has the value 0.0010826269 25, and Re is the earth’s radius.
ΔFx , ΔFy , ΔFz are the perturbations including high order nonspherical earth perturbations,
third-body perturbations, atmospheric drag perturbations, solar radiation perturbations, and
other perturbations, which are considered as process noises wt.
3.2. Celestial Orbit Determination and Its Measurement

The celestial OD method is based on the fact that the position of a celestial body in the inertial
frame at a certain time is known and that its position measured in the spacecraft body frame
is a function of the satellite’s position. To earth satellite, stars are distributed all over the sky,
and the positions of Earth are fixed at a certain time. The geometric relationship among stars,
the Earth, and satellite enables us to determine the position of the satellite 26.
Satellite celestial OD methods can be broadly separated into two major approaches:
directly sensing horizon method and indirectly sensing horizon method. In this paper, the
directly sensing horizon method is used.
The angle between a star and the earth, α, as shown in Figure 2, is a kind of directly
sensing horizon measurement of satellite celestial OD system, which is measured by star
sensor and earth sensor. The measurement model using the star-earth angle is given by 27

−s · r
α arccos να , 3.3
r
Star
Satellite
Earth
Figure 2: The measurement of celestial OD system.
where r is the position vector of the satellite, which is the same as that in 3.2, s is the position
vector of the star in the earth-centered inertial frame, να is the measurement noise.
Assuming a measurement Z1 α and measurement noise V1 vα , 3.2 can be
written as a general measurement equation
Z1 t h1 Xt, t V1 t. 3.4
3.3. Geomagnetic Orbit Determination and Its Measurement

Geomagnetic OD system relies on measurements from a three-axis magnetometer to
determine satellite position and orbit. It uses a model of Earth’s magnetic field and a model of
orbital dynamics to predict the time-varying magnitude of Earth’s magnetic field vector at the
space. OD system compares the time history of the predicted magnitude and the measured
magnitude time history in filter sense to obtain the optimal estimated state position and
velocity 14.
3.3.1. Magnetic Model

Two main models used for describing Earth’s magnetic vector in the geodetic reference frame
are World Magnetic Model WMM and International Geomagnetic Reference Field IGRF
28. The WMM 2005 is selected in this paper for geomagnetic orbit determination 29.
According to the WMM model 2005, the vector field B can be written as the gradient
of a potential function
Br, λ, θ, t −∇V r, λ, θ, t, 3.5
where r, λ, θ represent the radius, the longitude, and the colatitude in a spherical, geocentric
reference frame, respectively.
This potential V can be expanded in terms of spherical harmonics:
&
N & & N n 1 &
n
a n
> m ? m
V r, λ, θ, t Vnm a gn t cosmλ hm
n t sinmλ P̆n cos θ,
n1 m0 n1
r m0
3.6
Geographical Magnetic
north pole north pole
Bn
Geographical Magnetic
north pole north pole Be
Bn
Satellite Bv
Be
S
Bt N
Satellite Satellite
orbit
Bv
Earth
Figure 3: The measurement of geomagnetic OD system.
where N 12 is the degree of the expansion of the WMM, a is the standard Earth’s magnetic
reference radius, gnm t and hmn t are the time-dependent Gauss coefficients of degree n and
order m, and P̆nm cos θ are the Schmidt normalized associated Legendre polynomials.
3.3.2. Magnetic Measurement Model

Based on the relationship between magnetic vector, which is obtained by the magnetometer,
and the earth magnetic model, the measurement model can be written as
Bs Asb Abi Ait Bt vB , 3.7
where Bs is the magnetic vector of local position in sensor coordinates, which can be
obtained from vector magnetometer system consisting of three mutually orthogonal, single-
axis magnetometers. Bt Bn Be Bv T is the magnetic vector of local position in geocentric
coordinates, and it can be obtained from WMM according to local longitude, latitude, and
height, as shown in Figure 3; Asb , Abi , and Ait are the transformation matrices from satellite
body coordinates to sensor coordinates, from earth inertial coordinates to satellite body
coordinates, and from earth inertial coordinates to geocentric coordinates, respectively. vB
is the measurement noise.
Assuming a measurement Z2 Bs and measurement noise V2 vB , 3.7 can be
written as a general measurement equation as
Z2 t h2 Xt, t V2 t. 3.8
4. Analysis and Comparison

4.1. Simulation Condition
The trajectory used in the following simulation is a LEO satellite whose orbital parameters
are semimajor axis a 7136.635444 km, eccentricity e 1.809 × 10−3 , inclination i 65◦ , right
ascension of the ascending node Ω 30◦ , and the argument of perigee ω 30◦ . The orbit
Position estimation error Velocity estimation error

3500 6
3000 5
Velocity error (m/s)

Position error (m)
2500
4
2000
3
1500
2
1000
500 1
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time (min) Time (min)
EKF EKF
UKF UKF
UPF UPF
a b
Figure 4: Three filtering methods results of celestial OD system T 3 s.

3000 4.5
4
2500
3.5
Position error (m)
2000 3
2.5
1500
2
1000 1.5
1
500
0.5
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
EKF EKF
UKF UKF
UPF UPF
a b
Figure 5: Three filtering methods results of geomagnetic OD system T 3 s.
and attitude data of the satellite are produced by the Satellite Tool Kit STK software 30.
The accuracy of star sensor and earth sensor is selected 3 and 0.02◦ , respectively. The stellar
database used in simulation is the Tycho stellar catalog 31. The magnetometer measurement
and geomagnetic model accuracy is considered as 100 nT 32.
4.2. Performances under Different Sampling Intervals

Figures 4 and 5 show the performances comparison among the EKF, UKF, and UPF methods
of celestial OD system and geomagnetic OD system, respectively. Data is obtained with a 3 s
Table 1: Performance of celestial OD system under different sampling intervals.
RMS after convergence Maximum after convergence

Sampling interval
Position error/m Velocity error/m/s Position error/m Velocity error/m/s
EKF 203.318211 0.196622 532.987272 0.486705
T 3s UKF 161.312723 0.162900 357.545600 0.407503
UPF 159.756079 0.160780 354.555359 0.402378
EKF 271.640953 0.287641 703.000581 0.613708
T 15 s UKF 245.939302 0.219864 593.829098 0.563749
UPF 245.229683 0.219566 593.802048 0.563396
EKF 934.238939 0.976641 2457.895170 2.348656
T 60 s UKF 736.876288 0.699942 2322.291896 2.302674
UPF 735.166932 0.698808 2314.761249 2.294924
Table 2: Performance of geomagnetic OD system under different sampling intervals.

Sampling interval
EKF 591.253628 0.601946 1129.365510 1.061735
T 3s UKF 376.894372 0.371566 877.909403 0.799659
UPF 376.516863 0.366538 861.513975 0.783449
EKF 1481.673752 1.358867 2851.660177 2.607626
T 15 s UKF 705.765450 0.648228 1325.501267 1.276198
UPF 705.161263 0.647876 1323.976107 1.274585
EKF 4343.783162 4.308921 14643.275741 11.712547
T 60 s UKF 3904.890544 3.633798 10403.528541 10.328902
UPF 3904.747892 3.633460 10401.279139 10.328554
sampling interval during the 600 min period 6 orbits. Tables 1 and 2 present the details of
the simulation results of celestial OD system and geomagnetic OD system under different
sampling intervals, respectively.
The simulations in Figures 4 and 5 suggest that the EKF-based OD system performance
is the worst. In contrast, UPF-based OD system provides the highest OD accuracy. As the
details in Tables 1 and 2, regardless the celestial OD and geomagnetic OD system, the
different sampling intervals can strongly affect the OD accuracy. OD performance is degraded
remarkably with increasing sample interval. However, under the same sampling interval, the
EKF method is the most sensitive to the sampling interval, for the nonlinear error increases
rapidly with the longer sampling interval. In contrast, the UKF and UPF perform distinctly
better.
4.3. Performance under Different Noise Distributions

This subsection reports how different noise distributions affect the OD performances using
three filters. We selected three common noise distributions in navigation, and they are normal
distribution, student’s t distribution, and uniform distribution 33.

5000 8
4500 7
4000
6

Position error (m)
3500
3000 5
2500 4
2000 3
1500
2
1000
500 1
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
EKF EKF
UKF UKF
UPF UPF
a b
Figure 6: Celestial OD results of three filtering methods under Student’s t noise distributions.

8000 8
7000 7
6000 6
Position error (m)
5000 5
4000 4
3000 3
2000 2
1000 1
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
EKF EKF
UKF UKF
UPF UPF
a b
Figure 7: Geomagnetic OD results of three filtering methods under Student’s t noise distributions.
Figures 6 and 7 show the OD results of celestial OD system and geomagnetic OD

system using three filters under student’s t noise distributions, respectively. All performance
curves were obtained with 15 s sampling interval during the 600 min period 6 orbits. Tables
3 and 4 present the details of the simulation results of celestial OD system and geomagnetic
OD system under three different noise distributions, respectively.
As the results in Figures 6 and 7 showed, the UPF-based geomagnetic OD system
provides the highest OD accuracy. As the details in Tables 3 and 4 demonstrated, OD
performance under different noise distribution is similar. In general, the EKF performance
Table 3: Performance of celestial OD system under different noise distributions.

Noise distribution
EKF 271.640953 0.287641 703.000581 0.613708
Normal
distribution UKF 245.939302 0.219864 593.829098 0.563749
UPF 245.229683 0.219566 593.802048 0.563396
EKF 387.618044 0.411669 827.358300 0.972190
Student’s t
UPF 246.294197 0.250511 587.154229 0.639735
EKF 385.803609 0.398842 994.714264 0.902353
Uniform
UPF 349.537325 0.338473 880.417206 0.782641
Table 4: Performance of geomagnetic OD system under different noise distributions.

Noise distribution
EKF 1481.673752 1.358867 2851.660177 2.607626
Normal
UPF 705.161263 0.647876 1323.976107 1.274585
EKF 1059.874849 0.828895 2900.800178 3.184135
Student’s t
UPF 669.959004 0.604513 1593.085809 1.436468
EKF 1246.529130 1.207728 4029.187774 3.253167
Uniform
UPF 776.526144 0.767452 1985.704699 1.781594
is the worst and the UPF performance is the best, no matter what measurement errors are
chosen.
4.4. Computation Cost of Three Methods

Besides the accuracy, the computation cost is another essential requirement to evaluate the
performance of filtering methods. Table 5 gives the computation cost of the three methods
for the celestial orbit determination system and the geomagnetic orbit determination system,
respectively. As in the theoretical value of computation cost, where Φ is the process Jacobian,
n is the order of the Φ, and in the simulation n equals 6. The simulation results presented
here were run on a 2.66 GHz Inter Core2 Duo CPU with 32-bit Windows 7 system. The
simulation time of celestial OD system in Table 5 demonstrates that the UPF demands the
highest computation time, which is almost twenty times sample number higher than UKF,
and EKF requires almost a quarter of the computation time of UKF. However, the simulation
time of geomagnetic OD system is not the same amount as celestial OD system, and the EKF-
based geomagnetic OD system takes significantly longer time, since the time for computing
measurement Jacobians takes a lot of computer resource.
Table 5: Comparison of computation cost.
Theoretical value of computation cost Simulation value of computation cost

Computation Computation
Filter
time per orbit time per orbit
method Φ is full matrix Φ is diagonal matrix
of celestial of geomagnetic
OD systems OD system s

EKF CEKF 3n3 3n2 4n CEKF 6n2 4n 0.3 9.5

UKF CUKF 2n 12n 14n 5
3 2
CUKF 2n 12n 14n 5
3 2
1.6 11.7
UPF

sample CUPF sample num · CUKF CUPF sample num · CUKF 39.5 313.9
num 20
5. Conclusion
The problem of choosing a suitable filtering method for the orbit determination application
has been studied here. Three filtering methods for the autonomous orbit determination using
either celestial or geomagnetic measurements have been studied and their performances have
been compared for the estimation problem.
The algorithms are tested with STK satellite orbit data, and the simulation results
demonstrate that UPF yields the best OD accuracy and the EKF yields the worst under the
same condition. The main reason is that the state equations and measurement equations
for autonomous orbit determination system are significantly nonlinear as well as the non-
Gaussian errors.
In addition, the paper analyzed the computation cost of the three filtering methods,
and UPF-based OD system can provide the highest OD accuracy, though it requires the largest
computation time. However, the UPF can finally meet the real-time requirements, as with the
development of computer technology.
Acknowledgments
The work described in this paper was supported by the National Natural Science Foundation
of China 60874095 and Hi-Tech Research and Development Program of China. The
authors would like to thank all members of Science and Technology on Inertial Laboratory
and Fundamental Science on Novel Inertial Instrument & Navigation System Technology
Laboratory, for their useful comments regarding this work effort.
References
1 D.-J. Lee and K.T. Alfriend, “Precise real-time satellite orbit estimation using the unscented kalman
filter,” AAS/AIAA Space Flight Mechanics Symposium AAS 03-230, pp. 853–1872, Advances in the
Astronautical Sciences, Ponce, Puerto Rico, 2003.
2 P. C. P. M. Pardal, H. K. Kuga, and R. V. de Moraes, “Recursive least squares algorithms applied to
satellite orbit determination, using GPS signals,” in Proceedings of the 8th International Conference on
Signal Processing, Robotics and Automation, pp. 167–172, World Scientific and Engineering Academy
and Society, Cambridge, UK, 2009.
3 O. Montenbruck and P. Ramos-Bosch, “Precision real-time navigation of LEO satellites using global
positioning system measurements,” GPS Solutions, vol. 12, no. 3, pp. 187–198, 2008.
4 B. Maurizio, L. J. John, S. J. Peter, and T. Denver, “Advanced stellar compass: onboard autonomous
orbit determination, preliminary performance,” Annals of the New York Academy of Sciences, vol. 1017,
pp. 393–407, 2004.
5 D.-J. Lee and K. T. Alfriend, “Sigma point filters for efficient orbit estimation,” in Proceedings of the
AAS/AIAA Astrodynamics Conference, vol. 116, pp. 349–372, Advances in the Astronautical Sciences,
Big Sky, Mont, USA, August 2003.
6 P. C. P. M. Pardal, H. K. Kuga, and R. V. de Moraes, “Nonlinear sigma point Kalman filter applied
to orbit determination using GPS measurement,” in Proceedings of the 22nd International Meeting of the
Satellite Division of The Institude of Navigation, Savannah, Ga, USA, September 2009.
7 D.-J. Lee and T. A. Kyle, “Adaptive sigma point filtering for state and parameter estimation,” in
Proceedings of the AIAA/AAS Astrodynamics Specialist Conference and Exhibit, Providence, Rhode Island,
USA, August 2004.
8 S. J. Julier and J. K. Uhlmann, “A new extension of the Kalman filter to nonlinear systems,” in
Proceedings of the International Society for Optical Engineering (SPIE ’97), vol. 3, no. 1, pp. 182–193, April
1997.
9 S. Julier, J. Uhlmann, and H. F. Durrant-Whyte, “A new method for the nonlinear transformation
of means and covariances in filters and estimators,” Institute of Electrical and Electronics Engineers.
Transactions on Automatic Control, vol. 45, no. 3, pp. 477–482, 2000.
10 M. S. Arulampalam, S. Maskell, N. Gordon, and T. Clapp, “A tutorial on particle filters for online
nonlinear/non-Gaussian Bayesian tracking,” IEEE Transactions on Signal Processing, vol. 50, no. 2, pp.
174–188, 2002.
11 D. Watzenig, M. Brandner, and G. Steiner, “A particle filter approach for tomographic imaging based
on different state-space representations,” Measurement Science and Technology, vol. 18, no. 1, pp. 30–40,
2007.
12 R. van der Merwe, A. Doucet, N. de Freitas, and E. Wan, “The unscented particle filter,” Tech. Rep.
CUED/F-INFENG/TR 380, Cambridge University Engineering Department, Cambridge, UK, 2000.
13 O. Payne and A. Marrs, “An unscented particle filter for GMTI tracking,” in Proceedings of the IEEE
Aerospace Conference, vol. 3, pp. 1869–1875, March 2004.
14 M. L. Psiaki, L. Huang, and S. M. Fox, “Ground tests of magnetometer-based autonomous navigation
MAGNAV for low-earth-orbiting spacecraft,” Journal of Guidance, Control, and Dynamics, vol. 16, no.
1, pp. 206–214, 1993.
15 H. Jung and M. L. Psiaki, “Tests of magnetometer-sun-sensor orbit determination using flight data,”
in Proceedings of the AIAA Guidance, Navigation, and Control Conference and Exhibit, Montreal, Canada,
August 2001.
16 C. J. Gramling and A. C. Long, “Autonomous navigation using the TDRSS Onboard Navigation
System TONS,” Advances in Space Research, vol. 16, no. 12, pp. 77–80, 1995.
17 A. C. Long, D. Leung, D. Folta, and C. Gramling, “Autonomous navigation of high-earth satellites
using celestial objects and Doppler measurements,” in Proceedings of the AIAA/AAS Astrodynamics
Specialist Conference, Denver, Colo, USA, August 2000.
18 G. M. Levine, “A method of orbital navigation using optical sightings to unknown landmarks,” AIAA
Journal, vol. 4, no. 11, pp. 1928–1931, 1966.
19 W. L. Brogan and J. L. Lemay, “Orbit navigation with known landmark tracking,” in Proceedings of
the Joint Automatic Control Conference of American Automatic Control Council, Institute of Electrical and
Electronics Engineers, New York, NY, USA, 1968.
20 S. I. Sheikh, The use of variable celestial X-Ray sources for spacecraft navigation, Ph.D. thesis, University of
Maryland, College Park, Maryland, 2005.
21 A. A. Emadzadeh and J. E. Speyer, Navigation in Space by X-Ray Pulsar, Springer, New York, NY, USA,
2011.
22 G. Welch and G. Bishop, “An introduction to the Kalman filter,” Tech. Rep. TR95-041, University of
North Carolina, Chapel Hill, NC, USA, 1995.
23 E. A. Wan and R. van der Merwe, “The unscented Kalman filter for nonlinear estimation,” in
Proceedings of the IEEE Adaptive Systems for Signal Processing, Communications, and Control Symposium,
pp. 153–158, 2000.
24 S. J. Julier and J. K. Uhlmann, “A general method for approximating nonlinear transformations of
probability distributions,” Tech. Rep. RRG, Department of Engineering Science, University of Oxford,
1996.
25 D. A. Vallado, Fundamentals of Astrodynamics and Applications, Springer, New York, NY, USA, 3rd
edition, 2007.
26 R. H. Battin, An Introduction to the Mathematics and Methods of Astrodynamics, American Institute
Aeronautics and Astronautics, New York, NY, USA, 1999.
27 X. L. Ning and J. C. Fang, “An autonomous celestial navigation method for LEO satellite based on
unscented Kalman filter and information fusion,” Aerospace Science and Technology, vol. 11, no. 2-3, pp.
222–228, 2007.
28 I. A. O. Geomagnetism, W. G. V. P. Aeronomy, C. C. Finlay et al., “International Geomagnetic Refer-
ence Field: the eleventh generation,” Geophysical Journal International, vol. 183, no. 3, pp. 1216–1230,
2010.
29 S. Mclean, S. Macmillan, and S. Maus, “The US/UK World Magnetic Model for 2005–2010,” Tech.
Rep. NESDIS/NGDC-2, NOAA, 2005.
30 Satellite Tools Kit Suite Version 8.1.1 Help System, Analytical Graphics, New York, NY, USA, 2007.
31 The Hipparcos and Tycho Catalogues (ESA SP-1200) ESA, Noordwijk, The Netherlands, 1997.
32 M. N. Filipski and E. J. Abdullah, “Nanosatellite navigation with the WMM2005 geomagnetic field
model,” Turkish Journal of Engineering and Environmental Sciences, vol. 30, no. 1, pp. 43–55, 2006.
33 S. P. Mertikas, “Error distributions and accuracy measures in navigation: an overview,” Tech. Rep.
113, 1985.
doi:10.1155/2012/405870
Research Article
The Orbital Dynamics of Synchronous Satellites:
Irregular Motions in the 2 : 1 Resonance
Jarbas Cordeiro Sampaio,1 Rodolpho Vilhena de Moraes,2

and Sandro da Silva Fernandes3
1
Departamento de Matemática, Universidade Estadual Paulista (UNESP),
12516-410 Guaratinguetá-SP, Brazil
2
Instituto de Ciência e Tecnologia, Universidade Federal de São Paulo (UNIFESP),
3
Departamento de Matemática, Instituto Tecnológico de Aeronáutica (ITA),
Correspondence should be addressed to Jarbas Cordeiro Sampaio, jarbascordeiro@gmail.com
Received 7 July 2011; Accepted 27 September 2011
Copyright q 2012 Jarbas Cordeiro Sampaio et al. This is an open access article distributed under
The orbital dynamics of synchronous satellites is studied. The 2 : 1 resonance is considered; in other
words, the satellite completes two revolutions while the Earth completes one. In the development
of the geopotential, the zonal harmonics J20 and J40 and the tesseral harmonics J22 and J42 are con-
sidered. The order of the dynamical system is reduced through successive Mathieu transforma-
tions, and the final system is solved by numerical integration. The Lyapunov exponents are used
as tool to analyze the chaotic orbits.
1. Introduction
Synchronous satellites in circular or elliptical orbits have been extensively used for naviga-
tion, communication, and military missions. This fact justifies the great attention that has
been given in literature to the study of resonant orbits characterizing the dynamics of these
satellites since the 60s 1–14. For example, Molniya series satellites used by the old Soviet
Union for communication form a constellation of satellites, launched since 1965, which have
highly eccentric orbits with periods of 12 hours. Another example of missions that use eccen-
tric, inclined, and synchronous orbits includes satellites to investigate the solar magneto-
sphere, launched in the 90s 15.
The dynamics of synchronous satellites are very complex. The tesseral harmonics
of the geopotential produce multiple resonances which interact resulting significantly in
nonlinear motions, when compared to nonresonant orbits. It has been found that the orbital
elements show relatively large oscillation amplitudes differing from neighboring trajectories
11.
Due to the perturbations of Earth gravitational potential, the frequencies of the lon-
gitude of ascending node Ω and of the argument of pericentre ω can make the presence of
small divisors, arising in the integration of equation of motion, more pronounced. This
phenomenon depends also on the eccentricity and inclination of the orbit plane. The impor-
tance of the node and the pericentre frequencies is smaller when compared to the mean
anomaly and Greenwich sidereal time. However, they also have their contribution in the reso-
nance effect. The coefficients l, m, p which define the argument φlmpq in the development of
the geopotential can vary, producing different frequencies within the resonant cosines for the
same resonance. These frequencies are slightly different, with small variations around the
considered commensurability.
In this paper, the 2 : 1 resonance is considered; in other words, the satellite completes
two revolutions while the Earth carries one. In the development of the geopotential, the zonal
harmonics J20 and J40 and the tesseral harmonics J22 and J42 are considered. The order of the
dynamical system is reduced through successive Mathieu transformations, and the final sys-
tem is solved by numerical integration. In the reduced dynamical model, three critical angles,
associated to the tesseral harmonics J22 and J42 , are studied together. Numerical results show
the time behavior of the semimajor axis, argument of pericentre and of the eccentricity. The
Lyapunov exponents are used as tool to analyze the chaotic orbits.
2. Resonant Hamiltonian and Equations of Motion

In this section, a Hamiltonian describing the resonant problem is derived through successive
Mathieu transformations.
Consider 2.1 to the Earth gravitational potential written in classical orbital elements
16, 17
μ & ∞ & l & l & −∞

μ ae l
V Jlm Flm IGlpq e cos φlmpq M, ω, Ω, θ , 2.1
2a l2 m0 p0 q ∞ a a
where μ is the Earth gravitational parameter, μ 3.986009 × 1014 m3 /s2 , a, e, I, Ω, ω, M are

the classical keplerian elements: a is the semimajor axis, e is the eccentricity, I is the in-
clination of the orbit plane with the equator, Ω is the longitude of the ascending node, ω is the
argument of pericentre, and M is the mean anomaly, respectively; ae is the Earth mean
equatorial radius, ae 6378.140 km, Jlm is the spherical harmonic coefficient of degree l and
order m, Flmp I and Glpq e are Kaula’s inclination and eccentricity functions, respectively.
The argument φlmpq M, ω, Ω, θ is defined by
π
φlmpq M, ω, Ω, θ qM l − 2p ω mΩ − θ − λlm l − m , 2.2
2
where θ is the Greenwich sidereal time, θ ωe t ωe is the Earth’s angular velocity, and t is
the time, and λlm is the corresponding reference longitude along the equator.
In order to describe the problem in Hamiltonian form, Delaunay canonical variables

are introduced,

√
L μa, G μa1 − e2 , H μa1 − e2 cosI,
2.3

M, g ω, h Ω.
L, G, and H represent the generalized coordinates, and

, g, and h represent the conjugate
momenta.
9
Using the canonical variables, one gets the Hamiltonian F,
μ2 &∞ & l
F9 Rlm , 2.4
2L2 l2 m0
with the disturbing potential Rlm given by
&
l &
∞

Rlm Blmpq L, G, H cos φlmpq
, g, h, θ . 2.5
p0 q−∞
The argument φlmpq is defined by
π
φlmpq
, g, h, θ q
l − 2p g mh − θ − λlm l − m , 2.6
2
and the coefficient Blmpq L, G, H is defined by
&∞ & l & l & −∞

μ2 μae l
Blmpq Jlm Flmp L, G, HGlpq L, G. 2.7
l2 m0 p0 q ∞
L2 L2
The Hamiltonian F9 depends explicitly on the time through the Greenwich sidereal
time θ. A new term ωe Θ is introduced in order to extend the phase space. In the extended
E is given by
phase space, the extended Hamiltonian H
E F9 − ωe Θ.
H 2.8
For resonant orbits, it is convenient to use a new set of canonical variables. Consider
the canonical transformation of variables defined by the following relations:
X L, Y G − L, Z H − G, Θ Θ,
2.9
x
g h, y g h, z h, θ θ,
where X, Y, Z, Θ, x, y, z, θ are the modified Delaunay variables.

E , resulting from the canonical transformation defined by 2.9,

The new Hamiltonian H
is given by
2 &∞ & l
E μ − ωe Θ
H Rlm , 2.10
2X 2 l2 m0
where the disturbing potential Rlm is given by
&
l &
∞

Rlm
Blmpq X, Y, Z cos φlmpq x, y, z, θ . 2.11
p0 q−∞
Now, consider the commensurability between the Earth rotation angular velocity ωe
and the mean motion n μ2 /X 3 . This commensurability can be expressed as
qn − mωe ∼
0, 2.12
considering q and m as integers. The ratio q/m defining the commensurability will be de-
noted by α. When the commensurability occurs, small divisors arise in the integration of the
E with frequencies qn −
equations of motion 9. These periodic terms in the Hamiltonian H
mωe are called resonant terms. The other periodic terms are called short- and long-period
terms.
The short- and long-period terms can be eliminated from the Hamiltonian H E by ap-
plying an averaging procedure 12, 18:
- . 2π 2π
E 1
H E dξsp dξlp .
H 2.13
4π 2 0 0
The variables ξsp and ξlp represent the short- and long-period terms, respectively, to be elimi-
nated of the Hamiltonian H E .
The long-period terms have a combination in the argument φlmpq which involves only
the argument of the pericentre ω and the longitude of the ascending node Ω. From 2.10 and
2.11, these terms are represented by the new variables in the following equation:
&
∞ &
l &
l &
∞

J
H
Blmpq X, Y, Z cos l − 2p y − z mz . 2.14
lp
l2 m0 p0 q−∞
The short-period terms are identified by the presence of the sidereal time θ and mean
J
anomaly M in the argument φlmpq ; in this way, from 2.10 and 2.11, the term H
sp in the new
variables is given by the following equations:
&
∞ &
l &
l &
∞

J
H
sp

Blmpq X, Y, Z cos q x − y − mθ ζp . 2.15
l2 m0 p0 q−∞
The term ζp represents the other variables in the argument φlmpq , including the argument of
the pericentre ω and the longitude of the ascending node Ω, or, in terms of the new variables,
y − z and z, respectively.
A reduced Hamiltonian H Er is obtained from the Hamiltonian H E when only secular
E
and resonant terms are considered. The reduced Hamiltonian Hr is given by
2 &∞
Er μ − ωe Θ B
H 2j,0,j,0 X, Y, Z
2X 2 j1
2.16
&
∞ &
l &
l

Blmpαm X, Y, Z cos φlmpαm x, y, z, θ .
l2 m2 p0
Several authors, 11, 15, 19–22, also use this simplified Hamiltonian to study the resonance.
The dynamical system generated from the reduced Hamiltonian, 2.16, is given by

dX, Y, Z, Θ ∂HEr d x, y, z, θ ∂HEr
, − . 2.17
dt ∂ x, y, z, θ dt ∂X, Y, Z, Θ
The equations of motion dX/dt, dY/dt, and dZ/dt defined by 2.17 are
dX & ∞ & l & l

−α mBlmpαm X, Y, Z sin φlmpαm x, y, z, θ , 2.18
dt l2 m2 p0
dY &∞ & l & l

− l − 2p − mα Blmpαm X, Y, Z sin φlmpαm x, y, z, θ , 2.19
dt l2 m2 p0
dZ & ∞ & l & l

l − 2p − m Blmpαm X, Y, Z sin φlmpαm x, y, z, θ . 2.20
dt l2 m2 p0
From 2.18 to 2.20, one can determine the first integral of the system determined by
Er .
the Hamiltonian H
Equation 2.18 can be rewritten as
1 dX &∞ & l & l

− mBlmpαm X, Y, Z sin φlmpαm x, y, z, θ . 2.21
α dt l2 m2 p0
Adding 2.19 and 2.20,
dY dZ &∞ & l & l

α − 1 mBlmpαm X, Y, Z sin φlmpαm x, y, z, θ , 2.22
dt dt l2 m2 p0
and substituting 2.21 and 2.22, one obtains
dY dZ 1 dX
−α − 1 . 2.23
dt dt α dt
Now, 2.23 is rewritten as

1 dX dY dZ
1− 0. 2.24
α dt dt dt
Er has the first

In this way, the canonical system of differential equations governed by H
integral generated from 2.24:

1
1− X Y Z C1 , 2.25
α
where C1 is an integration constant.

Using this first integral, a Mathieu transformation

X, Y, Z, Θ, x, y, z, θ −→ X1 , Y1 , Z1 , Θ1 , x1 , y1 , z1 , θ1 2.26
can be defined.
This transformation is given by the following equations:

1
X1 X, Y1 Y, Z1 1− X Y Z, Θ1 Θ,
α
2.27
1
x1 x − 1 − z, y1 y − z, z1 z, θ1 θ.
α
The subscript 1 denotes the new set of canonical variables. Note that Z1 C1 , and the z1 is an
ignorable variable. So the order of the dynamical system is reduced in one degree of freedom.
Substituting the new set of canonical variables, X1 , Y1 , Z1 , Θ1 , x1 , y1 , z1 , θ1 , in the
reduced Hamiltonian given by 2.16, one gets the resonant Hamiltonian. The word “reso-
nant” is used to denote the Hamiltonian Hrs which is valid for any resonance. The periodic
terms in this Hamiltonian are resonant terms. The Hamiltonian Hrs is given by
μ2 &
∞
Hrs − ωe Θ1 B2j,0,j,0 X1 , Y1 , C1
2X12 j1
2.28
&
∞ &
l &
l

Blmp,αm X1 , Y1 , C1 cos φlmpαm x1 , y1 , θ1 .
l2 m2 p0
The Hamiltonian Hrs has all resonant frequencies, relative to the commensurability α,
where the φlmpαm argument is given by

φlmpαm mαx1 − θ1 l − 2p − αm y1 − φlmpαm0 , 2.29
with
π
φlmpαm0 mλlm − l − m . 2.30
2
The secular and resonant terms are given, respectively, by B2j,0,j,0 X1 , Y1 , C1 and
Blmpαm X1 , Y1 , C1 .
Each one of the frequencies contained in dx1 /dt, dy1 /dt, dθ1 /dt is related, through
the coefficients l, m, to a tesseral harmonic Jlm . By varying the coefficients l, m, p and keeping
q/m fixed, one finds all frequencies dφ1,lmpαm /dt concerning a specific resonance.
From Hrs , taking, j 1, 2, l 2, 4, m 2, α 1/2, and p 0, 1, 2, 3, one gets
2
E1 μ − ωe Θ1 B1,2010 X1 , Y1 , C1 B1,4020 X1 , Y1 , C1
H
2X12

B1,2201 X1 , Y1 , C1 cos x1 − 2θ1 y1 − 2λ22

B1,2211 X1 , Y1 , C1 cos x1 − 2θ1 − y1 − 2λ22
2.31
B1,2221 X1 , Y1 , C1 cos x1 − 2θ1 − 3y1 − 2λ22

B1,4211 X1 , Y1 , C1 cos x1 − 2θ1 y1 − 2λ42 π

B1,4221 X1 , Y1 , C1 cos x1 − 2θ1 − y1 − 2λ42 π

B1,4231 X1 , Y1 , C1 cos x1 − 2θ1 − 3y1 − 2λ42 π .
The Hamiltonian H E1 is defined considering a fixed resonance and three different criti-
cal angles associated to the tesseral harmonic J22 ; the critical angles associated to the tesseral
harmonic J42 have the same frequency of the critical angles associated to the J22 with a dif-
ference in the phase. The other terms in Hrs are considered as short-period terms.
Table 1 shows the resonant coefficients used in the Hamiltonian H E1 .
Finally, a last transformation of variables is done, with the purpose of writing the
resonant angle explicitly. This transformation is defined by
X4 X1 , Y 4 Y1 , Θ4 Θ1 2X1 ,
2.32
x4 x1 − 2θ1 , y4 y1 , θ4 θ1 .
Table 1: Resonant coefficients.

Degree l Order m p q
2 2 0 1
2 2 1 1
2 2 2 1
4 2 1 1
4 2 2 1
4 2 3 1
So, considering 2.31 and 2.32, the Hamiltonian H4 is found to be
μ2
H4 − ωe Θ4 − 2X4 B4,2010 X4 , Y4 , C1 B4,4020 X4 , Y4 , C1
2X42

B4,2201 X4 , Y4 , C1 cos x4 y4 − 2λ22

B4,2211 X4 , Y4 , C1 cos x4 − y4 − 2λ22
2.33
B4,2221 X4 , Y4 , C1 cos x4 − 3y4 − 2λ22

B4,4211 X4 , Y4 , C1 cos x4 y4 − 2λ42 π

B4,4221 X4 , Y4 , C1 cos x4 − y4 − 2λ42 π

B4,4231 X4 , Y4 , C1 cos x4 − 3y4 − 2λ42 π ,
with ωe Θ4 constant and

μ4 3 C1 2X4 2 1 3 −Y4 2 − 2X4 Y4
B4,2010 2
ae J20 − 1 , 2.34
X46 4 X4 Y4 2 4 2 X4 2
⎛ 2 ⎞
μ6 105 C1 2X4 2 3 15 C1 2X4 2 ⎠
B4,4020 10 ae 4 J40 ⎝ 1− −
X4 64 X4 Y4 2 2 8 X4 Y4 2
2.35

−Y4 2 − 2X4 Y4
× 1 5 ,
X4 2

21 4 2 C1 2X4 2
B4,2201 μ a J
e 22 1 −Y4 2 − 2X4 Y4 , 2.36
8X47 X4 Y4
3 3 3 C1 2X4 2
B4,2211 μ4 ae 2 J22 − −Y4 2 − 2X4 Y4 , 2.37
2X47 2 2 X4 Y4 2

3 C1 2X4 2
B4,2221 − μ 4 2
a J
e 22 1 − −Y4 2 − 2X4 Y4 , 2.38
8X47 X4 Y4

9 35 C1 2X4 2
B4,4211 μ6 ae 4 J42 1− C1 2X4
2X411 27 X4 Y4 2

C1 2X4
× 1 X4 Y4 −1 2.39
X4 Y4

15 C1 2X4 2
− 1 −Y4 2 − 2X4 Y4 ,
8 X4 Y4

5 6 4 105 C1 2X4 2 C1 2X4 2
B4,4221 μ ae J42 1− 1−3
2X411 16 X4 Y4 2 X4 Y4 2

2.40
15 15 C1 2X4 2
− −Y4 2 − 2X4 Y4 ,
4 4 X4 Y4 2

μ6 35 C1 2X4 2
B4,4231 4
ae J42 − 1− C1 2X4
X410 27 X4 Y4 2

C1 2X4 −1 15 C1 2X4 2
× 1− X4 Y4 − 1−
X4 Y4 8 X4 Y4 2.41
⎛
⎞
⎜1 −Y 4
2
− 2X Y
4 4 33 −Y4 − 2X4 Y4 ⎟
2
×⎝ ⎠.
2 X4 16 X4 2
Since the term ωe Θ4 is constant, it plays no role in the equations of motion, and a new
Hamiltonian can be introduced,
E4 H4 ωe Θ4 .
H 2.42
E4 is given by
The dynamical system described by H

dX4 , Y4 E4
∂H d x4 , y4 E4
∂H
, − . 2.43
dt ∂ x4 , y4 dt ∂X4 , Y4
Table 2: The zonal and tesseral harmonics.

Zonal harmonics Tesseral harmonics
J20 1.0826 × 10−3 J22 1.8154 × 10−6
J40 −1.6204 × 10−6 J42 1.6765 × 10−7
The zonal harmonics used in 2.34 and 2.35 and the tesseral harmonics used in 2.36
to 2.41 are shown in Table 2.
The constant of integration C1 in 2.34 to 2.41 is given, in terms of the initial values
of the orbital elements, ao , eo , and Io , by

D
C1 μao 1 − eo2 cosIo − 2 2.44
or, in terms of the variables X4 and Y4 ,
C1 X4 cosIo − 2 Y4 cosIo . 2.45
In Section 4, some results of the numerical integration of 2.43 are shown.
3. Lyapunov Exponents
The estimation of the chaoticity of orbits is very important in the studies of dynamical
systems, and possible irregular motions can be analyzed by Lyapunov exponents 23.
In this work, “Gram-Schmidt’s method,” described in 23–26, will be applied to
compute the Lyapunov exponents. A brief description of this method is presented in what
follows.
The dynamical system described by 2.43 can be rewritten as
dX4
P1 X4 , Y4 , x4 , y4 ; C1 ,
dt
dY4
P2 X4 , Y4 , x4 , y4 ; C1 ,
dt 3.1
dx4
P3 X4 , Y4 , x4 , y4 ; C1 ,
dt
dy4
P4 X4 , Y4 , x4 , y4 ; C1 .
dt
Introducing
⎛ ⎞
X4
⎜ ⎟
⎜ Y4 ⎟
⎜ ⎟
z ⎜ ⎟,
⎜ x4 ⎟
⎝ ⎠
y4
⎛ ⎞ 3.2
P1
⎜ ⎟
⎜P2 ⎟
⎜ ⎟
Z ⎜ ⎟.
⎜P3 ⎟
⎝ ⎠
P4
Equations 3.2 can be put in the form
dz
Zz. 3.3
dt
The variational equations, associated to the system of differential equations 3.3, are given
by
dζ
Jζ, 3.4
dt
where J ∂Z/∂z is the Jacobian.

The total number of differential equations used in this method is nn 1, n represents
the number of the motion equations describing the problem, in this case four. In this way,
there are twenty differential equations, four are motion equations of the problem and sixteen
are variational equations described by 3.4.
The dynamical system represented by 3.3 and 3.4 is numerically integrated and the
neighboring trajectories are studied using the Gram-Schmidt orthonormalization to calculate
the Lyapunov exponents.
The method of the Gram-Schmidt orthonormalization can be seen in 25, 26 with more
details. A simplified denomination of the method is described as follows.
Considering the solutions to 3.4 as uκ t, the integration in the time τ begins from
initial conditions uκ t0 eκ t0 , an orthonormal basis.
At the end of the time interval, the volumes of the κ-dimensional κ 1, 2, . . . , N
produced by the vectors uκ are calculated by
C C
CK C
Cκ C
Vκ C
C uj t C,
C 3.5
Cj1 C
L
where is the outer product and · is a norm.
26568
26566
26564
26562
a (km) 26560
26558
26556
26554
0 1000 2000 3000 4000 5000 6000 7000
t (days)
a(0)= 26555 km a(0)= 26563.5 km

a(0)= 26561.7 km a(0)= 26565 km
a(0)= 26562.4 km
Figure 1: Time behavior of the semimajor axis for different values of C1 given in Table 3.
1800
1600
1400
1200
x4 (degrees)
1000
800
600
400
200
0
−200
0 1000 2000 3000 4000 5000 6000 7000
t (days)
a(0)= 26555 km a(0)= 26563.5 km

a(0)= 26561.7 km a(0)= 26565 km
a(0)= 26562.4 km
Figure 2: Time behavior of x4 angle for different values of C1 given in Table 3.
In this way, the vectors uκ are orthonormalized by Gram-Schmidt method. In other

words, new orthonormal vectors eκ t0 τ are calculated, in general, according to
6
uκ − κ−1
j1 uκ · ej ej
eκ C 6 C 3.6
C C.
Cuκ − κ−1
j1 uκ · ej ej C
Table 3: Values of the constant of integration C1 for e 0.001, I 55◦ and different values for semimajor
axis.
a0 × 103 m C1 × 1011 m2 /s
26555.000 −1.467543158
26561.700 −1.467728282
26562.400 −1.467747623
26563.500 −1.467778013
26565.000 −1.467819454
100
50
0
ω (degrees)
−50
−100
−150
−200
0 1000 2000 3000 4000 5000 6000 7000
t (days)
a(0)= 26555 km a(0)= 26563.5 km

a(0)= 26561.7 km a(0)= 26565 km
a(0)= 26562.4 km
Figure 3: Time behavior of the argument of pericentre for different values of C1 given in Table 3.
The Gram-Schmidt method makes invariant the κ-dimensional subspace produced by

the vectors u1 , u2 , u3 , . . . , uκ in constructing the new κ-dimensional subspace spanned by the
vectors e1 , e2 , e3 , . . . , eκ .
With new vector uκ t0 τ eκ t0 τ, the integration is reinitialized and carried
forward to t t0 2τ. The whole cycle is repeated over a long-time interval. The theorems
guarantee that the κ-dimensional Lyapunov exponents are calculated by 25, 26:

1 &n ln Vκ t0 jτ
λκ lim . 3.7
j1 ln Vκ t0 j − 1 τ
n → ∞ nτ
The theory states that if the Lyapunov exponent tends to a positive value, the orbit is
chaotic.
In the next section are shown some results about the Lyapunov exponents.
0.025
0.02
0.015
e
0.01
0.005
0
0 1000 2000 3000 4000 5000 6000 7000
t (days)
a(0)= 26555 km a(0)= 26563.5 km

a(0)= 26561.7 km a(0)= 26565 km
a(0)= 26562.4 km
Figure 4: Time behavior of the eccentricity for different values of C1 given in Table 3.
26580
26575
26570
26565
a (km)
26560
26555
26550
26545
26540
0 1000 2000 3000 4000 5000 6000 7000
t (days)
J22 and J42

J22
4. Results
Figures 1, 2, 3, and 4 show the time behavior of the semimajor axis, x4 angle, argument of
perigee and of the eccentricity, according to the numerical integration of the motion equa-
tions, 2.43, considering three different resonant angles together: φ2201 , φ2211 , and φ2221 associ-
ated to J22 , and three angles, φ4211 , φ4221 , and φ4231 associated to J42 , with the same frequency
of the resonant angles related to the J22 , but with different phase. The initial conditions corre-
sponding to variables X4 and Y4 are defined for eo 0.001, Io 55◦ , and ao given in Table 3.
1000
500
−500
x4 (degrees)
−1000
−1500
−2000
−2500
−3000
0 1000 2000 3000 4000 5000 6000 7000
t (days)
J22 and J42

J22
100
−100
−200
ω (degrees)
−300
−400
−500
−600
−700
0 1000 2000 3000 4000 5000 6000 7000
t (days)
J22 and J42

J22
Table 4: Values of the constant of integration C1 for e 0.05, I 10◦ , and different values for semimajor
axis.
a0 × 103 m C1 × 1011 m2 /s
26555.000 −1.045724331
26565.000 −1.045921210
26568.000 −1.045980267
26574.000 −1.046098370
The initial conditions of the variables x4 and y4 are 0◦ and 0◦ , respectively. Table 3 shows the
values of C1 corresponding to the given initial conditions.
0.058
0.056
0.054
0.052
e
0.05
0.048
0.046
0.044
0 1000 2000 3000 4000 5000 6000 7000
t (days)
J22 and J42

J22
26570
26568
26566
26564
a (km)
26562
26560
26558
26556
26554
26552
0 1000 2000 3000 4000 5000 6000 7000
t (days)
a(0)= 26555 km a(0)= 26564 km

a(0)= 26558 km a(0)= 26568 km
a(0)= 26562 km
Figures 5, 6, 7, and 8 show the time behavior of the semimajor axis, x4 angle, argument
of perigee and of the eccentricity for two different cases. The first case considers the critical
angles φ2201 , φ2211 , and φ2221 , associated to the tesseral harmonic J22 , and the second case
considers the critical angles associated to the tesseral harmonics J22 and J42 . The angles associ-
ated to the J42 , φ4211 , φ4221 , and φ4231 , have the same frequency of the critical angles associated
to the J22 with a different phase. The initial conditions corresponding to variables X4 and Y4
are defined for eo 0.05, Io 10◦ , and ao given in Table 4. The initial conditions of the varia-
bles x4 and y4 are 0◦ and 60◦ , respectively. Table 4 shows the values of C1 corresponding to
the given initial conditions.
2000
1500
1000
x4 (degrees)
500
−500
−1000
0 1000 2000 3000 4000 5000 6000 7000
t (days)
a(0)= 26555 km a(0)= 26564 km

a(0)= 26558 km a(0)= 26568 km
a(0)= 26562 km
80
60
40
20
0
ω (degrees)
−20
−40
−60
−80
−100
−120
−140
0 1000 2000 3000 4000 5000 6000 7000
t (days)
a(0)= 26555 km a(0)= 26564 km

a(0)= 26558 km a(0)= 26568 km
a(0)= 26562 km
Analyzing Figures 5–8, one can observe a correction in the orbits when the terms re-
lated to the tesseral harmonic J42 are added to the model. Observing, by the percentage, the
contribution of the amplitudes of the terms B4,4211 , B4,4221 , and B4,4231 , in each critical angle
studied, is about 1,66% up to 4,94%. In fact, in the studies of the perturbations in the arti-
ficial satellites motion, the accuracy is important, since adding different tesseral and zonal
harmonics to the model, one can have a better description about the orbital motion.
Figures 9, 10, 11, and 12 show the time behavior of the semimajor axis, x4 angle, argu-
ment of perigee and of the eccentricity, according to the numerical integration of the motion
equations, 2.43, considering three different resonant angles together; φ2201 , φ2211 , and φ2221
0.025
0.02
0.015
e
0.01
0.005
0
0 1000 2000 3000 4000 5000 6000 7000
t (days)
a(0)= 26555 km a(0)= 26564 km

a(0)= 26558 km a(0)= 26568 km
a(0)= 26562 km
0.1
0.01
Lyapunov exponents λ (κ)
0.001
0.0001
1e−005
1e−006
1e−007
10 100 1000 10000 100000
Time (days)
λ (1)(a(0)= 26563.5 km) λ (2)(a(0)= 26563.5 km)

λ (1)(a(0)= 26565 km) λ (2)(a(0)= 26565 km)
Figure 13: Lyapunov exponents λ1 and λ2, corresponding to the variables X4 and Y4 , respectively, for
C1 −1.467778013 × 1011 m2 /s and C1 −1.467819454 × 1011 m2 /s, x4 0◦ and y4 0◦ .
associated to J22 and three angles φ4211 , φ4221 , and φ4231 associated to J42 . The initial conditions
corresponding to variables X4 and Y4 are defined for eo 0.01, Io 55◦ , and ao given in Table
5. The initial conditions of the variables x4 and y4 are 0◦ and 60◦ , respectively. Table 5 shows
the values of C1 corresponding to the given initial conditions.
Analyzing Figures 1–12, one can observe possible irregular motions in Figures 1–4,
specifically considering values for C1 −1.467778013 × 1011 m2 /s and C1 −1.467819454 ×
1011 m2 /s, and, in Figures 9–12, for C1 −1.467765786 × 1011 m2 /s and C1 −1.467821043 ×
1011 m2 /s. These curves will be analyzed by the Lyapunov exponents in a specified time
verifying the possible regular or chaotic motions.
0.1
0.01

0.001
0.0001
1e−005
1e−006
1e−007
10 100 1000 10000 100000
Time (days)
λ (1)(a(0)= 26562 km) λ (2)(a(0)= 26562 km)

λ (1)(a(0)= 26564 km) λ (2)(a(0)= 26564 km)
C1 −1.467765786 × 1011 m2 /s and C1 −1.467821043 × 1011 m2 /s, x4 0◦ and y4 60◦ .
Table 5: Values of the constant of integration C1 for e 0.01, I 55◦ , and different values for semimajor
axis.
a0 × 103 m C1 × 1011 m2 /s
26555.000 −1.467572370
26558.000 −1.467655265
26562.000 −1.467765786
26564.000 −1.467821043
26568.000 −1.467931552
Figures 13 and 14 show the time behavior of the Lyapunov exponents for two different
cases, according to the initial values of Figures 1–4 and 9–12. The dynamical system in-
volves the zonal harmonics J20 and J40 and the tesseral harmonics J22 and J42 . The method
used in this work for the study of the Lyapunov exponents is described in Section 3. In
Figure 13, the initial values for C1 , x4 , and y4 are C1 −1.467778013 × 1011 m2 /s and C1
−1.467819454 × 1011 m2 /s, x4 0◦ and y4 0◦ , respectively. In Figure 14, the initial values for
C1 , x4 , and y4 are C1 −1.467765786 × 1011 m2 /s and C1 −1.467821043 × 1011 m2 /s, x4 0◦
and y4 60◦ , respectively. In each case are used two different values for semimajor axis
corresponding to neighboring orbits shown previously in Figures 1–4 and 9–12.
Figures 13 and 14 show Lyapunov exponents for neighboring orbits. The time used in
the calculations of the Lyapunov exponents is about 150.000 days. For this time, it can be ob-
served in Figure 13 that λ1, corresponding to the initial value a0 26565.0 km, tends to a
positive value, evidencing a chaotic region. On the other hand, analyzing the same Figure 13,
λ1, corresponding to the initial value a0 26563.5 km, does not show a stabilization
around the some positive value, in this specified time. Probably, the time is not sufficient for
a stabilization in some positive value, or λ1, initial value a0 26563.5 km, tends to a nega-
tive value, evidencing a regular orbit. Analyzing now Figure 14, it can be verified that λ1,
corresponding to the initial value a0 26564.0 km, tends to a positive value, it contrasts
0.0001

1e−005
1e−006
1e−007
10000 100000
Time (days)
λ (1)(a(0)= 26563.5 km) λ (2)(a(0)= 26563.5 km)

λ (1)(a(0)= 26565 km) λ (2)(a(0)= 26565 km)
C1 −1.467778013 × 1011 m2 /s and C1 −1.467819454 × 1011 m2 /s, x4 0◦ and y4 0◦ .
0.0001
1e−005
1e−006
1e−007
10000 100000
Time (days)
λ (1)(a(0)= 26562 km) λ (2)(a(0)= 26562 km)

λ (1)(a(0)= 26564 km) λ (2)(a(0)= 26564 km)
C1 −1.467765786 × 1011 m2 /s and C1 −1.467821043 × 1011 m2 /s, x4 0◦ and y4 60◦ .
with λ1, initial value a0 26562.0 km. Comparing Figure 13 with Figure 14, it is observed
that the Lyapunov exponents in Figure 14 has an amplitude of oscillation greater than the Ly-
apunov exponents in Figure 13. Analyzing this fact, it is probable that the necessary time for
the Lyapunov exponent λ2, in Figure 14, to stabilize in some positive value is greater than
the necessary time for the λ2 in Figure 13.
Rescheduling the axes of Figures 13 and 14, as described in Figures 15 and 16, respec-
tively, the Lyapunov exponents tending to a positive value can be better visualized.
5. Conclusions
In this work, the dynamical behavior of three critical angles associated to the 2 : 1 resonance
problem in the artificial satellites motion has been investigated.
The results show the time behavior of the semimajor axis, argument of perigee and e-
ccentricity. In the numerical integration, different cases are studied, using three critical angles
together: φ2201 , φ2211 , and φ2221 associated to J22 and φ4211 , φ4221 , and φ4231 associated to the
J42 .
In the simulations considered in the work, four cases show possible irregular motions
for C1 −1.467778013 × 1011 m2 /s, C1 −1.467819454 × 1011 m2 /s, C1 −1.467765786 ×
1011 m2 /s, and C1 −1.467821043 × 1011 m2 /s. Studying the Lyapunov exponents, two cases
show chaotic motions for C1 −1.467819454 × 1011 m2 /s and C1 −1.467821043 × 1011 m2 /s.
Analyzing the contribution of the terms related to the J42 , it is observed that, for the
value of C1 −1.045724331 × 1011 m2 /s, the amplitudes of the terms B4,4211 , B4,4221 , and B4,4231
are greater than the other values of C1 . In other words, for bigger values of semimajor axis, it
is observed a smaller contribution of the terms related to the tesseral harmonic J42 .
The theory used in this paper for the 2 : 1 resonance can be applied for any resonance
involving some artificial Earth satellite.
Acknowledgments
This work was accomplished with support of the FAPESP under the Contract no. 2009/00735-
5 and 2006/04997-6, SP Brazil, and CNPQ Contracts 300952/2008-2 and 302949/2009-7.
References
1 M. B. Morando, “Orbites de resonance des satellites de 24h,” Bulletin of the American Astronomical
Society, vol. 24, pp. 47–67, 1963.
2 L. Blitzer, “Synchronous and resonant satellite orbits associated with equatorial ellipticity,” Journal of
Advanced Robotic Systems, vol. 32, pp. 1016–1019, 1963.
3 B. Garfinkel, “The disturbing function for an artificial satellite,” The Astronomical Journal, vol. 70, no.
9, pp. 688–704, 1965.
4 B. Garfinkel, “Tesseral harmonic perturbations of an artificial satellite,” The Astronomical Journal, vol.
70, pp. 784–786, 1965.
5 B. Garfinkel, “Formal solution in the problem of small divisors,” The Astronomical Journal, vol. 71, pp.
657–669, 1966.
6 G. S. Gedeon and O. L. Dial, “Along-track oscillations of a satellite due to tesseral harmonics,” AIAA
Journal, vol. 5, pp. 593–595, 1967.
7 G. S. Gedeon, B. C. Douglas, and M. T. Palmiter, “Resonance effects on eccentric satellite orbits,”
Journal of the Astronautical Sciences, vol. 14, pp. 147–157, 1967.
8 G. S. Gedeon, “Tesseral resonance effects on satellite orbits,” Celestial Mechanics, vol. 1, pp. 167–189,
1969.
9 M. T. Lane, “An analytical treatment of resonance effects on satellite orbits,” Celestial Mechanics, vol.
42, pp. 3–38, 1988.
10 A. Jupp, “A solution of the ideal resonance problem for the case of libration,” The Astronomical Journal,
vol. 74, pp. 35–43, 1969.
11 T. A. Ely and K. C. Howell, “Long-term evolution of artificial satellite orbits due to resonant tesseral
harmonics,” Journal of the Astronautical Sciences, vol. 44, pp. 167–190, 1996.
12 D. M. Sanckez, T. Yokoyama, P. I. O. Brasil, and R. R. Cordeiro, “Some initial conditions for disposed
satellites of the systems GPS and galileo constellations,” Mathematical Problems in Engineering, vol.
2009, Article ID 510759, 22 pages, 2009.
13 L. D. D. Ferreira and R. Vilhena de Moraes, “GPS satellites orbits: resonance,” Mathematical Problems
in Engineering, vol. 2009, Article ID 347835, 12 pages, 2009.
14 J. C. Sampaio, R. Vilhena de Moraes, and S. S. Fernandes, “Artificial satellites dynamics: resonant ef-
fects,” in Proceedings of the 22nd International Symposium on Space Flight Dynamics, São José dos Cam-
pos, Brazil, 2011.
15 A. G. S. Neto, Estudo de Órbitas Ressonantes no Movimento de Satélites Artificiais, Tese de Mestrado, ITA,
2006.
16 J. P. Osorio, Perturbações de Órbitas de Satélites no Estudo do Campo Gravitacional Terrestre, Imprensa
Portuguesa, Porto, Portugal, 1973.
17 W. M. Kaula, Theory of Satellite Geodesy: Applications of Satellites to Geodesy, Blaisdel, Waltham, Mass,
USA, 1966.
18 A. E. Roy, Orbital Motion, Institute of Physics Publishing Bristol and Philadelphia, 3rd edition, 1988.
19 P. H. C. N. Lima Jr., Sistemas Ressonantes a Altas Excentricidades no Movimento de Satélites Artificiais, Tese
de Doutorado, Instituto Tecnológico de Aeronáutica, 1998.
20 P. R. Grosso, Movimento Orbital de um Satélite Artificial em Ressonância 2:1, Tese de Mestrado, Instituto
Tecnológico de Aeronáutica, 1989.
21 J. K. S. Formiga and R. Vilhena de Moraes, “Dynamical systems: an integrable kernel for resonance
effects,” Journal of Computational Interdisciplinary Sciences, vol. 1, no. 2, pp. 89–94, 2009.
22 R. Vilhena de Moraes, K. T. Fitzgibbon, and M. Konemba, “Influence of the 2:1 resonance in the orbits
of GPS satellites,” Advances in Space Research, vol. 16, no. 12, pp. 37–40, 1995.
23 F. Christiansen and H. H. Rugh, “Computing lyapunov spectra with continuous gram-schmidt
orthonormalization,” Nonlinearity, vol. 10, no. 5, pp. 1063–1072, 1997.
24 L. Qun-Hong and T. Jie-Yan, “Lyapunov exponent calculation of a two-degree-of-freedom vibro-
impact system with symmetrical rigid stops,” Chinese Physics B, vol. 20, no. 4, Article ID 040505, 2011.
25 I. Shimada and T. Nagashima, “A numerical approach to ergodic problem of dissipative dynamical
systems,” Progress of Theoretical Physics, vol. 61, no. 6, pp. 1605–1616, 1979.
26 W. E. Wiesel, “Continuous time algorithm for Lyapunov exponents,” Physical Review E, vol. 47, no. 5,
pp. 3686–3697, 1993.
doi:10.1155/2012/351759
Research Article
Application of Periapse Maps for
the Design of Trajectories Near the Smaller
Primary in Multi-Body Regimes
Kathleen C. Howell, Diane C. Davis, and Amanda F. Haapala

School of Aeronautics and Astronautics, Purdue University, Armstrong Hall, West Lafayette,
IN 47907, USA
Correspondence should be addressed to Kathleen C. Howell, howell@purdue.edu
Copyright q 2012 Kathleen C. Howell et al. This is an open access article distributed under
The success of the Genesis spacecraft, as well as the current Artemis mission, continue to generate
interest in expanding the trajectory options for future science and exploration goals throughout the
solar system. Incorporating multi-body dynamics into the preliminary design can potentially offer
flexibility and influence the maneuver costs to achieve certain objectives. In the current analysis,
attention is focused on the development and application of design tools to facilitate preliminary
trajectory design in a multi-body environment. Within the context of the circular restricted three-
body problem, the evolution of a trajectory in the vicinity of the smaller primary is investigated.
Introduced previously, periapse Poincaré maps have emerged as a valuable resource to predict
both short- and long-term trajectory behaviors. By characterizing the trajectories in terms of radius
and periapse orientation relative to the P1 -P2 line, useful trajectories with a particular set of desired
characteristics can be identified and computed.
1. Introduction
Spacecraft exploration activities increasingly involve trajectories that reach the vicinity of
the libration points in various types of three-body systems. Libration point trajectories and
low-energy transfers, in particular, have garnered much recent attention because of their
potential to incorporate the natural dynamics, to generate new types of design options and,
for some spacecraft applications, reduce propellant. A number of successful missions within
the last decade have successfully exploited the natural dynamics of multiple gravity fields,
for example, NASA’s Genesis mission, launched in 2001 with a return to Earth in 2004
1, 2. In a Sun-Earth rotating view in Figure 1, the trajectory for the Genesis spacecraft
leveraged the gravity of the Earth, Sun, and the Moon to supply a gravitational balance and
Solar wind collection

1
in halo orbit about L1
(29.3 mos.)
Return and
0.5
recovery
Millions of kilometers
Lunar orbit
Positioning for
(5.3 mos.)
Moon daylight reentry
L1 Earth
Sun
Halo Outward leg

orbits (2.7 mos.)
−0.5 (5)
Total flight time
(37.3 mos.)
−1
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
Millions of kilometers
Figure 1: Genesis trajectory as viewed in the Sun-Earth rotating frame. Samples of solar material were
collected on the spacecraft over two years in an L1 libration point orbit and returned to Earth. During the
return, the lunar gravity was also leveraged as the spacecraft shifted to the L2 region prior to Earth reentry.
CoutesyNASA/JPL-Caltech, http://genesismission.jpl.nasa.gov/gm2/mission/halo.htm.
deliver a trajectory that met the goals and satisfied the constraints with a path that does
not emerge within the context of a two-body problem. In another example from an ongoing
mission, Artemis involves two identical spacecraft identified as P1 and P2, originally two
of the five Themis spacecraft 3–5. Employing the Sun-Earth-Moon multi-body dynamical
environment, these two vehicles were directed from the outermost of five elliptical Earth
orbits to eventually arrive in Earth-Moon libration point orbits on August 25, 2010, and
October 10, 2010. The transfer phase appears in Figure 2a as viewed in the Sun-Earth
rotating fame; both spacecraft are to be inserted into elliptical lunar orbits see Figure 2b
on June 27 and July 17, 2011, respectively.
Although much has been learned about the design space for such missions in the
last few decades, as is evident from Genesis and Artemis as well as a number of other
libration point missions, trajectory design in this type of regime remains a nontrivial problem.
Typical challenges in the use of non-Keplerian orbits in a multi-body environment include i
complexity: there are many destinations and competing goals; ii the search for solutions
in new dynamical environments with frequent attempts to blend arcs from various models
with different levels of fidelity; iii new types of scenarios that are explored to offer options
for extended missions and contingencies. To exploit the gravity of multiple bodies requires a
capability to deliver the trajectory characteristics that meet the requirements for a particular
mission. Without analytical solutions, increasing insight into the dynamical structure in the
three-body problem has been developed, beginning with Hénon and resulting in a wide
variety of investigations, frequently with a focus toward applications 6–27. In many of
these analyses, the invariant manifolds associated with the L1 and L2 Lyapunov orbits have
been increasingly used to predict the behavior of trajectories that originate near the smaller
primary in the circular restricted three-body problem. In addition, the use of Poincaré maps to
identify trajectories with various short- and long-term behaviors is effective. As noted in some
of these recent publications, however, a major barrier to the development of a simple orbit
Libration orbit (1-rev)

Flyby number 2
(Feb 13, 2010) Moon Flyby number 1
(Jan 31, 2010)
Sun-Earth line Earth
Transfer trajectory Postraised

elliptical orbit
Libration orbit (Aug 2010)

Earth BBR axes
25 Jun 2011 10:33:36 Time step: 57600 s
a
L2 to L1 transfer orbit
Direction Moon
to Earth Artemis-P1
here on
L1 August 25th
L2
Direction of
motion
Artemis L1 orbit
Artemis L2 orbit
Moon’s orbit
Moon inertial axes
22 Aug 2010 07:10:00 Time step: 600 s
b
Figure 2: a Artemis trajectory for one spacecraft during transfer from Earth to lunar vicinity viewed in
the Sun-Earth rating frame. The trajectory reaches the vicinity of the Sun-Earth L1 libration point after two
relatively high energy lunar flybys a trajectory “backflip” to eventually reach a low-energy trajectory
in the vicinity of the Moon. http://www.nasa.gov/mission pages/artemis/news/lunar-orbit.html. b
Artemis trajectory at arrival in lunar vicinity viewed in Earth-Moon rotating frame. Both spacecraft employ
Earth-Moon L1 and L2 libration point orbits to modify the path and eventually insert into lunar orbit later
this year. http://www.nasa.gov/mission pages/artemis/news/lunar-orbit.html.
design process is the overwhelming nature of the design space. The trajectory design process
remains challenging due to the varied, and sometimes chaotic, nature of trajectories that are
simultaneously influenced by two gravitational bodies. To effectively select a trajectory to
satisfy a given requirement, it is necessary to simplify and organize the design space as
much as possible. The invariant manifold structure associated with the collinear libration
points, in particular, has offered a geometrical framework for this dynamical environment.
Yet, harnessing this information to supply relatively quick and efficient options for trajectory
designers is a formidable challenge. Thus, the design difficulties remain significant, but a
wider range of tools is emerging.
The motivation for this investigation originated from a general need to repeatedly
develop design concepts for potential applications. A major challenge involved in orbit
design within the context of the circular restricted three-body problem CR3BP is the
organization of the vast set of options that is available within the design space; it is difficult
to locate the specific initial conditions that lead to a trajectory with a particular set of
characteristics. The invariant manifolds and the corresponding phase space yield a rich
dynamical structure, and one method for visualizing the space involves the use of Poincaré
maps, which reduce the dimensionality of the problem. Such maps are successfully employed
in a number of analyses including Koon et al. 11, 12, Gómez et al. 13, 17, Howell et al.
14, Topputo et al. 18, and Anderson and Lo 27. However, an alternative representation
is sought to capture this knowledge and further facilitate trajectory design. A different
parameterization of a Poincaré map involves the surface of section at the plane corresponding
to periapsis. Its advantage is based on the fact that the map is viewed within position
space. This type of map is denoted by a periapse Poincaré map and was first defined and
introduced by Villac and Scheeres 28 to relate a trajectory escaping the vicinity of P2 back
to its previous periapsis in the planar Hill problem. Paskowitz and Scheeres 29, 30 extend
this analysis, using periapse Poincaré maps to define lobes corresponding to the first four
periapse passages after capture into an orbit about P2 . For application to the Europa orbiter
problem, the authors define “safe zones” where a spacecraft is predicted to neither escape
nor impact the surface of the satellite for a specified period of time. Davis and Howell
31, 32 build on the analysis for short- and long-term trajectories to illustrate some of the
structures associated with manifold tubes corresponding to the L1 and L2 Lyapunov orbits
as viewed in terms of periapse maps. Long-term periapsis Poincaré maps aid in organizing
large numbers of trajectories and can deliver initial conditions that yield trajectories with
specific characteristics. Davis and Howell 32 as well as Haapala and Howell 33 employ
such maps to compute specific types of trajectories in the region near P2 . For trajectory design
in this regime, good initial guesses are critical and a technique that supplies geometrical
insight concerning the structure and delivers good approximate solutions is a practical
alternative to construct trajectory options. Poincaré maps generally do require large amounts
of computation, but such capabilities are expanding very quickly. In addition, as more
design is accomplished within an interactive visual environment, techniques that are easily
adaptable to visual interfaces are appealing and likely to be incorporated into the next
generation of design tools. Ultimately, any of the design approaches that successfully leverage
computational speed and visual interfaces possess advantages.
This analysis is focused on exploring periapse maps to construct trajectory options
in multi-body regimes. The CR3BP frequently serves as a basis for the preliminary analysis
in such problems, and, thus, for this investigation, the primary focus is the region near the
smaller primary, P2 . The goal is a strategy that facilitates preliminary trajectory design in
the CR3BP but still embraces the invariant manifold framework. Periapse Poincaré maps are
defined and the structure that is apparent in such types of maps is summarized. The map
appears in position space, and different parameters at periapsis are represented depending
on the application. Examples serve to demonstrate how the maps can be exploited to deliver
different types of solutions. Periapse maps can also be used in conjunction with other types of
analysis tools. Such an approach has proved useful to isolate specific arcs and, in some cases,
blend them with other arcs for design.
2. Background
2.1. Dynamical Model
The dynamical model that is assumed for this analysis is consistent with the formulation
in the circular restricted three-body problem CR3BP, where the motion of a particle of
infinitesimal mass, P3 , is examined as it moves in the vicinity of two larger primary bodies, P1
and P2 . A rotating frame, centered at the system barycenter, B, is defined such that the rotating
9
x-axis is directed from the larger primary P1 to the smaller P2 , the z9-axis is parallel to the
direction of the angular velocity of the primary system with respect to the inertial frame, and
9
the y-axis completes the dextral orthonormal triad. Let the nondimensional mass μ be the
ratio of the small mass of P2 to the total system mass. The system then admits five equilibrium
solutions comprised of the three collinear points L1 , L2 , and L3 and two equilateral points
L4 and L5 as depicted in Figure 3a. Note that the magnitude of the Hill radius 8 is
1/3
μ
rH , 2.1
3
and is nearly equal to the distance between P2 and the Lagrange points L1 and L2 . A single
integral of motion exists in the CR3BP. Known as the Jacobi integral, it is evaluated as

2μ 2 1 − μ
C x y
2 2
− v2 , 2.2
r23 r13
where v is the magnitude of the spacecraft velocity relative to the rotating frame. The scalar
nondimensional distances r13 and r23 reflect the distance to P3 from the primaries P1 and
P2 , respectively. Then, the Jacobi constant restricts the motion of the spacecraft to regions in
space, where v2 ≥ 0; these regions are bounded by surfaces of zero velocity. In the planar
problem, the surfaces reduce to the zero velocity curves ZVCs. For values of the Jacobi
integral higher than that associated with the L1 libration point, the ZVCs form closed regions
around the two primaries. As the energy of the spacecraft is increased, the Jacobi value
decreases until, at the L1 value of the Jacobi integral, that is, CL1 , the ZVCs open at the L1
libration point, exposing a gateway. The particle P3 is now free to move between the two large
primaries. Similarly, when the value of the Jacobi integral decreases to the value associated
with L2 , CL2 , the ZVCs open at L2 and the particle, that is, spacecraft, may escape entirely from
the vicinity of the primaries. For values of such that CL3 < C < CL2 , the ZVCs appear as in
Figure 3b; the gray area cannot be reached by P3 at this Jacobi level and is thus “forbidden.”
L4
L3 B L1 L2
x
P1 P2
L5
a Schematic of libration points
1.5
Exterior region
1 Forbidden region
0.5
y (dimensionless)
P1 P2
0 ∗ ∗
Interior region P2 region

−0.5
−1
−1.5
−1.5 −1 −0.5 0 0.5 1 1.5
x (dimensionless)
b Sample ZVC
Figure 3: Regions of position space delineated by the zero velocity curves.
The exterior and interior regions in the figure are denoted to be consistent with Koon et al.
11.
2.2. Invariant Manifolds

For unstable periodic orbits in the CR3BP, in particular the periodic Lyapunov orbits in the
planar problem, higher-dimensional manifold structures exist and supply a framework for
this region via the stable and unstable manifolds. For L1 and L2 Lyapunov orbits to exist at a
given level of Jacobi constant, both the L1 and L2 gateways are open. Let to be the initial time
and the symbol T identify the time for one period. Assume that λs < 1 and λu 1/λ2 are the
stable and unstable eigenvalues from the monodromy matrix, Φto T, to , as determined for
0.2
0.1
y (dimensionless)
0
−0.1
−0.2
0.8 1 1.2
x (dimensionless)
Figure 4: Stable blue and unstable red manifolds associated with periodic L1 and L2 Lyapunov orbits
for C 3.1672 in the Earth-Moon three-body problem.
a given Lyapunov orbit. Let ws and wu be the associated eigenvectors, computed by solving
the equation Φto T, to ws λs ws , Φto T, to wu λu wu . Define w , w− as the
U− S−
two directions associated with an eigenvector. The local half-manifolds, Wloc and Wloc , are
∗
approximated by introducing a perturbation relative to a fixed point, x , along the periodic
orbit in the direction w−u and w−s , respectively. Likewise, a perturbation relative to x∗ in the
direction w u and w s , respectively, produces the local half-manifolds Wloc U S
and Wloc . The
step along the direction of the eigenvector is denoted d and the initial states along the local
stable and unstable manifolds are evaluated as x±s x∗ ± d · w±s . The local stable manifolds
are globalized by propagating the states x s and x−s in reverse time in the nonlinear model.
S−
This process yields the numerical approximation for the global manifolds WxS ∗ and W ∗ ,
x
respectively. Propagating the state x±u x∗ ± d · w±u in forward-time yields the unstable global
manifolds WxU ∗ and WxU−∗ . The collection of the stable and unstable manifolds corresponding
to each fixed point along sample L1 and L2 Lyapunov orbits in the Earth-Moon system appear
in Figure 4 in configuration space.
3. Periapse Poincaré Maps

3.1. Creation of Periapse Maps
Connecting arcs in the CR3BP by exploiting the invariant manifold structures and the use of
Poincaré maps has been successfully demonstrated by Koon et al., Gómez et al., and others
11–14, 16, 17. A Poincaré map is commonly used to interpret the behavior of groups of
trajectories, relating the states at one point in time to a future state forward along the path. By
fixing the value of the Jacobi integral and selecting a surface of section, the dimensionality of
the problem is reduced by two; the four-dimensional planar problem is thus reduced to two
dimensions. Poincaré maps, with various parameterizations, have proven to be a useful tool
for trajectory analysis and design. An alternative parameterization that facilitates exploration
of the design space and selection of certain types of characteristics is the periapse Poincaré
map, first defined and applied by Villac and Scheeres 28. In this type of map, the surface of
section is the plane of periapse passage. Villac and Scheeres, 28 as well as Paskowitz and
Scheeres, 29 employ periapse Poincaré maps to explore the short-term behavior of escaping
trajectories as well as captured trajectories within the context of the Hill three-body problem
with applications to the Jupiter-Europa system. Building on these results, the short- and long-
term behavior of trajectories in the CR3BP as viewed in terms of periapse maps is explored
by Davis and Howell 31, 32 as well as Haapala and Howell 33. Periapsis and apoapsis
relative to P2 are defined such that the radial velocity, ṙ, of P3 relative to P2 is zero and are
distinguishable by the direction of radial acceleration, r̈ ≥ 0 at periapsis and r̈ ≤ 0 at apoapsis
34.
The periapse maps are relatively simple to create. Each point within the periapse
region corresponds to an initial condition that is associated with a specific trajectory about
P2 at a specified level of Jacobi constant. The initial condition always reflects a periapsis.
Given an initial position, velocity can be selected to produce a prograde or retrograde path;
all initial velocities are assumed prograde in this analysis. The initial state corresponding
to each trajectory is then propagated forward in time for a specified number of revolutions
to generate a series of subsequent periapse points. After the first revolution, the state is
evaluated against four possible outcomes: the particle impacts P2 , the particle escapes out
the L1 gateway; the particle escapes through the L2 gateway, or the particle remains captured
near P2 , that is, it continues to evolve within the ZVCs. Impact trajectories are defined as
those possessing a position vector, at any time, that passes on or within the radius of the body
P2 . Escape trajectories are identified by an x-coordinate lying more than 0.01 nondimensional
units beyond either L1 or L2 . Finally, the initial periapse position is colored consistent with the
outcome. Any states that continue to evolve are evaluated after the return to the next periapse
condition and the process continues until a predetermined time or number of revolutions.
Thus, maps are created to isolate certain types of behavior. Maps are produced in the Sun-
Saturn system for both short- and long-term propagations in Davis and Howell 32 and
Haapala and Howell 33. Once a region is isolated, relationships between other periapsis
parameters are also exploited 32, 33.
3.2. Defining Regions in the Maps
As an example, the periapse structures in the Sun-Saturn system, associated with the
invariant manifolds corresponding to the planar Lyapunov orbits and a specified Jacobi
constant value, appear in Figure 5. Note that the Sun and Saturn are simply a representative
system. The manifolds in Figure 5a are propagated through their first periapses which
are indicated as blue points along the manifold trajectories. Observe that the periapses
along the manifold tubes define well-defined lobes that identify the escaping trajectories.
These lobes are analogous to the lobes defined for the Hill restricted three-body problem in
Villac and Scheeres 28 and Paskowitz and Scheeres 29. Consistent with the nature of the
stable/unstable manifolds that are associated with the Lyapunov orbits as separatrices for the
flow, these lobes represent regions in which a periapsis occurs just prior to direct escape from
the vicinity of Saturn; any trajectory with a periapsis within one of these lobes escapes prior
to reaching its next periapsis. Conversely, a trajectory with periapsis lying outside a lobe
does not escape before its next periapse passage. These lobes can, therefore, be considered
gateways to escape: all escaping trajectories pass through one of these regions at the final
periapse passage prior to escape. Note that, for some trajectories, the first periapsis actually
occurs in the vicinity of the libration point orbit near L1 or L2 , however, these periapses are
neglected in this investigation. The boundaries of the lobes, that is, the periapses along the
manifolds, appear as contours in Figure 5a. To isolate the structures in the figure, a naming
0.04 0.04
0.03 0.03
S
y (dimensionless)
0.02
y (dimensionless)
0.02 L1 ,2
0.01 r̈ = 0
0.01 S S
WLS+ L1 ,1 S L1 ,1
0 ∗ 1
∗ 0 ∗ L1 ,3 ∗
S
WLS− S
−0.01
L2 ,1 2 −0.01 L2 ,1
S S
−0.02 −0.02 L2 ,2 L2 ,3
−0.03 −0.03
−0.04 −0.04
0.94 0.96 0.98 1 1.02 1.04 1.06 1.06 0.94 0.96 0.98 1 1.02 1.04 1.06
x (dimensionless) x (dimensionless)
t+ t+
L1 ,1 L2 ,1
t+ t+
L1 ,2 L2 ,2
t+ t+
L1 ,3 L2 ,3
a Map of first periapses along WL1S and WL2S- for b First three periapses along WLS
1
, WLS−
2
, and along L1
C3.0173 blue dots forward-time escapes, and L2 forward-time escapes for C
3.0174
0.04 0.04
0.03 0.03
y (dimensionless)
U
y (dimensionless)
0.02 L2 ,2 U
0.02
L2 ,3
0.01 U 0.01
L2 ,1
0 ∗ U
L1 ,3 ∗ 0 ∗ ∗
U
L1 ,1
−0.01 −0.01
−0.02 U
r̈ = 0 −0.02
L1 ,2
−0.03 −0.03
−0.04 −0.04
0.94 0.96 0.98 1 1.02 1.04 1.06 0.94 0.96 0.98 1 1.02 1.04 1.06
t− t−
L1 ,1 L2 ,1
t− t−
L1 ,2 L2 ,2
t− t−
L1 ,3 L2 ,3
c First three periapses along WLU

1
, WLU−
2
and L1 reverse- d Long-term periapse map associated with WLS blue
1
time escapes, and L2 reverse-time escapes for C 3.0174 and WLU− red for C 3.0174
2
Figure 5: Manifold and periapse structures in the Sun-Saturn system.
convention similar to one that appears in Koon et al. and Gómez et al. 11, 17 is employed.
Let ΓSLi ,m denote the periapse contour formed by the mth intersection of the stable manifold
tube associated with the Li Lyapunov orbit in the P2 region, and ΓU Lj ,n denote the periapse
contour formed by the nth intersection of the unstable manifold tube associated with the Lj
Lyapunov orbit in the P2 region. Then, to examine a periapse Poincaré map, consider the
map in Figure 5b. For the Lyapunov orbits at the given value of Jacobi constant, the first
three periapses along each manifold WLS 1
and WLS−
2
appear, as marked by blue dots in the
figure, and distinct regions appear. Recall that the delineation between regions of allowed
periapses and apoapses occur where r̈ 0, and this boundary is plotted as a dotted black
line in the figures. Then, for a large number of arbitrary initial periapse locations, escaping
trajectories are examined in both forward and reverse time. Those trajectories that cross
0.15
0.04
0.03 0.1
y (dimensionless)
y (dimensionless)
0.02 0.05
0.01
0 ∗ ∗
0 ∗ ∗
−0.01 −0.05
−0.02
−0.1
−0.03
−0.04
0.94 0.96 0.98 1 1.02 1.04 1.06 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
a Sun-Saturn system; μ 2.858 × 10−4 , C 3.016 b Earth-Moon system; μ 1.215 × 10−2 , C 3.160
Figure 6: Comparison of periapse regions in different systems.
the boundary x xL1 − 0.01 are defined as forward-time L1 escape trajectories, while those
that cross x xL2 0.01 are defined as forward-time L2 escapes. Colors represent the fate
of these initial periapse states. The colored areas in Figure 5b identify the locations of the
first three periapses along these L1 and L2 forward-time escape trajectories and are denoted
L1 ,1 → 3 , and ΠL2 ,1 → 3 , respectively, where superscript t indicates a forward-time escape. The
Πt t
map associated with the first three periapses along the unstable manifolds WLU 1
and WLU−2

is simply the reflection of the stable manifold contours from Figure 5b across the x-axis, 9
as demonstrated in Figure 5c, and represent entry or capture into the specified region.
The colored areas in Figure 5c represent locations of the first three periapses along the L1
and L2 reverse-time escape trajectories and are denoted Πt− L1 ,1 → 3 and ΠL2 ,1 → 3 , respectively,
t−
where superscript t− indicates reverse-time escape. Propagating WLS 1

and WLU− 2
for a longer
interval and plotting all the manifold periapses together, the periapse structures appearing
in Figure 5d emerge. The patterns apparent in the colored periapse regions are a function
of the mass ratio as well as the value of the Jacobi integral. However, due entirely to the
structure of the invariant manifolds associated with Lyapunov orbits, the patterns reappear
in different systems as is apparent in Figure 6. The Sun-Saturn mass ratio is μ 2.858 × 10−5
as compared to the Earth-Moon system for which μ 1.215 × 10−2 . The values of C differ, of
course, but similar patterns are apparent in the two different systems as expected.
Ultimately, these maps represent pathways through the system. To highlight the paths
that are available in this type of map, consider Figure 7, where both the Sun-Saturn and
Earth-Moon systems are represented. In Figures 7a and 7b, the regions corresponding
to the first six periapses along L1 forward-time escape trajectories are identified by color.
Contours ΓSLi ,m appear in blue for m within ∼33 orbits of the primaries. The values of C in
each system are selected for visual comparison and simply represent a similar opening of
the gateway at L2 . Sample paths are over plotted on the maps and periapses are marked by
dots in Figures 7c and 7d. Assume that the initial periapsis occurs in the yellow region

ΠtL1 ,6 . The propagated path then moves to the next periapses in the green region ΠtL1 ,5 ,

and so on. The final periapsis occurs in the orange ΠtL1 ,1 region with a subsequent escape
through L1 . In Figures 7a and 7b, it is also apparent that some white regions exist. These
are regions outside the lobes, that is, outside of the manifolds. Because the initial states in
these white regions lie outside the manifolds, periapses in these regions can correspond to
long-term behavior in the system. Not all white regions in Figure 7 correspond to long-term
0.15
0.04
0.03 0.1
y (dimensionless)
y (dimensionless)
0.02
0.05
0.01
0 ∗ ∗ 0 ∗ ∗
−0.01 −0.05
−0.02
−0.1
−0.03
−0.04
0.94 0.96 0.98 1 1.02 1.04 1.06 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
t+ t+ t+ t+
L1 ,1 L1 ,4 L1 ,1 L1 ,4
t+ t+ t+ t+
L1 ,2 L1 ,5 L1 ,2 L1 ,5
t+ t+ t+ t+
L1 ,3 L1 ,6 L1 ,3 L1 ,6
a Sun-Saturn escape map for six periapses at a value b Earth-Moon escape map for six periapses at a value
of Jacobi constant C 3.0174 of Jacobi constant C 3.17212
0.15
0.03 0.1
y (dimensionless)
0.02
y (dimensionless)
0.05
0.01
0 ∗ ∗ 0 ∗ ∗
−0.01 −0.05
−0.02
−0.1
−0.03
−0.04
0.94 0.96 0.98 1 1.02 1.04 1.06 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
c Sun-Saturn sequence originates with Πt
L1 ,6 through d Earth-Moon sequence originates with Πt
L1 ,6 to
Πt
L1 ,1 to escape through the L1 gateway
escape through the L1 gateway
Figure 7: Comparison of periapse maps for different Jacobi values and different systems.
capture in the vicinity of Saturn. At this value of Jacobi in the Sun-Saturn system, a relatively
large white region appears near P2 and a zoom of the same region near P2 at C 3.0173
appears in Figure 8. In Figure 8, the periapses along the manifold associated with the L1
Lyapunov orbit appear in blue WLS 1
and those associated with the L2 orbit appear in red
WL2 . The lobes reflecting escapes appear in white in Figure 8. The black dots correspond to
S−
periapses representing periodic orbits and other trajectories that evolve in this system for as
long as 1000 years and it is apparent that there is significant structure in this long-term map.
Numerous orbits and quasiperiodic trajectories that yield certain characteristics are selected
directly from the map in Davis 35.
4. Applications of Periapse Maps

Exploiting Poincaré maps to construct trajectories has certainly been accomplished by others.
Producing trajectories with certain characteristics can be facilitated with these types of
periapse maps as well. Maps at different energy levels can also be employed together to blend
0.25
0.2
0.15
0.1
0.05
yp (rH ) 0
−0.05
−0.1
−0.15
−0.2
−0.25
−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.2 0.25
xp (rH )
Figure 8: Long-term trajectories in a periapse Poincaré map; Sun-Saturn system; C 3.0173.
arcs. Examples illustrating the process appear below. In some examples, known solutions
emerge quickly.
4.1. Example: Transit through Both L1 and L2 Gateways

The unstable manifold tubes corresponding to the L1 and L2 Lyapunov orbits delineate
regions in the periapse maps that correspond to trajectories that enter the vicinity of P2
through the L1 or L2 gateways. As previously noted, the periapses of the unstable manifold
trajectories are the mirror image reflected across the x-axis 9 of the stable manifold apses.
A trajectory that lies both within a stable L1 tube and an unstable L2 tube can represent a
“double transit” trajectory, that is, a trajectory that transits through both gateways. Such a
trajectory enters the P2 vicinity through L2 and subsequently escapes, after an unspecified
number of revolutions about P2 , through L1 11, 13, 17, 36. For a particle to move from
the exterior to the interior region requires such a path. Similarly, a transit trajectory may
enter through the L1 gateway and depart through L2 . A sample transit trajectory L2 → L1
appears in Figure 9 in the Sun-Saturn system. The first two lobes representing periapses
within the L2 unstable manifold appear in red; the first two lobes associated with the L1 stable
manifold appear in the figure in blue. A periapse state is selected that lies within both of the
tubes; it appears as a black dot. Thus, the two lobes Πt− L2,2 and ΠL1,2 overlap and the selected
t
point appears in the intersection. The result is a transit trajectory that enters the vicinity of
Saturn through L2 and completes three periapse passages before escaping through the L1
gateway, passing through three periapse lobes in sequence that highlight its passage. One
advantage of the periapse Poincaré map for trajectory design applications is that the maps
exist in configuration space, allowing the selection of initial conditions based on the physical
location of periapsis. This type of application is explored in Davis and Haapala 35, 36.
Haapala and Howell 33 further explore the use of periapse Poincaré maps as a transit
trajectory design tool. By selecting initial conditions that correspond to periapses within the
region inside both contours ΓSLi ,m and ΓULj ,n , that is, within the intersections of regions ΠLi ,m
t
0.5
y (rH )
0
−0.5
−1 −0.5 0 0.5 1
x (rH )
Figure 9: Transit from the exterior region to the interior region through gateways at both L2 and L1 .
0.04
0.03
0.02
y (dimensionless)
0.01
0 ∗ ∗
−0.01
−0.02
−0.03
−0.04
0.94 0.96 0.98 1 1.02 1.04 1.06
x (dimensionless)
t+ t+
L1 ,1 L1 ,4
t+ t+
L1 ,2 L1 ,5
t+ t+
L1 ,3 L1 ,6
Figure 10: Arrival contour ΓU

L2 ,1 in red and contours ΓL1 ,1 → 10 in blue; C 3.0174 Saturn at 10x.
S
and Πt− Lj ,n , and propagating in both forward- and reverse-time, a transit orbit passing through
the Li gateway in forward time, and the Lj gateway in reverse time is produced. Defining
one revolution about P2 as consisting of one periapsis and one apoapsis, the transit trajectory
experiences a number of revolutions about P2 equal to p m n − 3/2. Thus, to design
transit trajectories with some desired behavior in the vicinity of P2 , the contours and/or
intersections are selected such that m n p 3/2, where for an Interior-to-Interior I-to-
I transit i j 1, for an exterior-to-exterior E-to-E transit i j 2, for an Interior-
to-Exterior I-to-E transit i 2, j 1, and lastly for an Exterior-to-Interior E-to-I transit
i 1, j 2. Reconsider an E-to-I transfer using the map in Figure 10. After entering the
L2 gateway, all E-to-I transit trajectories reach their first periapses within the first contour
ΓUL2 ,1 . Likewise, the last periapsis before exiting through the L1 gateway occurs within the first
×10−3
3.5 ×107
S 5
3 L1 ,3
y (dimensionless)
2.5 S
L1 ,4
y (km)
2
S
0 ∗ ∗
1.5 L1 ,5
U
L2 ,1
1
0.5 −5
0.992 0.993 0.994 0.995 0.996 1.35 1.4 1.45 1.5
×109
x (dimensionless) x (km)
a Zoom view of ΓU
L2 ,1 b p 2.5 tP2 4.51 revs 21.14 years
×107 ×107
5 5
y (km)
y (km)
0 ∗ ∗ 0 ∗ ∗
−5 −5
1.35 1.4 1.45 1.5 1.35 1.4 1.45 1.5
×109 ×109
x (km) x (km)
c p 3.5 tP2 6.10 revs 28.59 years d p 4.5 tP2 8.71 revs 40.81 years
Figure 11: Transit trajectories of varying numbers of revolutions about Saturn appear with periapses
marked in red.
contour ΓSL1 ,1 . Selecting n 1, m 10, it is possible to obtain trajectories with a maximum of

p 9.5 revolutions about P2 , although increasing n, m, or both could certainly render p ≥ 9.5.
The scenario from Figure 7a is repeated in Figure 10 but the manifold periapses are plotted
for fewer crossings. The figure includes contours ΓSL1 ,1 → 10 as blue dots, and ΓU L2 ,1 as a contour
in red indicated by a red arrow. The six regions ΠL1 ,1 → 6 prior to escape out L1 are colored
t
appropriately. Within the red L2 entry lobe, a certain structure is apparent when considering
the intersections with the L1 escaping contours as noted by Haapala and Howell 33 as well
as Haapala 36. Identifying the overlaps of the lobe regions yields transits that correspond
to p 2.5, 3.5, 4.5 and are colored in magenta, navy, and green in Figure 11a. Note that
the manifold contours at this particular value of Jacobi constant further split the arrival lobe
into different subregions that reflect different types of E-to-I transit paths in terms of p, that
is, the number of revolutions about P2 . Representative E-to-I transit trajectories are generated
from the initial conditions marked as red points in Figure 11a and are plotted in Figures
11b–11d. The time interval in the figures corresponds to the time required to pass from
x xL2 to x xL1 and appears in the captions.
4.2. Example: Earth-Moon Transfers
Periapse Poincaré maps are also applied to the problem of designing a low-energy ballistic
lunar transfer from low Earth orbit. Examined by various researchers 10, 12, 24, 37, 38, a
ballistic lunar transfer utilizes the gravity of the Sun to naturally raise the periapsis of an
Earth-centered trajectory, lowering the Δv required to reach the orbital radius of the Moon as
compared to a Hohmann transfer. Initial condition or periapse maps simplify the problem of
determining both the Δv and the orientation in the Sun-Earth frame that yield the appropriate
periapse raise.
Consider a spacecraft in a 167 km circular parking orbit centered at Earth. At this
energy level, the ZVCs are completely closed and the Sun has little effect on the orbit. A
maneuver applied at an appropriate location in the parking orbit decreases Jacobi Constant
and shifts the spacecraft to periapsis of a large Earth-centered orbit. This larger orbit is
affected significantly by the Sun, and the subsequent periapsis is raised to the radius of the
lunar orbit. If RE is the radius of the Earth, to reach the Moon’s orbital radius, the periapsis
must be raised from RE 167 km to 384,400 km, corresponding to Δrp 377, 855 km
0.2525 rH , in terms of the Sun-Earth Hill radius. To determine the required Δv, periapse maps
are created for a series of post-Δv Jacobi values. For each value of Δv, a Δrp map is created.
These maps allow a quick visualization of the orientation that produces the largest increase in
periapse radius for each Δv. The ZVCs and the trajectories corresponding to approximately
the largest periapse increase for a set of eight values of Δv appear in Figure 12a. For
Δv < 3.199 km/s, the ZVCs constrain the apoapsis to a radius too small to allow solar gravity
to raise periapsis sufficiently. For Δv ≥ 3.2 km/s, however, the ZVCs are sufficiently open
that is, the low value of Jacobi constant allows open gateways to result in a periapse raise
sufficiently large to reach the lunar orbit. This value agrees well with a theoretical minimum
Δv determined by Sweetser 39 as 3.099 km/s for transfer from a 167-km parking orbit at
Earth, as well as with optimized Earth-Moon transfer Δv values calculated by Parker and
Born 38, who computes Δv ≈ 3.2 km/s for a transfer from a 185 km parking orbit in a Sun-
Earth-Moon gravity model.
A maneuver of 3.2 km/s shifts the value of Jacobi Constant from C 3.068621, the
value of C corresponding to the low-Earth orbit, to C 3.000785 in the Sun-Earth system,
the value of C associated with the transfer orbit. The postmaneuver initial condition map
corresponding to Δv 3.2 km/s, or C 3.000785, appears in a full view in Figure 12b and a
zoomed view in Figure 12c; both are in the Sun-Earth rotating frame. The colors in Figures
12b and 12c simplify the options over the next revolution, that is, at this Jacobi constant,
C 3.00078518, depending on the location of the post-Δv periapsis along the parking orbit,
the spacecraft can impact Earth black or escape the vicinity of the Earth entirely red or
blue. However, in this application, the focus is on the trajectories that remain in the vicinity
of the Earth for at least one revolution, but with a significant rise in radius at the second
periapsis. To locate the orientation for the appropriate periapse raise, a Δrp map is produced.
The map is recolored in Figure 12d to reflect the value of Δrp over one revolution. Note
that black indicates a decrease in Δrp and blue → red indicates an increasing magnitude of
Δrp . The 167 km parking orbit is again marked in green on the map in Figure 12d. Four
bands of initial periapse angles exist that correspond to Δrp 0.2525 rH ; these bands are
marked by purple rays. By selecting an initial periapsis at the intersection of a purple ray
with the green parking orbit, the orientation is determined that will result in the desired
raise in periapsis. Four such trajectories appear in Figure 12e corresponding to the four
initial conditions marked on the map in Figure 12d. Clearly, each trajectory originates from
y (rH ) 0.5 0.5
y (rH )
0 0
−0.5
−0.5
−1
−1 0 1 −1 −0.5 0 0.5 1
x (rH ) x (rH )
v = 3.194 km/s v = 3.198 km/s

v = 3.195 km/s v = 3.199 km/s
v = 3.196 km/s v = 3.2 km/s
v = 3.197 km/s v = 3.201 km/s
a Trajectories at a series of Δv values, each b Sun-Earth initial condition map corresponding
oriented for the largest periapse raise; lunar orbital to C 3.000785
radius marked in green
×10−3 ×10−3
5 5
IVb
IIa
y (rH )
y (rH )
0 0
IVa
IIb
−5 −5
−5 0 5 −5 0 5
x (rH ) x (rH )
c Zoom near Earth; 167-km altitude; Earth orbit d Initial condition map colored in terms of Δrp
is green over one rev; desired Δrp for Earth-Moon transfer
denoted by purple rays; four selected ICs
0.5 IIa
y (rH )
IVb
0 IIb
−0.5 IVa
−1
−1 0 1
x (rH )
e Four Earth-Moon transfer trajectories correlated
to selected points in map d; Moon’s orbit is green
Figure 12: Low-energy Earth-Moon transfer.

the 167-km parking orbit; after a Δv 3.2 km/s, the trajectory reaches apoapsis well beyond
the radius of the Moon’s orbit. But the subsequent periapsis along each orbit is precisely at
the radius of the Moon’s orbit. Given that the data for the maps is available, computed ahead
or in real time, the appropriate point is selected directly from the map; in a simple three-body
model, the map is a fast easy method for preliminary design.
4.3. Example: Arrival through the L1 Gateway

Sun-Saturn System
Consider arrivals in the vicinity of Saturn either by natural objects or spacecraft. The arrival
lobe from either gateway can serve to deliver different types of arrival trajectories. Consider
first an arrival through the L1 gateway in the Sun-Saturn system as represented in Figure 13.
The goal is an arrival periapsis that will produce a path that remains in the P2 vicinity.
In Figure 13a, the arrival contour ΓU L1 ,1 appears in red. Also plotted in the figure are the
contours representing ΓSL2 ,1 → n . In this example, the points within Πt− L1 ,1 are colored based on
time to escape from P2 through either gateway, although not all trajectories escape within
the time of propagation. Thus, periapses within the arrival region, Πt− L1 ,1 , are colored such
that red → blue indicates increasing time to escape. Selection of a periapsis point that is in a
region of the deepest blue color yields a trajectory that will remain in the P2 vicinity for an
extended time. Such a point is indicated in white with a white arrow in Figure 13a. The
corresponding path appears in Figure 13b. Once propagated, it remains near Saturn for at
least 1000 years. Arrival through L2 could be accomplished with the same process.
Earth-Moon System
The same type of arrival lobe can yield a path for a different application in the Earth-Moon
system in Figure 14. Many investigators have examined transfers from Earth to a periodic L1
Lyapunov or halo orbit including Perozzi and Di Salvo 37 and Parker and Born 23. The L1
arrival contour for a given value of Jacobi Constant C 3.17212 appears in Figure 14a. A
periapsis is selected within this lobe as indicated by the red dot and arrow that also lies on
a stable invariant manifold associated with the Lyapunov orbit. The selected lunar periapse
point represents the gold trajectory arc in Figures 14b and 14c. Earth departure occurs as
a result of a maneuver, Δv1 3.105 km/s, from a 200 km circular Earth parking orbit in black.
The black arc then intersects the manifold gold corresponding to an L1 Lyapunov orbit
Δv2 0.630 km/s. After a close pass by the Moon at 100 km altitude, the path eventually
asymptotically approaches the L1 orbit as seen in Figures 14b and 14c. The final result
is a total Δv 3.735 km/s in the CR3BP. The periapsis is selected specifically to be along a
manifold and is located on the blue contour ΓSL1 ,1 , however, this path could also produce a
capture into a 100-km lunar orbit by including a 0.631 km/s maneuver at periapsis.
In Figure 15, the same objective is achieved with more excursions about the Moon
prior to arrival in the L1 orbit. Again, consider the arrival contour ΓU L1 ,1 as plotted as the
red lobe in Figure 14a. Within this contour, select a periapsis from the green region, which
represents the periapses corresponding to escaping trajectories from Πt L1,5 . Propagating an
initial condition from this region yields a path with four revolutions about the Moon prior
to departure again through L1 . If the periapsis initial condition that is selected in the green
region also lies on the contour ΓSL1 ,5 , as noted in the figure, the green trajectory arc in Figure 15
×10−3 S
L1 ,1
0 S
L1 ,2
−1
S
L1 ,3
y (dimensionless)
−2 S
L1 ,4
−3
−4
−5
S U
L1 ,6 L1 ,1
−6
1.001 1.002 1.003 1.004 1.005 1.006 1.007 1.008 1.009
x (dimensionless)
a Sun-Saturn system C 3.0174; arrival contour ΓU L1 ,1 overlaps with
contours ΓSL1 ,1 → m ; escape periapses colored by time to escape
0.04
0.03
0.02
y (dimensionless)
0.01
0 ∗ ∗
−0.01
−0.02
−0.03
−0.04
0.94 0.96 0.98 1 1.02 1.04 1.06
x (dimensionless)
b Selection of arrival periapsis in deep blue region results in a path that
remains in the vicinity of P2 for over 1000 years
Figure 13: Arrival through the Sun-Saturn L1 gateway; periapsis selected to yield a trajectory that remains
in the P2 vicinity for an extended time.
is generated. In comparison to the path in Figures 14b and 14c, the green path in Figure 15
completes four revolutions about the Moon, passing the Moon at a higher altitude with a
result that the total Δv 3.794 km/s with a corresponding increase in the time of flight.
Other periapses yield opportunities to insert into alterative lunar trajectories as well.
5. Summary and Concluding Remarks

Trajectory design in the multi-body regime remains a nontrivial problem. The goal in
any design process is to exploit the gravity of multiple bodies and deliver a trajectory
with characteristics that meet the requirements for a particular mission. Without analytical
solutions, increasing insight into the dynamical structure in the three-body problem
Arrival
trajectory Periapsis
arc S
0.005 L1 ,1
0 S
L1 ,2
−0.005
y (dimensionless)
S
L1 ,3
−0.01
S
−0.015 L1 ,4
−0.02
Periapsis
−0.025 S
S L1 ,6
−0.03 L1 ,5 U
L1 ,1
−0.035
0.98 0.99 1 1.01 1.02 1.03 1.04

x (dimensionless)
a Arrival contour ΓU
L1 ,1 and points within the lobe ΠL1,1 colored in terms
t−
of overlaps with regions Πt L1 ,1 → 6 as determined from contours ΓL1 ,1 → 6 . A

S
periapse point is selected red dot and the trajectory computed in forward
and backward time from that periapsis in gold
0.15
y (dimensionless)
0.5
y (dimensionless)
0.1
0 ∗ ∗ 0.05
0 ∗ ∗
−0.5
−0.05
−1 −0.1
−1 −0.5 0 0.5 1 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
b Earth departure as a result of a maneuver from a c Zoom: close pass by the Moon and approach to the
200-km circular parking orbit onto the black arc that Lyapunov orbit
intersects an L1 Lyapunov orbit manifold
Figure 14: Earth-Moon system C 3.17212; manifold trajectory connects to the black arc that departs from
200-km altitude Earth orbit; lunar periapsis at 100-km altitude; asymptotic approach to L1 Lyapunov orbit.
introduces more options for mission design as well as some new and exotic trajectories
that may enable future opportunities. However, the design space is very large and the
tradeoffs are not well defined. Many investigators are working to develop techniques to
create viable solutions for applications. The invariant manifold structure associated with the
collinear libration points, in particular, has supplied a geometrical framework and a number
of approaches are now available to represent this information. This current analysis explores
0.15
y (dimensionless)
0.5
y (dimensionless)
0.1
0 0.05
0
−0.5
−0.05
−1 −0.1
−1 −0.5 0 0.5 1 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
a b
Figure 15: Earth-Moon system; C 3.17212; manifold trajectory connects to the black arc that departs from
200-km altitude Earth orbit; asymptotic approach to L1 Lyapunov orbit after four revolutions about the
Moon.
an additional type of map to add another perspective for selecting arcs in support of the
design process. Within the context of the CR3BP, the periapse maps are an efficient design
tool. Using a combination of strategies to manage the information and deliver the results
is almost always necessary, however, automating these techniques is a priority; a visual
interface is a longer-term goal.
Acknowledgments
The authors wish to thank the Purdue University Graduate School, the College of
Engineering, and the School of Aeronautics and Astronautics for support of this effort.
Additional support is also appreciated through a GAANN Fellowship Graduate Assistance
in Areas of National Need, a Zonta International Amelia Earhart Fellowship, and a Purdue
Forever Fellowship. Assistance for the computational facilities is provided through the
Eliasen Visualization Laboratory. The paper has been presented at the 6th International
Workshop and Advanced School “Spaceflight Dynamics and Control,” Covilha, Portugal,
March 28-30, 2011, http://www.aerospace.ubi.pt/workshop2011/.
References
1 K. C. Howell, B. T. Barden, and M. W. Lo, “Application of dynamical systems theory to trajectory
design for a libration point mission,” Journal of the Astronautical Sciences, vol. 45, no. 2, pp. 161–178,
1997.
2 M. W. Lo, B. G. Williams, W. E. Bollman et al., “Genesis mission design,” Journal of the Astronautical
Sciences, vol. 49, no. 1, pp. 169–184, 2001.
3 V. Angelopoulos, “The THEMIS mission,” Space Science Reviews, vol. 141, no. 1–4, pp. 5–34, 2008.
4 V. Angelopoulos, “The ARTEMIS mission,” Space Science Reviews. In press.
5 M. Woodard, D. Folta, and D. Woodfork, “ARTEMIS: the first mission to the lunar libration points,” in
Proceedings of the 21st International Symposium on Space Flight Dynamics, Toulouse, France, September-
October 2009.
6 M. Hénon, “Exploration of the restricted problem. V. Hill’s case: periodic orbits and their stability,”
Astronomy and Astrophysics, vol. 1, pp. 223–238, 1969.
7 M. Hénon, “Exploration of the restricted problem. VI. Hill’s case: non-periodic orbits,” Astronomy and
Astrophysics, vol. 9, pp. 24–36, 1970.
8 D. P. Hamilton and J. A. Burns, “Orbital stability zones about asteroids,” Icarus, vol. 92, no. 1, pp.
118–131, 1991.
9 H. Yamakawa, J. Kawaguchi, N. Ishil, and H. Matsuo, “On earth-moon transfer trajectory with
gravitational capture,” in Proceedings of the AAS/AIAA Astrodynamics Specialist Conference, Victoria,
Canada, August 1993, Paper No. AAS 93-633.
10 E. A. Belbruno and J. K. Miller, “Sun-perturbed earth-to-moon transfers with ballistic capture,” Journal
of Guidance, Control, and Dynamics, vol. 16, no. 4, pp. 770–775, 1993.
11 W. S. Koon, M. W. Lo, J. E. Marsden, and S. D. Ross, “Heteroclinic connections between periodic orbits
and resonance transitions in celestial mechanics,” Chaos, vol. 10, no. 2, pp. 427–469, 2000.
12 W. S. Koon, M. W. Lo, J. E. Marsden, and S. D. Ross, “Low energy transfer to the moon,” Celestial
Mechanics and Dynamical Astronomy, vol. 81, no. 1-2, pp. 63–73, 2001.
13 G. Gómez, W. S. Koon, M. W. Lo, J. E. Marsden, J. Masdemont, and S. D. Ross, “Invariant
manifolds, the spatial three-body problem and space mission design,” in Proceedings of the AAS/AIAA
Astrodynamics Specialist Conference, Quebec, Canada, August 2001, Paper No. AAS 2001-301.
14 K. C. Howell, B. G. Marchand, and M. W. Lo, “Temporary satellite capture of short-period Jupiter
family comets from the perspective of dynamical systems,” American Astronautical Society. Journal of
15 A. F. Prado, “Third-body perturbation in orbits around natural satellites,” Journal of Guidance, Control,
16 F. Topputo, M. Vasile, and A. E. Finzi, “Combining two and three-body dynamics for low energy
transfer trajectories of practical interest,” in Proceedings of the 55th International Astronautical Congress,
pp. 584–596, Vancouver, Canada, October 2004, IAC-04-A.7.02.
17 G. Gómez, W. S. Koon, M. W. Lo, J. E. Marsden, J. Masdemont, and S. D. Ross, “Connecting orbits
and invariant manifolds in the spatial restricted three-body problem,” Nonlinearity, vol. 17, no. 5, pp.
1571–1606, 2004.
18 F. Topputo, M. Vasile, and F. Bernelli-Zazzera, “Low energy interplanetary transfers exploiting
invariant manifolds of the restricted three-body problem,” Journal of the Astronautical Sciences, vol.
53, no. 4, pp. 353–372, 2005.
19 R. L. Anderson and M. W. Lo, “Virtual exploration by computing global families of trajectories with
supercomputers,” in Proceedings of the AAS/AIAA Spaceflight Mechanics Meeting, Copper Mountain,
Colo, USA, January 2005, Paper No. AAS 05-220.
20 R. P. Russell, “Global search for planar and three-dimensional periodic orbits near Europa,” Journal of
21 C. F. de Melo and O. C. Winter, “Alternative paths to Earth-moon transfer,” Mathematical Problems in
Engineering, vol. 2006, Article ID 34317, 20 pages, 2006.
22 F. Garcı́a and G. Gómez, “A note on weak stability boundaries,” Celestial Mechanics and Dynamical
Astronomy, vol. 97, no. 2, pp. 87–100, 2007.
23 J. S. Parker and G. H. Born, “Direct lunar halo orbit transfers,” in Proceedings of the 17th AAS/AIAA
Space Flight Mechanics Conference, Sedona, Ariz, USA, 2007, Paper AAS 07-229.
24 J. S. Parker and G. H. Born, “Modeling a low-energy ballistic lunar transfer using dynamical systems
theory,” Journal of Spacecraft and Rockets, vol. 45, no. 6, pp. 1269–1281, 2008.
25 G. A. Tsirogiannis, E. A. Perdios, and V. V. Markellos, “Improved grid search method: an efficient tool
for global computation of periodic orbits,” Celestial Mechanics and Dynamical Astronomy, vol. 103, no.
1, pp. 49–78, 2009.
26 E. Belbruno, M. Gidea, and F. Topputo, “Weak stability boundary and invariant manifolds,” SIAM
Journal on Applied Dynamical Systems, vol. 9, no. 3, pp. 1061–1089, 2010.
27 R. L. Anderson and M. W. Lo, “A dynamical systems analysis of resonant flybys: ballistic case,” The
Journal of the Astronautical Sciences, vol. 58, no. 2, 2011.
28 B. F. Villac and D. J. Scheeres, “Escaping trajectories in the Hill three-body problem and applications,”
29 M. E. Paskowitz and D. J. Scheeres, “Robust capture and transfer trajectories for planetary satellite
orbiters,” Journal of Guidance, Control, and Dynamics, vol. 29, no. 2, pp. 342–353, 2006.
30 M. E. Paskowitz and D. J. Scheeres, “Design of science orbits about planetary satellites: Application
to Europa,” Journal of Guidance, Control, and Dynamics, vol. 29, no. 5, pp. 1147–1158, 2006.
31 D. C. Davis and K. Howell, “Long term evolution of trajectories near the smaller primary in the
restricted problem,” in Proceedings of the 20th AAS/AIAA Space Flight Mechanics Meeting, San Diego,
Calif, USA, February 2010.
32 D. C. Davis and K. Howell, “Trajectory evolution in the multi-body problem with applications in
the saturnian system,” in Proceedings of the 61st International Astronautical Congress, Prague, Czech
Republic, September 2010.
33 A. F. Haapala and K. C. Howell, “Trajectory design using Poincaré maps and invariant manifolds,”
in Proceedings of the 21st AAS/AIAA Space Flight Mechanics Meeting, New Orleans, La, USA, February
2011.
34 B. F. Villac and D. J. Scheeres, “On the concept of periapsis in Hill’s problem,” Celestial Mechanics and
Dynamical Astronomy, vol. 90, no. 1-2, pp. 165–178, 2004.
35 D. C. Davis, Multi-body trajectory design strategies based on periapsis Poincaré maps, Ph.D. dissertation,
School of Aeronautics and Astronautics, Purdue University, West Lafayette, Ind, USA, 2011.
36 A. F. Haapala, Trajectory design using periapsis Poincaré maps and invariant manifolds, M.S. thesis, School
of Aeronautics and Astronautics, Purdue University, West Lafayette, Ind, USA, 2011.
37 E. Perozzi and A. Di Salvo, “Novel spaceways for reaching the Moon: an assessment for exploration,”
Celestial Mechanics and Dynamical Astronomy, vol. 102, no. 1–3, pp. 207–218, 2008.
38 J. Parker and G. Born, “Targeting low-energy ballistic lunar transfers,” in Proceedings of the AAS George
H. Born Symposium, Boulder, Colo, USA, May 2010, Paper No. AAS 10-1859.
39 T. H. Sweetser, “An estimate of the global minimum ΔV needed for earth-moon transfer,” in
Proceedings of the AAS/AIAA Spaceflight Mechanics Meeting, Houston, Tex, USA, February 1991, Paper
No. AAS 91-101.
doi:10.1155/2012/985429
Research Article
Unscented Kalman Filter Applied to the Spacecraft
Attitude Estimation with Euler Angles
Roberta Veloso Garcia,1 Helio Koiti Kuga,1

and Maria Cecilia F. P. S. Zanardi2
1
Space Mechanic and Control Division, INPE, 12227-010 São José dos Campos, SP, Brazil
2
Department of Mathematics, FEG, UNESP, 12516-410 Guaratinguetá, SP, Brazil
Correspondence should be addressed to Roberta Veloso Garcia, rvelosogarcia@yahoo.com.br
Copyright q 2012 Roberta Veloso Garcia et al. This is an open access article distributed under
The aim of this work is to test an algorithm to estimate, in real time, the attitude of an artificial
satellite using real data supplied by attitude sensors that are on board of the CBERS-2 satellite
China Brazil Earth Resources Satellite. The real-time estimator used in this work for attitude
determination is the Unscented Kalman Filter. This filter is a new alternative to the extended
Kalman filter usually applied to the estimation and control problems of attitude and orbit.
This algorithm is capable of carrying out estimation of the states of nonlinear systems, without
the necessity of linearization of the nonlinear functions present in the model. This estimation
is possible due to a transformation that generates a set of vectors that, suffering a nonlinear
transformation, preserves the same mean and covariance of the random variables before the
transformation. The performance will be evaluated and analyzed through the comparison between
the Unscented Kalman filter and the extended Kalman filter results, by using real onboard data.
1. Introduction
The attitude of a spacecraft is defined by its orientation in space related to some reference
system. The importance of determining the attitude is related not only to the performance
of attitude control but also to the precise usage of information obtained by payload
experiments performed by the satellite. The attitude estimation is the process of calculating
the orientation of the spacecraft in relation to a reference system from data supplied by
attitude sensors. Chosen the vector of reference, an attitude sensor measures the orientation
of these vectors with respect to the satellite reference system 1. Once these one or more
vectors measurements are known, it is possible to compute the orientation of the satellite
processing these vectors, using methods of attitude estimation. There are several methods
for determining the attitude of a satellite. Each method is appropriate to a particular type of
application and meets the needs such as available time for processing and accuracy to be
attained. However, all methods need observations that are obtained by means of sensors
installed on the satellite. The sensors are essential for attitude estimation, because they
measure its orientation relative to some referential, for example, the Earth, the sun, or a star.
In this work, the satellite attitude is described by Euler angles, due to its easy geometric
interpretation, and the method to estimate the attitude used is the Unscented Kalman Filter.
This method is capable of performing state estimation in nonlinear systems, besides taking
into account measurements provided by different attitude sensors. In this work there were
considered real data supplied by gyroscopes, infrared Earth sensors, and digital sun sensors.
These sensors are on board of the CBERS-2 satellite China-Brazil Earth Resources Satellite,
and the measurements were collected by the Satellite Control Centre of INPE Brazilian
Institute for Space Research.
2. Representation of Attitude by Euler Angles

The attitude of an artificial satellite is directly related to its orientation in space. Through
the attitude one can know the spatial orientation of the satellite, since in most cases it can
be considered as a rigid body, where the attitude is expressed by the relationship between
two coordinate systems, one fixed on the satellite and another associated with a reference
system, for example, inertial system 2. For a good performance of a mission it is essential
that the satellite be stabilized in relation to a specified attitude. The attitude stabilization is
achieved by the on-board attitude control, which is designed to acquire and maintain the
satellite in a predefined attitude. The CBERS-2 attitude is stabilized in three axes nominally
geo-pointed and can be described with respect to the orbital system. In this reference system,
the movement around the direction of the orbital velocity is called roll. The movement around
the direction normal to the orbit is called pitch, and finally the movement around the direction
Nadir/Zenith is called yaw. To transform a vector represented in a given reference to another
it is necessary to define a matrix of direction cosines R, where its elements are written in
terms of Euler angles φ, θ, ψ 3. The rotation sequence used in this work for the Euler
angles was the 3-2-1, where the coordinate system fixed in the body of the satellite x, y,
z is related to the orbital coordinate system xo, yo, zo through the following sequence of
rotations:
i 1st rotation of an angle ψ yaw angle around the zo axis,
ii 2nd rotation of an angle θ roll angle around an intermediate axis y ,
iii 3rd rotation of an angle φ pitch angle around the x-axis.
The matrix obtained through the 3-2-1 rotation sequence is given by
⎡ ⎤
CθC
ψ ψ −Sθ
CθS
R ⎣S φ SθC
ψ − S ψ C φ S φ SθS ψ CφCψ S φ Cθ
⎦
, 2.1
SθC ψ C φ S φ S ψ C φ SθS ψ − S φ C ψ CθC φ
where R is the matrix of direction cosines with S sin, C cos, and T tan.
By representing the attitude of a satellite with Euler angles, the set of kinematic
equations are given by 4
⎡ ⎤
⎡ ⎤ 1 S φ Tθ C φ T θ ⎧⎡ ⎤ ⎡ ⎤⎫
9x
φ̇ ⎢0 C φ −S φ ⎥⎨ w 0 ⎬
⎣ θ̇ ⎦ ⎢ ⎥ ⎣w ⎦ ⎣ ⎦
⎣ S φ ⎦⎩ 9 y − R −ω0 ⎭, 2.2
ψ̇ 0 C φ Cθ 9z
w 0
Cθ
9 x, w
where ω0 is the orbital angular velocity and w 9 y , and w
9 z are the components of the angular
velocity on the satellite system.
3. The Measurements System of Satellite

In order to estimate the satellite attitude accurately, several types of sensors, including gyros,
earth sensors, and solar sensors, are used in the attitude determination system. The equations
of these sensors are introduced here.
3.1. The Model for Gyros

The advantage of a gyro is that it can provide the angular displacement and/or angular
velocity of the satellite directly. However, gyros have an error due to drifting, meaning that
their measurement error increases with time. In this work, the rate-integration gyros RIGs
are used to measure the angular velocities of the roll, pitch, and yaw of the satellite. The
mathematical model of the RIGs is 4
Δt

ΔΘi ωi εi dt i x, y, z , 3.1
0
where ΔΘ are the angular displacement of the satellite in a time interval Δt, and εi are
components of bias of the gyroscope.
Thus, the measured components of the angular velocity of the satellite are given by

ΔΘ
9
ω − ε9 − η1 g − ε − ηatt , 3.2
Δt
where g t is the output vector of the gyroscope, and η1 t represents a Gaussian white noise
process covering all the remaining unmodelled effects:
+ ,
T
E ηatt ηatt σatt
2
. 3.3
3.2. The Measurement Model for Infrared Earth Sensors (IRESs)

One way to compensate for the drifting errors present in gyros is to use the earth sensors.
These sensors are located on the satellite and aligned with their axes of roll and pitch. In the
work, two earth sensors are used, with one measuring the roll angle and the other measuring
the pitch angle. In principle, an earth sensor cannot measure the yaw angle. The measurement
equations for the earth sensors are given as 5
φH φ ηφH ,
3.4
θH θ ηθH ,
where ηφH and ηθH are the white noise representing the small remaining misalignment,
installation, and/or assembly errors assumed to be gaussian:
+ , + ,
E ηφH ηφT H E ηθH ηθTH σIRES
2
. 3.5
3.3. The Measurement Model for Digital Solar Sensors (DSSs)

Since an earth sensor is not able to measure the yaw angle, the solar sensors are used by the
Attitude Control System in order to overcome this problem. However, these sensors do not
provide direct measurements but coupled angle of pitch αθ and yaw αψ . The measurement
equations for the solar sensor are obtained as follows 5:

−Sy
αψ tan−1 ηαψ 3.6
Sx cos60 Sz cos150◦
◦
when |Sx cos60◦ Sz cos150◦ | ≥ cos60◦ , and

Sx
αθ 24◦ − tan−1 ηαθ 3.7
Sz
when |24◦ − tan−1 Sx /Sz | < 60◦ , where ηαψ and ηαθ are the white noise representing the small
remaining misalignment, installation, and/or assembly errors assumed Gaussian:
+ , + ,
E ηαψ ηαTψ E ηαθ ηαTθ σDSS
2
. 3.8
The conditions are such that the solar vector is in the field of view of the sensor, and Sx ,
Sy , and Sz are the components of the unit vector associated to the sun vector in the satellite
system and given by
Sx S0x ψS 9 0z ,
9 0y − θS
Sy S0y − ψS 9 0z ,
9 0x φS 3.9
9 0y θS
Sz S0z − φS 9 0z ,
where S0x , S0y , and S0z are the components of the sun vector in the orbital coordinate system
9 θ,
and φ, 9 and ψ9 are the Euler angles-estimated attitude.
4. Attitude Estimation Methods

The goal of an estimator is to calculate the state vector attitude based on a set of observations
sensors 6. In other words, it is an algorithm capable of processing measurements to
produce, according with a given criterium, a minimum error estimate of the state of a system.
In this paper, the real-time estimator used to estimate the satellite attitude is a variant of the
Kalman filter, applied to problems that present some nonlinearity. This estimator is described
as follows.
4.1. Unscented Kalman Filter

The basic premise behind the Unscented Kalman Filter UKF is that it is easier to
approximate a Gaussian distribution than it is to approximate an arbitrary nonlinear function.
Instead of linearizing to first order using Jacobian matrices, the UKF uses a rational
deterministic sampling approach to capture the mean and covariance estimates with a
minimal set of sample points. The nonlinear function is applied to each point, in turn, to yield
a cloud of transformed points. The statistics of the transformed points can then be calculated
to form an estimate of the nonlinearly transformed mean and covariance. We present an
algorithmic description of the UKF omitting some theoretical considerations, left to 7, 8.
Consider the system model given by
ẋk 1 fxk , k Gk wk ,
4.1
yk hxk , k νk ,
where xk is the n × 1 state vector and yk is the m × 1 measurement vector. We assume that the
process noise wk and measurement-error noise νk are zero-mean Gaussian noise processes
with covariances given by Qk and Rk , respectively. In this work the state vector at time k is
defined by the Euler angles and gyro biases:
> ?T
x9k φ, θ, ψ, εx , εy , εz . 4.2
Performing the necessary simplifications small Euler angles in the set of 2.2, the
attitude angles and gyro angular velocity biases are modelled as follows:
9 x θ9ω
φ̇t ω0 sin ψ9 ω 9 z,
9 y φ9ω
θ̇t ω0 cos ψ9 ω 9 z,
4.3
ψ̇t ω0 θ9 sin ψ9 − φ9 cos ψ9 ω9 z φ9ω
9y,
ε˙ t 0.
Given the state vector at step k − 1, we compute a collection of sigma-points, stored in

the columns of the n × 2n 1 sigma point matrix χk−1 , where n is the dimension of the state
vector. In our case, n 6, so χk−1 is a 6 × 13 matrix. The columns of χk−1 are computed by

χk−1 0 x9k−1 ,

χk−1 i x9k−1 n λPk−1 , i 1, . . . , n,
4.4

i

χk−1 i x9k−1 − n λPk−1 , i n 1, . . . , 2n,
i−n
D
in which λR , n λPk−1 i is the ith column of the matrix square root of n λPk−1 .
Once χk−1 computed, we perform the prediction step by first predicting each column
of χk−1 through time by Δt using

χk i
f χk−1 i
, i 0, . . . , 2n, 4.5
where f is differential equation defined in 2.2 or 4.3. In our formulation, we perform a

numerical Runge-Kutta integration.
With χk i being calculated, the a priori state estimate is
&
2n

x9k− Wim χk i , 4.6
i0
where Wim are weights defined by
λ
W0m ,
n λ
4.7
1
Wim i 1, . . . , 2n.
2n λ
As the last part of the prediction step, we calculate the a priori error covariance with
&
2n
> ?> ?T
Pk− Wic χk i − x9k− χk i − x9k− Qk , 4.8
i0
where Qk is the process error covariance matrix, and the weights are defined by
λ
W0c 1 − α2 β2 ,
n λ
4.9
1
Wic i 1, . . . , 2n,
2n λ
where α is a scaling parameter which determines the spread of the sigma points and β is a
parameter used to incorporate any prior knowledge about the distribution of x 9.
To compute the correction step, we first must transform the columns of χk through the
measurement function to Yk . In this way

Yk i h χk i , i 0, . . . , 2n,
&
2n 4.10
y9k− Wim Yk i .
i0
With the mean measurement vector y9k− , we compute the a posteriori state estimate using

x9k x9k− Kk yk − y9k− , 4.11
where Kk is the Kalman gain. In the UKF formulation, Kk is defined by
Kk Px9k ,y9k Py9−1k ,y9k , 4.12
where
&
2n
c > ?> ?T
Py9k ,y9k Wi Yk i − y9k− Yk i − y9k− Rk ,
i0
4.13
&
2n
c > ?> ?T
Px9k ,y9k Wi χk i − x9k− Yk i − y9k− ,
i0
where Rk represents the measurement error covariance matrix.

Finally, the last calculation in the correction step is to compute the a posteriori estimate
of the error covariance given by
Pk Pk− − Kk Py9k ,y9k KkT . 4.14
5. Results
Here, the results and the analysis from the algorithm developed to estimate the attitude are
presented. To validate and to analyze the performance of the estimators, real sensors data
from the CBERS-2 satellite were used see 10, 11. The CBERS-2 satellite was launched
on October 21, 2003. The measurements are for the month of April 2006, available to the
ground system at a sampling rate of about 8.56 sec. The algorithm was implemented through
MATLAB software. To check the performance the UKF, their results were compared with the
estimated attitude by the more conventional EKF Extended Kalman Filter, considering the
following set of initial conditions:
i initial attitude: φ θ ψ 0 deg;

ii initial bias of gyro: εx 5.56 deg/hour, and εy 0.87 deg/hour, εz 6.12 deg/hour;
iii initial covariance P : σφ,θ,ψ
2
0.5 deg2 error related to the attitude, and σbg
2

1 deg/hour error related to the drift of gyro;
2
Measurements
Digital sun sensors
40
(deg)
20
0
−20
13.76 13.78 13.8 13.82 13.84 13.86 13.88 13.9 13.92 13.94
Time (hours)
DSS1
DSS2
a
Infrared earth sensor
Measurements
−0.35
(deg)
−0.45
−0.55
13.76 13.78 13.8 13.82 13.84 13.86 13.88 13.9 13.92 13.94
Time (hours)
IRES1
IRES2
b
−4
×10
Measurements
Incremental measures of gyroscope

10
(deg)
5
0
−5
13.76 13.78 13.8 13.82 13.84 13.86 13.88 13.9 13.92 13.94
Time (hours)
Gyrox
Gyroy (· 102 )
Gyroz
c
Figure 1: Real measurements supplied by attitude sensors.
iv observation noise covariance R: σDSS

2
0.3 deg2 sun sensor, are σIRES
2

0.03 deg earth sensor;
2
v dynamic noise covariance Q: σatt2

0.1 deg2 noise related to the attitude,
σDbgx σDbgy 0.01 deg/hour , and σDbg
2 2 2 2
z
0.005 deg/hour2 noise related to
the drifting of gyro.
The real measurements obtained by the attitude sensors digital sun sensors, infrared
Earth sensors, and gyroscopes are shown in Figure 1.
In Figures 2 and 3 are observed the behavior of attitude and the biases of gyro during
the period analyzed. The average estimated values for the axes of roll and pitch, considering
the Unscented Kalman Filter, are in the order of −0.47 deg and −0.45 deg, respectively, and
their standard deviations are about 0.02 deg. For the yaw axis the estimate seems not to behave
randomly and its average estimated value is about −1.47 deg with standard deviation 0.3 deg.
The attitude estimated by the extended Kalman filter had their values for the axes roll and
pitch in the order of about −0.49 deg and −0.43 deg with standard deviation about 0.05 deg.
For the axis pitch, its average value is −1.42 deg and standard deviation is 0.36 deg.
Estimated roll angle Estimated pitch angle
Attitude (deg)
Attitude (deg)
−0.3 −0.8
−0.4 −0.6
−0.5 −0.4
−0.6 −0.2
−0.7 0
13.76 13.8 13.84 13.88 13.92 13.76 13.8 13.84 13.88 13.92
a b
Estimated yaw angle

Attitude (deg)
0
−0.5
−1
−1.5
−2
13.76 13.8 13.84 13.88 13.92
Time (hours)
UKF
EKF
c
Figure 2: Attitude estimated by the Unscented Kalman filter end Extended Kalman filter.
Estimated gyro bias in x Estimated gyro bias in y

εx (deg/hour)
εy (deg/hour)
5.8 1.4
1.2
5.75 1
0.8
5.7
13.76 13.8 13.84 13.88 13.92 13.76 13.8 13.84 13.88 13.92
a b
Estimated gyro bias in z

εz (deg/hour)
6.4
6.3
6.2
6.1
13.76 13.8 13.84 13.88 13.92
Time (hours)
UKF
EKF
c
Figure 3: Bias estimated by the Unscented Kalman filter end Extended Kalman filter.
Attitude error estimation Attitude error estimation

0.2 0.2
σφ (deg)
σθ (deg)
0.15 0.15
0.1 0.1
0.05 0.05
0 0
13.76 13.8 13.84 13.88 13.92 13.76 13.8 13.84 13.88 13.92
Time (hour) Time (hour)
a b
Attitude error estimation

0.5
σψ (deg)
0.4
0.3
0.2
13.76 13.8 13.84 13.88 13.92
Time (hour)
UKF
EKF
c
Figure 4: Attitude errors Standard Deviation estimated.
Biases errors estimation in x Biases estimation errors in y

σεx (deg/hour)
σεy (deg/hour)
1.02 1.02
1.01 1.01
1 1
0.99 0.99
0.98 0.98
13.76 13.8 13.84 13.88 13.92 13.76 13.8 13.84 13.88 13.92
a b
Biases estimation errors in z

σεz (deg/hour)
1.02
1.01
1
0.99
0.98
13.76 13.8 13.84 13.88 13.92
Time (hour)
UKF
EKF
c
Figure 5: Bias errors Standard Deviation estimated.

Digital sun sensor (DSS1 ) Digital sun sensor (DSS2 )

0.4 0.8
Residuals (deg)
Residuals (deg)
0.2 0.6
0
−0.2 0.4
−0.4 0.2
−0.6
−0.8 0
−1 −0.2
13.76 13.8 13.84 13.88 13.92 13.76 13.8 13.84 13.88 13.92
UKF UKF
EKF EKF
a b
Figure 6: Residuals related to DSS attitude sensors.
Infrared earth sensor (IRES1 ) Infrared earth sensor (IRES2 )

0.2 0.2
Residuals (deg)
0.1
Residuals (deg)
0 0
−0.1
−0.2
−0.2
−0.4 −0.3
−0.4
−0.6 −0.5
13.76 13.8 13.84 13.88 13.92 13.76 13.8 13.84 13.88 13.92
UKF UKF
EKF EKF
a b
Figure 7: Residuals related to IRES attitude sensors.
Figures 4 and 5 present the standard deviations for both estimators for the attitude
and the bias of the gyro. It is observed that the attitude standard deviations and the standard
deviations of the gyro bias decrease with a tendency to stabilize around a value for both
filters. However, the graphs show the superiority of UKF, because in most cases it works
within a range of protection better than the EKF.
In Figures 6 and 7, we can see that the residues of sun sensors and Earth sensors have
the same behavior for both estimators. For the Earth sensors, the residuals obtained by the
both estimators are smaller and show a tendency to zero mean. However, the residues of
UKF are still lower, being at about −0.009 deg for IRES1 and 0.004 deg for IRES2 . Already
the residues of the EKF are approximately 0.01 deg and −0.027 deg for IRES1 and IRES2 ,
respectively.
These results seem consistent because in this case it is not possible to compare the
estimated values with true values, since these values are not known.
6. Final Comments
In this paper, the Unscented Kalman Filter estimator applied in nonlinear systems was
presented for use in real-time attitude estimation. The main objective was to estimate the
attitude of a CBERS-2 like satellite, using real data provided by sensors that are on board
of the satellite. To verify the consistency of the estimator, the attitude was estimated by two
different methods. The usage of real data from on-board attitude sensors poses difficulties
like mismodelling, mismatch of sizes, misalignments, unforeseen systematic errors, and
postlaunch calibration errors. However, it is observed that, although the EKF and UKF have
roughly the same accuracy, the UKF leads to a convergence of the state vector faster than the
EKF. This fact was expected, since the UKF prevents the linearizations needed for EKF, when
the system has some nonlinearity in their equations.
References
1 V. L. Pisacane and R. C. Moore, Fundamentals of Space Systems, Oxford University Press, New York,
NY, USA, 1994.
2 M. D. Shuster, “Survey of attitude representations,” Journal of the Astronautical Sciences, vol. 41, no. 4,
pp. 439–517, 1993.
3 M. C. Zanardi, Dinâmica de atitude de satelites artificiais, Ph.D. thesis, FEG-UNESP, Guaratinguetá, São
Paulo, Brazil, 2005.
4 J. R. Wertz, Spacecraft Attitude Determination and Control, D. Reidel, Dordrecht, The Netherlands, 1978.
5 H. Fuming and H. K. Kuga, “CBERS simulator mathematical models,” CBTT Project, CBTT /2000
/MM /001. INPE, S ão José dos Campos, 1999.
6 R. G. Brown and P. Y. C. Hwang, Introduction to Random Signals and Applied Kalman Filterin, Wiley,
New York, NY, USA, 1996.
7 S. J. Julier and J. K. Uhlmann, “Reduced sigma point filters for the propagation of means and
covariances through nonlinear transformations,” in Proceedings of the American Control Conference, vol.
2, pp. 887–892, Anchorage, Alaska, USA, May 2002.
8 S. J. Julier and J. K. Uhlmann, “Unscented filtering and nonlinear estimation,” Proceedings of the IEEE,
vol. 92, no. 3, pp. 401–422, 2004.
9 S.J. Julier and J. K. Uhlmann, “A new extension of the kalman filter for nonlinear systems,” in
Proceedings of the International Symposium on Aerospace/Defense Sensing, Simulation and Controls, (SPIE
’99), SPIE, Orlando, Fla, USA, April 1997.
10 H. K. Kuga, R. V. F. Lopes, and A. R. Silva, “On board attitude reconstitution of CBERS-2 Satellite,” in
Proceedings of the Proceedings of XIII Brazilian Coloquium on Orbital Dynamics, p. 109, 2006.
11 R. V. F. Lopes and H. K. Kuga, “CBERS-2: on ground attitude determination from telemetry data,”
Internal report C-ITRP, INPE, 2005.
doi:10.1155/2012/971983
Research Article
Optimal Two-Impulse Trajectories with
Moderate Flight Time for Earth-Moon Missions

and Cleverson Maranhão Porto Marinho
Departamento de Matemática, Instituto Tecnológico de Aeronáutica,
Correspondence should be addressed to Sandro da Silva Fernandes, sandro@ief.ita.br
Received 10 June 2011; Accepted 4 September 2011
Copyright q 2012 S. da Silva Fernandes and C. Maranhão Porto Marinho. This is an open access
article distributed under the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.
A study of optimal two-impulse trajectories with moderate flight time for Earth-Moon missions is
presented. The optimization criterion is the total characteristic velocity. Three dynamical models
are used to describe the motion of the space vehicle: the well-known patched-conic approximation
and two versions of the planar circular restricted three-body problem PCR3BP. In the patched-
conic approximation model, the parameters to be optimized are two: initial phase angle of space
vehicle and the first velocity impulse. In the PCR3BP models, the parameters to be optimized are
four: initial phase angle of space vehicle, flight time, and the first and the second velocity impulses.
In all cases, the optimization problem has one degree of freedom and can be solved by means of
an algorithm based on gradient method in conjunction with Newton-Raphson method.
1. Introduction
In the last two decades, new types of trajectories have been proposed to transfer a spacecraft
from an Earth orbit to a Moon orbit which reduce the cost of the traditional Hohmann transfer
based on the two-body dynamics 1. The new trajectories are designed using more realistic
models of the motion of the spacecraft such as the PCR3BP 2, 3 or the planar bicircular four
body problem 4. These models describing the motion of the spacecraft exhibit very complex
dynamics that are used to design new Earth-to-Moon trajectories 5–10. The most of the
proposed approaches to calculate the new trajectories are based on the concept of the weak
stability boundary introduced by Belbruno 11, and, usually, involve large flight time. Only
few works consider the minimization of the total cost or the time for two-impulse trajectories
8–10, 12, 13.
In this paper, the problem of transferring a space vehicle from a circular low Earth orbit
LEO to a circular low Moon orbit LMO with minimum fuel consumption is considered.
It is assumed that the propulsion system delivers infinite thrusts during negligible times
such that the velocity changes are instantaneous, that is, the propulsion system is capable of
delivering impulses. The class of two impulse trajectories is considered: a first accelerating
velocity impulse tangential to the space vehicle velocity relative to Earth is applied at a
circular low Earth orbit and a second braking velocity impulse tangential to the space vehicle
velocity relative to Moon is applied at a circular low Moon orbit 12. The minimization of
fuel consumption is equivalent to the minimization of the total characteristic velocity which
is defined by the arithmetic sum of velocity changes 14.
Three dynamical models are used to describe the motion of the space vehicle: the well-
known patched-conic approximation 1 and two versions of the planar circular restricted
three-body problem PCR3BP. One version of PCR3BP assumes the Earth is fixed in space;
this version will be referred as simplified version of PCR3BP, and it is same one used by
Miele and Mancuso 12. The second version of PCR3BP assumes the Earth moves around the
center of mass of the Earth-Moon system, that is, it corresponds to the classical formulation 2,
3. In the patched-conic approximation model, the parameters to be optimized are two: initial
phase angle of space vehicle and the first velocity impulse. In this approach, the two-point
boundary value problem involves only one final constraint. In this model, the flight time and
the second velocity impulse are determined from the two-body dynamics after solving the
two-point boundary value problem. In the PCR3BP models, the parameters to be optimized
are four: initial phase angle of space vehicle, flight time, and the first and the second
velocity impulses. In these formulations, the two-point boundary value problem involves
three final constraints. In all cases, the optimization problem has one degree of freedom
and can be solved by means of an algorithm based on gradient method 15 in conjunction
with Newton-Raphson method 16. The analysis of optimal trajectories is then carried out
considering several final altitudes of a clockwise or counterclockwise circular low Moon orbit.
All trajectories departure from a counterclockwise circular low Earth orbit corresponding
to the altitude of the Space Station. Maneuvers with direct ascent and multiple revolutions
around the Earth are considered in the analysis. The results for maneuvers with direct ascent
are compared to the ones obtained by Miele and Mancuso 12 who used the sequential
gradient-restoration algorithm for solving the optimization problem 15. The results for
maneuvers with multiple revolutions show that fuel can be saved if a lunar swing-by occurs.
2. Optimization Problem Based on Patched-Conic Approximation

In this section, the optimization problem based on the patched-conic approximation is
formulated. A detailed presentation of the patched-conic approximation can be found in
Bate et al. 1.
The following assumptions are employed.
1 The Earth is fixed in space.
2 The eccentricity of the Moon orbit around the Earth is neglected.
3 The flight of the space vehicle takes place in the Moon orbital plane.
4 The gravitational fields of Earth and Moon are central and obey the inverse square
law.
5 The trajectory has two distinct phases: geocentric and selenocentric trajectories. The
geocentric phase corresponds to the portion of the trajectory which begins at the
point of application of the first impulse and extends to the point of entering the
Moon’s sphere of
influence
r2
Moon
λ1 at time t1
r1
γ1 D
Earth
Moon
ϕ0 at time t0
γ0
r0
v0
Figure 1: Geometry of the geocentric phase.
Moon’s sphere of influence. The selenocentric phase corresponds to the portion of

trajectory in the Moon’s sphere of influence and ends at the point of application
of the second impulse. In each one of these phases, the space vehicle is under the
gravitational attraction of only one body, Earth or Moon.
6 The class of two impulse trajectories is considered. The impulses are applied
tangentially to the space vehicle velocity relative to Earth first impulse and Moon
second impulse.
An Earth-Moon trajectory is completely specified by four quantities: r0 —radius of
circular LEO; v0 —velocity of the space vehicle at the point of application of the first impulse
after the application of the impulse; ϕ0 —flight path angle at the point of application of the
first impulse; γ0 —phase angle at departure. These quantities must be determined such that
the space vehicle is injected into a circular LMO with specified altitude after the application of
the second impulse. It is particularly convenient to replace γ0 by the angle λ1 which specifies
the point at which the geocentric trajectory crosses the Moon’s sphere of influence.
Equations describing each phase of an Earth-Moon trajectory are briefly presented in
what follows. It is assumed that the geocentric trajectory is direct and that lunar arrival occurs
prior to apoapsis of the geocentric orbit. Figure 1 shows the geometry of the geocentric phase.
For a given set of initial conditions r0 , v0 , ϕ0 , energy and angular momentum of the
geocentric trajectory can be determined from the equations
1 2 μE
E v − ,
2 0 r0 2.1
h r0 v0 cos ϕ0 ,
where μE is the Earth gravitational parameter.

v1
ϕ2
λ1
r1 v2
r2
−vMoon
Moon
at time t1
Moon’s sphere of
influence
Figure 2: Geometry of the selenocentric phase.
From the geometry of the geocentric phase Figure 1, one finds

r1 D2 R2S − 2DRS cos λ1 ,

2.2
RS
sin γ1 sin λ1 ,
r1
where D is the distance from the Earth to the Moon and RS is the radius of the Moon’s sphere
of influence. Subscript 1 denotes quantities of the geocentric trajectory calculated at the edge
of the Moon’s sphere of influence.
From energy and angular momentum of the geocentric trajectory, one finds

μE
v1 2 E ,
r1
2.3
h
cos ϕ1 .
r1 v1
The selenocentric phase begins at the point at which the geocentric trajectory crosses
the Moon’s sphere of influence. Figure 2 shows the geometry of the selenocentric phase for a
clockwise arrival to LMO. Thus,
r2 RS , 2.4
v2 v1 − vM , 2.5
where vM is the velocity vector of the Moon relative to the center of the Earth. Subscript 2
denotes quantities of the selenocentric trajectory calculated at the edge of the Moon’s sphere
of influence.
From 2.5, one finds

v2 v12 vM
2
− 2v1 vM cos ϕ1 − γ1 ,
2.6
v2 sin ϕ1 − γ1
tan λ1 ± ϕ2 − .
vM − v2 cos ϕ1 − γ1
The upper sign refers to clockwise arrival to LMO and the lower sign refers to counter-
clockwise to LMO.
The semimajor axis af and eccentricity ef of the selenocentric trajectory are given by
r2
af ,
2 − Q2

2.7
ef 1 Q2 Q2 − 2cos2 ϕ2 ,
where Q2 r2 v22 /μM and μM is the Moon gravitational parameter.

The second impulse is applied at the periselenium of the selenocentric trajectory such
that the terminal conditions, before the impulse, are defined by

rpM af 1 − ef ,

μM 1 e f 2.8
vpM .
af 1 − ef
Equations 2.1–2.8 lead to the following two-point boundary value problem: for a
specified value of λ1 and a given set of initial parameters r0 and ϕ0 0◦ the impulse is
applied tangentially to the space vehicle velocity relative to Earth, determine v0 such that
the final condition rpM rf is satisfied, where r0 is the radius of LEO and rf is the radius of
LMO both orbits, LEO and LMO, are circular. This boundary value problem can be solved
by means of Newton-Raphson method 16.
After computing v0 , the velocity changes at each impulse can be determined
M
μE
Δv1 v0 − ,
r0

2.9
μM 1 e f μM
Δv2 − .
af 1 − ef rf
The total characteristic velocity is then given by
ΔvTotal Δv1 Δv2 . 2.10
Note that the total characteristic velocity is a function of λ1 for a given set of parameters
r0 , ϕ0 0◦ , rf . Accordingly, the following optimization problem can be formulated:
determine λ1 to minimize ΔvTotal . This minimization problem was solved using a classic
gradient method 15. The results are presented in Section 5.
The total flight time of an Earth-Moon trajectory is given by
T ΔtE ΔtM , 2.11
where ΔtE is the flight time of the geocentric trajectory and ΔtM is the flight time of the
selenocentric trajectory. These flight times are calculated from the well-known time of flight
equations of two-body dynamics as follows:

3
a0
ΔtE E1 − e0 sin E1 ,
μE
2.12

−a 3
f
ΔtM ef sinh F2 − F2 ,
μM
with eccentric anomaly E1 and hyperbolic eccentric anomaly F2 obtained, respectively, from

1 r1
cos E1 1− ,
e0 a0
2.13
1 r2
cosh F2 1− .
ef af
Since lunar arrival occurs prior to apoapsis of the geocentric trajectory, 0 < E1 ≤ 180◦ , and F2
is positive. So, equations above define E1 and F2 without ambiguity. The semimajor axis and
eccentricity of the geocentric trajectory are given by
r0
a0 ,
2 − Q0

2.14
e0 1 Q0 Q0 − 2cos2 ϕ0 ,
where Q0 r0 v02 /μE . Recall that the impulses are applied at the periapses of the geocentric
and selenocentric trajectories.
3. Optimization Problem Based on the Simplified Version of PCR3BP

In this section, the optimization problem based on the simplified PCR3BP is formulated. A
detailed presentation of this problem can be found in Miele and Mancuso 12.
The following assumptions are employed.
1 The Earth is fixed in space.
2 The eccentricity of the Moon orbit around the Earth is neglected.
3 The flight of the space vehicle takes place in the Moon orbital plane.
4 The space vehicle is subject to only the gravitational fields of Earth and Moon.
5 The gravitational fields of Earth and Moon are central and obey the inverse square
law.
6 The class of two impulse trajectories is considered. The impulses are applied
tangentially to the space vehicle velocity relative to Earth first impulse and Moon
second impulse.
Consider an inertial reference frame Exy contained in the Moon orbital plane: its origin
is the Earth center; the x-axis points towards the Moon position at the initial time t0 0 and
the y-axis is perpendicular to the x-axis. Figure 3 shows the inertial reference frame Exy.
In the Exy reference frame, the motion of the space vehicle P is described by the
following differential equations:
μE μM
ẍP − 3
xP − 3
xP − xM ,
rEP rMP
3.1
μE μM
ÿP − 3
yP − 3
yP − yM ,
rEP rMP
where rEP and rMP are, respectively, the radial distances of space vehicle from Earth E and
Moon M, that is, rEP2
xP −xE 2 yP −yE 2 and rMP
2
xP −xM 2 yP −yM 2 . Because the
origin of the inertial reference frame Exy is the Earth center, the position vector of the Earth
is defined by rE 0, 0. The position vector of the Moon in the inertial reference frame Exy is
defined by rM xM , yM . Since the eccentricity of the Moon orbit around Earth is neglected,
the Moon inertial coordinates are given by
xM t D cosωM t,

3.2
yM t D sinωM t,
where ωM μE /D3 is the angular velocity of the Moon.

The initial conditions of the system of differential equations correspond to the position
and velocity vectors of the space vehicle after the application of the first impulse. The initial
conditions t0 0 can be written as follows:
xP 0 xEP 0 rEP 0 cos θEP 0,
yP 0 yEP 0 rEP 0 sin θEP 0,

μE 3.3
ẋP 0 ẋEP 0 − Δv1 sin θEP 0,
rEP 0

μE
ẏP 0 ẏEP 0 Δv1 cos θEP 0,
rEP 0
Space vehicle
rEP rMP
θEP Moon
rM
x at time t0
Earth
Moon’s trajectory
Figure 3: Inertial reference frame Exy.
where Δv1 is the velocity change at the first impulse, rEP 0 rEP0 , and θEP t is the angle
defining the position of the space vehicle in the inertial reference frame Exy at time t, more
precisely the angle which the position vector rP forms with x-axis. It should be noted that
rEP 0 and vEP 0 or, equivalently, rP 0 and vP 0 are orthogonal, because the impulse is
applied tangentially to the circular LEO.
The final conditions of the system of differential equations correspond to the position
and velocity vectors of the space vehicle before the application of the second impulse. The
final conditions tf T can be written as follows:
xP T xMP T xM T rMP T cos θMP T xM T , 3.4
yP T yMP T yM T rMP T sin θMP T yM T , 3.5

μM
ẋP T ẋMP T ẋM T ± Δv2 sin θMP T ẋM T , 3.6
rMP T

μM
ẏP T ẏMP T ẏM T ∓ Δv2 cos θMP T ẏM T , 3.7
rMP T
where Δv2 is the velocity change at the second impulse, rMP T rMPf , and θMP t is the
angle which the position vector rMP forms with x-axis. The upper sign refers to clockwise
arrival to LMO and the lower sign refers to counterclockwise to LMO. Since the eccentricity
of the Moon orbit around Earth is neglected, it follows from 3.2 that the components of the
Moon inertial velocity at time T are given by
ẋM T −DωM sinωM T ,

3.8
ẏM T DωM cosωM T .
The angle θMP T is free and can be eliminated. After the problem has been solved,
the angle θMP T can be calculated from 3.4 and 3.5. So, combining 3.4–3.7, the final
conditions can be put in the following form:
2
xP T − xM T 2 yP T − yM T rMP T 2 , 3.9
2
2 2 μM
ẋP T − ẋM T ẏP T − ẏM T Δv2 , 3.10
rMP T

xP T − xM T ẏP T − ẏM T − yP T − yM T ẋP T − ẋM T
3.11
μM
∓ r MP T Δv2 .
rMP T
As before, the upper sign refers to clockwise arrival to LMO and the lower sign refers to
counterclockwise to LMO. It should be noted that constraint defined by 3.11 is derived
from the angular momentum considering a direct counterclockwise arrival or a retrograde
clockwise arrival orbit around the Moon.
The problem defined by 3.1–3.11 involves four unknowns Δv1 , Δv2 , T , and θEP 0
that must be determined in order to satisfy the three final conditions 3.9–3.11. Since
this problem has one degree of freedom, an optimization problem can be formulated as
follows: determine Δv1 , Δv2 , T , and θEP 0 which satisfy the final constraints 3.9–3.11
and minimize the total characteristic velocity ΔvTotal Δv1 Δv2 . This problem was
solved by Miele and Mancuso 12 using the sequential gradient-restoration algorithm for
mathematical programming problems developed by Miele et al. 15.
In this paper, the optimization problem described above is solved by means of an
algorithm based on gradient method 15 in conjunction with Newton-Raphson method
16, similarly to the one described in the previous section for the problem based on the
patched-conic approximation. The angle θEP 0 has been chosen as the iteration variable in
the gradient phase with Δv1 , Δv2 , and T calculated through Newton-Raphson method. The
results are presented in Section 5.
4. Optimization Problem Based on the Classical Version of PCR3BP

In this section, the optimization problem based on the classical version of PCR3BP is
formulated. The assumptions employed in this formulation are the same ones previously
presented in Section 3, except for assumption 1 which must be replaced by the following
one: Earth moves around the center of mass of the Earth-Moon system.
4.1. Problem Formulation in Inertial Frame

Consider an inertial reference frame Gxy contained in the Moon orbital plane: its origin is the
center of mass of the Earth-Moon system; the x-axis points towards the Moon position at the
initial time and the y-axis is perpendicular to the x-axis. Figure 4 shows the inertial reference
frame Gxy.
Space vehicle
Earth’s trajectory
rE P rMP
rP
rE θEP
rM Moon
x at time t0
Center of mass
Earth
Moon’s trajectory
Figure 4: Inertial reference frame Gxy.
In the Gxy reference frame, the motion of the space vehicle P is described by the
following differential equations:
μE μM
ẍP − 3
xP − xE − 3
xP − xM ,
rEP rMP
4.1
μE μM
ÿP − 3
yP − yE − 3
yP − yM ,
rEP rMP
where rEP and rMP are, respectively, the radial distances of space vehicle from Earth E and
Moon M, that is, rEP
2
xP − xE 2 yP − yE 2 and rMP
2
xP − xM 2 yP − yM 2 . Because
the origin of the inertial reference frame Gxy is the center of mass of Earth-Moon system,
the position vectors of the Earth and the Moon are, respectively, defined by rE xE , yE and
rM xM , yM . Since the eccentricity of the Moon orbit around Earth is neglected, the Earth
and Moon inertial coordinates are given by
xE t −μxM t,
yE t −μyM t,
D 4.2
xM t cosωt,
1 μ
D
yM t sinωt,
1 μ

D
where μ μM /μE and ω μE μM /D3 . Note that ω ωM 1 μ.
The initial conditions of the system of differential equations correspond to the position
and velocity vectors of the space vehicle after the application of the first impulse. The initial
conditions t0 0 can be written as follows:
xP 0 xEP 0 xE 0 rEP 0 cos θEP 0 xE 0,
yP 0 yEP 0 yE 0 rEP 0 sin θEP 0 yE 0,

μE 4.3
ẋP 0 ẋEP 0 ẋE 0 − Δv1 sin θEP 0 ẋE 0,
rEP 0

μE
ẏP 0 ẏEP 0 ẏE 0 Δv1 cos θEP 0 ẏE 0,
rEP 0
where Δv1 , rEP 0, and θEP t have the same meaning previously defined in Section 3 and,
from 4.2,
μD μDω
xE 0 − , yE 0 0, ẋE 0 0, ẏE 0 − . 4.4
1 μ 1 μ
It should be noted that rEP and vEP are orthogonal because the impulse is applied tangentially
to LEO, assumed circular.
The final conditions of the system of differential equations correspond to the position
and velocity vectors of the space vehicle before the application of the second impulse and
they are given by 3.4–3.7, with the final position and velocity vectors of Moon obtained
from 4.2, that is, given by
D D
xM T cosωT , yM T sinωT ,
1 μ 1 μ
4.5
Dω Dω
ẋM T − sinωT , ẏM T cosωT .
1 μ 1 μ
Accordingly, the final conditions can be put in the same form defined by 3.9–3.11.
Therefore, the optimization problem is the same one defined in Section 3, and it is
solved by the same algorithm previously described. The results are presented in Section 5.
4.2. Transformation to Rotating Frame

Consider a rotating reference frame Gξη contained in the Moon orbital plane: its origin is the
center of mass of the Earth-Moon; the ξ-axis points towards the Moon position at any time
t and the η-axis is perpendicular to the ξ-axis. In this rotating reference frame the Earth and
the Moon are at rest. Figure 5 shows the inertial and rotating reference frames, Gxy and Gξη.
To write the differential equations of motion of the space vehicle P in the rotating reference
frame, consider the coordinate transformation equations:
xk ξk cosωt − ηk sinωt,
4.6
yk ξk sinωt ηk cosωt,
or
ξk xk cosωt yk sinωt,
4.7
ηk −xk sinωt yk cosωt,
where k E, M, P . Thus, the new coordinates of the Earth and Moon are
ξE xE cosωt yE sinωt,
ηE −xE sinωt yE cosωt,

4.8
ξM xM cosωt yM sinωt,
ηM −xM sinωt yM cosωt.
Substituting 4.2 into 4.8, one finds the fixed positions of the Earth and the Moon in
the rotating reference frame:
μD
ξE − ,
1 μ
ηE 0,
4.9
D
ξM ,
1 μ
ηM 0.
Now, consider the inertial coordinates of the space vehicle written in terms of the rotating
coordinates
xP ξP cosωt − ηP sinωt,
4.10
yP ξP sinωt ηP cosωt.
Differentiating each of these equations twice and substituting into 4.1, one finds the new
equations of motion:
μE μM
ξ̈P − 2ωη̇P − ω2 ξP − 3
ξP − ξE − 3
ξP − ξM ,
rEP rMP
4.11
μE μM
η̈P 2ωξ̇P − ω2 ηP − 3
ηP − ηE − 3 ηP − ηM ,
rEP rMP
y
η
ωt
Moon
rM at time t
ξ
ωt
Center of mass x
Moon’s trajectory
Figure 5: Rotating reference frame Gxy.
where
2
2
rEP ξP − ξE 2 ηP − ηE ,
2 4.12
2
rMP ξP − ξM 2 ηP − ηM .
The system of differential equations above has a constant of motion, the so-called Jacobi
integral. In order to determine it, procede as follows. Multiply the first of 4.11 by ξ̇P , the
second by η̇P and add. It results in
μE μM
ξ̇P ξ̈P η̇P η̈P − ω2 ξ̇P ξP η̇P ηP − 3 ξ̇P ξP − ξE − 3 ξ̇P ξP − ξM
rEP rMP
4.13
μE μM
− 3
η̇P ηP − ηE − 3
η̇P ηP − ηM .
rEP rMP
This equation can be rewritten as
1 d 2 1 d 2 d μE μM

ξ̇P η̇P2 − ω2 ξP ηP2 . 4.14
2 dt 2 dt dt rEP rMP
Therefore,

v2 2Φ ξP , ηP − C, 4.15
Table 1: Lagrange points for the Earth-Moon system.
ξL k ηLk Ck
km km km2 /s2
L1 3.2171 × 105 0 3.3468
L2 4.4424 × 105 0 3.3298
L3 −3.8635 × 105 0 3.1618
L4 1.8753 × 105 3.3290 × 105 3.1365
L5 1.8753 × 105 −3.3290 × 105 3.1365
where
v2 ξ̇P2 η̇P2 ,
1 μE μM 4.16
Φ ξP , ηP ω2 ξP2 ηP2 ,
2 rEP rMP
and C is the so-called Jacobi constant. Therefore, from 4.15 it is seen that the Jacobi integral,

J ξP , ηP , ξ̇P , η̇P 2Φ ξP , ηP − v2 , 4.17
is equal to C during the motion 17.

The system of differential equations 4.11 has five equilibrium points. They are called
Lagrange points and are labelled L1 , . . . , L5 . The points L1 , L2 , L3 are located on the ξ-axis while
L4 and L5 form two equilateral triangles with the Earth and Moon in the plane of rotation.
See Figure 6. At each Lk , the corresponding value of Jacobi constant Ck is given by 4.15
substituting rLk ξLk , ηLk and vLk ξ̇Lk , η̇Lk 0, 0. These Ck are related to the regions in
the rotating reference frame Gξη where the spacecraft can move. See in Table 1 the position
and the correspondent value of the Jacobi constant per unit mass of each Lagrange point.
Observe that the right-hand side of the 4.15 must be nonnegative, since v2 ≥ 0. Thus,
an initial position ξP 0 , ηP 0 and an initial velocity ξ̇P 0 , η̇P 0 yield a Jacobi constant value C
and the motion of the spacecraft is possible only in positions satisfying the relation

2Φ ξP , ηP ≥ C. 4.18
The set of points in the ξ, η-plane defined by the inequality 4.18 is called Hill’s regions. See
in Figure 6 how the Hill’s regions white areas are related to the Ck values. The shaded areas
are the forbidden regions.
Finally, we note that the transformation to the rotating frame gives a better insight
about swing-by maneuvers with the Moon, as described in the results presented in Section 5.
5. Results
In this section, results are presented for some lunar missions using the three formulations
described in the preceding sections. Analysis of the results is discussed in two subsections:
in the first one, direct ascent maneuvers with flight time about 4.7 days are considered; in
5 L4
4 L4
3
2
Moon
η (105 km)
Earth
1
Earth Moon L3 L1 L2
0 L3 L2
L1 P
−1 P
−2
−3 L5
−4 L5
−5
−5 −4 −3 −2 −1 0 1 2 3 4 5
ξ (105 km)
a Lagrange points b C1 < C
L4 L4
Earth Moon Earth Moon

L3 L1 L2 L3 L1 L2
P P
L5 L5
c C2 < C < C1 d C3 < C < C2
L4 L4
Earth Moon Earth Moon

L3 L1 L2 L3 L1 L2
P P
L5 L5
e C4 < C < C3 f C < C4
Figure 6: Lagrange points and Hill’s regions for the Earth-Moon system.
Table 2: Lunar mission, counterclockwise LMO arrival, major parameters.
LMO altitude ΔvTotal Δv1 Δv2 T θEP 0

Model Feasibility
km km/s km/s km/s days deg
Patched-conic 3.8482 3.0655 0.7827 4.794 −115.723 Yes
100 Simplified PCR3BP 3.8758 3.0649 0.8109 4.564 −116.800 Yes
Classical PCR3BP 3.8777 3.0658 0.8119 4.573 −116.410 Yes
Miele and Mancuso 12 3.876 3.065 0.811 4.37 −118.98 —
the second subsection, some maneuvers with multiple revolutions and flight time about
14, 24, 32, 40, and 58 days are considered. Three final altitudes hLMO of a clockwise or
counterclockwise circular LMO and a specified altitude hLEO of a counterclockwise circular
LEO, which corresponds to the altitude of the Space Station 12, are considered. The
following data are used:

G 6.672 × 10−20 km3 kg−1 s−2 universal constant of gravitation ,
ME 5.9742 × 1024 kg mass of the Earth,
MM 7.3483 × 1022 kg mass of the Moon,
REM 384 400 km mean distance from the Earth to the Moon,
5.1
RE 6 378 km Earth radius,
RM 1 738 km Moon radius,
hLEO 463 km altitude of circular LEO,
hLMO 100, 200, 300 km altitude of circular LMO.
5.1. Direct Ascent Maneuvers

Table 2 shows the results for lunar missions with counterclockwise arrival at LMO, and
Table 3 shows the results for lunar missions with clockwise arrival at LMO. Recall that the
departure from LEO is counterclockwise for all missions. The major parameters that are
presented in these tables are the velocity changes Δv1 and Δv2 at each impulse, the total
characteristic velocity ΔvTotal Δv1 Δv2 , the flight time of lunar mission T , and the angular
position of the space vehicle with respect to Earth at the initial time defined by the angle
θEP 0.
Table 3: Lunar mission, clockwise LMO arrival, major parameters.
LMO altitude ΔvTotal Δv1 Δv2 T θEP 0

Model Feasibility
km km/s km/s km/s days deg
Simplified PCR3BP 3.8811 3.0677 0.8134 4.750 −114.215 Yes
100
Results in Tables 2 and 3 show good agreement. It should be noted the excellent results
were obtained using the patched-conic approximation model. In all missions, the patched-
conic approximation model yields very accurate estimate for the first impulse in comparison
to the results obtained using the PCR3BP models. For the second impulse, there exists a
small difference between the results given by the patched-conic approximation model and
the PCR3BP models. For all lunar missions, the values of the major parameters ΔvTotal , Δv1 ,
Δv2 , and T obtained using the simplified PCR3BP model are a little lesser than the values
obtained using the classical PCR3BP. In all cases, the trajectories are feasible, that is, the
spacecraft does not collide with the Moon. As described in the next subsection, collisions
can occur for maneuvers with multiple revolutions.
Tables also show a small difference in the flight time T and in the angle θEP 0
calculated by the three approaches. We suppose that the difference between the values
obtained in this paper and the values presented by 12 for the flight time T and the angle
θEP 0 calculated using the simplified PCR3BP model should be related to the accuracy
in the integration of differential equations and in the solution of the terminal constraints.
The algorithm based on gradient algorithm in conjunction with Newton-Raphson method,
described in this paper, uses a Runge-Kutta-Fehlberg method of orders 4 and 5, with step-size
control and relative error tolerance of 10−10 and absolute error tolerance of 10−11 , as described
in Stoer and Bulirsch 16 and Forsythe et al. 18. The terminal constraints are satisfied with
an error lesser than 10−8 . In all

simulations the following canonical units are used: 1 distance
unit RE and 1 time unit R3E /μE . On the other hand, the paper by Miele and Mancuso
12 does not describe the accuracy used in the calculations.
According to the results presented in Tables 2 and 3, we note, regardless the dynamical
model used in the analysis, that
1 lunar missions with clockwise LMO arrival spend more fuel than lunar missions
with counterclockwise LMO arrival;
2 the flight time is nearly the same for all lunar missions with clockwise LMO arrival,
independently on the final altitude of LMO. The differences between the flight
Table 4: Lunar mission, major parameters, hLEO 463 km, hLMO 100 km.
ΔvTotal Δv1 Δv2 T θEP 0

Model Maneuver Feasibility
km/s km/s km/s days deg
3.8758 3.0649 0.8109 4.564 −116.800 Yes
3.8778 3.0653 0.8125 14.175 9.119 Yes
Simplified Counterclockwise 3.8746 3.0647 0.8099 23.832 135.661 Yes
PCR3BP 3.8411 3.0585 0.7826 32.108 232.269 Yes
3.8444 3.0591 0.7853 40.871 347.205 Yes
3.7936 3.0498 0.7438 58.701 220.730 Yes
3.8811 3.0677 0.8134 4.750 −114.215 Yes
3.8829 3.0681 0.8149 14.791 17.307 Yes
Simplified Clockwise 3.8785 3.0672 0.8113 24.916 149.879 Yes
PCR3BP 3.8320 3.0586 0.7734 32.087 232.573 Yes
3.8352 3.0592 0.7760 40.862 347.608 Yes
3.7842 3.0498 0.7444 58.551 220.998 Yes
time of each mission are small, they are approximately lesser tan 0.004 days 5.8
minutes;
3 the flight time is nearly the same for all lunar missions with counterclockwise LMO
arrival, independently on the final altitude of LMO. The differences between the
flight time of each mission are small; they are approximately lesser than 0.010 days
14.4 minutes;
4 the first change velocity Δv1 is nearly independent of the LMO altitude;
5 the second change velocity Δv2 decreases with the LMO altitude;
6 the flight time for lunar missions with clockwise LMO arrival is larger than the
flight time for lunar missions with counterclockwise LMO arrival;
7 for the PCR3BP and patched-conic approximation models, the angle θEP 0 varies
with the LMO altitude for all lunar missions.
We note that some of these general results are quite similar to the ones described by 12.
For hLMO 100 km, the trajectory is shown in Figure 7 for counterclockwise LMO
arrival and in Figure 8 for clockwise LMO arrival. In both figures, trajectories are shown in
the inertial reference frame Exy and in the rotating reference frame Gξη. Only results obtained
through the classical PCR3BP model are depicted.
5.2. Multiple Revolution Ascent Maneuvers

Tables 4, 5, and 6 show the results for lunar missions with clockwise and counterclockwise
arrival at LMO, for hLMO 100, 200, 300 km, respectively, considering the simplified PCR3BP
model. Tables 7, 8, and 9 show similar results considering the classical PCR3BP model. The
major parameters that are presented in these tables are the same ones presented in Tables 3
and 2. The value of the Jacobi constant per unit mass for each mission is presented in Tables
7, 8, and 9.
For hLMO 100 km, the trajectories are shown in Figures 9 to 13 for counterclockwise
LMO arrival and in Figures 14 to 18 for clockwise LMO arrival, considering the classical
0.3
4
3.5 0.2
3
0.1
2.5
y (105 km)
y (105 km)
2 0
1.5
1 −0.1
0.5
−0.2
0
−0.5 −0.3
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 −0.2 −0.1 0 0.1 0.2 0.3
x (105 km) x (105 km)
a Earth-Moon trajectory, inertial coordinate b LEO departure, inertial coordinate frame
frame
5
3.4 4
3
3.38
2
3.36
y (105 km)
η (105 km)
1
3.34 0
3.32 −1
−2
3.3
−3
3.28 −4
−5
1.84 1.86 1.88 1.9 1.92 1.94 1.96 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
c LMO arrival, inertial coordinate frame d Earth-Moon trajectory, rotating coordinate
frame
0.6
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.05
−0.2
−0.4 −0.1
−0.6 −0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9 3.95
ξ (105 km) ξ (105 km)
e LEO departure, rotating coordinate frame f LMO arrival, rotating coordinate frame
Figure 7: Direct ascent, counterclockwise LMO arrival, Δv1 3.0658 km/s, Δv2 0.8119 km/s, T 4.573
days, θEP 0 −116.41 deg.
5 0.6
4
3 0.4
2
0.2
y (105 km)
y (105 km)
1
0 0
−1
−0.2
−2
−3 −0.4
−4
−5 −0.6
−5 −4 −3 −2 −1 0 1 2 3 4 5 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
x (105 km) x (105 km)
frame
5
3.55 4
3
3.5 2
y (105 km)
η (105 km)
1
3.45
0
3.4 −1
−2
3.35
−3
3.3 −4
−5
1.6 1.65 1.7 1.75 1.8 1.85 1.9 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6 0.15
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.2 −0.05
−0.4 −0.1
−0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9
ξ (105 km) ξ (105 km)
Figure 8: Direct ascent, clockwise LMO arrival, Δv1 3.0686 km/s, Δv2 0.8143 km/s, T 4.763 days,
θEP 0 −113.795 deg.

3.8614 3.0648 0.7966 4.562 −116.832 Yes
3.8635 3.0653 0.7983 14.168 9.002 Yes
PCR3BP 3.8263 3.0585 0.7678 32.110 232.270 Yes
3.8296 3.0591 0.7705 40.873 347.204 Yes
3.7826 3.0506 0.7320 58.398 216.284 Yes
3.8670 3.0677 0.7993 4.753 −114.187 Yes
3.8688 3.0681 0.8007 14.798 17.390 Yes
PCR3BP 3.8167 3.0586 0.7581 32.080 232.475 Yes
3.8200 3.0592 0.7608 40.863 347.613 Yes
3.7679 3.0498 0.7181 58.530 220.726 Yes

3.8483 3.0648 0.7835 4.560 −116.881 Yes
3.8504 3.0652 0.7852 14.157 8.851 Yes
PCR3BP 3.8127 3.0585 0.7542 32.111 232.260 Yes
3.8161 3.0591 0.7570 40.874 347.201 Yes
3.7633 3.0498 0.7135 58.685 220.409 Yes
3.8541 3.0678 0.7863 4.760 −114.110 Yes
3.8559 3.0681 0.7878 14.809 17.504 Yes
PCR3BP 3.8026 3.0586 0.7440 32.080 232.475 Yes
3.8060 3.0592 0.7467 40.864 347.620 Yes
3.7530 3.0498 0.7031 58.488 220.014 Yes
PCR3BP model. In both figures, trajectories are shown in the inertial reference frame Exy and
in the rotating reference frame Gξη.
For maneuvers with multiple revolutions, major comments are as follows.
1 All trajectories, regardless of the flight time, are feasible in the simplified PCR3BP
model.
2 In the classical PCR3BP model, all trajectories have Jacobi constant less than C4 .
Therefore, there are not forbidden regions for the motion of the spacecraft.
3 Trajectories with two revolutions around the Earth and flight time about 24.0
days collide with the Earth in the classical PCR3BP model. Figures 10 and 15
ΔvTotal Δv1 Δv2 T θEP 0 C

km/s km/s km/s days deg km2 /s2
3.8777 3.0658 0.8119 4.573 −116.410 2.4784 Yes
3.8732 3.0650 0.8082 14.330 12.466 2.4965 Yes
Classical Counterclockwise 3.8776 3.0658 0.8118 24.019 140.054 2.4788 No
PCR3BP 3.8428 3.0593 0.7835 31.910 232.464 2.6166 Yes
3.8379 3.0584 0.7795 40.742 348.882 2.6358 Yes
3.8300 3.0570 0.7730 58.415 229.239 2.6668 Yes
3.8829 3.0686 0.8143 4.763 −113.795 2.4187 Yes
3.8785 3.0678 0.8107 14.970 21.069 2.4363 Yes
Classical Clockwise 3.8821 3.0684 0.8136 25.103 154.315 2.4220 No
PCR3BP 3.8337 3.0595 0.7742 31.872 232.674 2.6138 Yes
3.8288 3.0586 0.7702 40.713 349.322 2.6331 Yes
3.7893 3.0513 0.7380 58.420 224.480 2.7871 Yes

3.8634 3.0658 0.7976 4.571 −116.451 2.4793 Yes
3.8587 3.0649 0.7938 14.322 12.340 2.4974 Yes
PCR3BP 3.8280 3.0593 0.7686 31.912 232.461 2.6166 Yes
3.8230 3.0584 0.7645 40.743 348.876 2.6358 Yes
3.8149 3.0570 0.7579 58.417 229.218 2.6669 Yes
3.8688 3.0686 0.8002 4.769 −113.742 2.4178 Yes
3.8643 3.0678 0.7965 14.981 21.198 2.4355 Yes
PCR3BP 3.8184 3.0595 0.7589 31.873 232.676 2.6137 Yes
3.8195 3.0597 0.7598 39.664 353.119 2.6094 Yes
3.7709 3.0509 0.7199 58.361 223.886 2.7957 Yes
depict the collision between the spacecraft and the Earth for lunar missions with
hLMO 100 km and counterclockwise or clockwise arrival to Moon, respectively.
The remaining trajectories are feasible.
4 According to the results in Tables 4–9, simplified and classical PCR3BP models
show that the group of trajectories with counterclockwise arrival to Moon is slightly
superior to the group of trajectories with clockwise arrival to Moon in terms of total
characteristic velocity and flight time for maneuvers with flight time smaller than
25.5 days.
5 According to the results in Tables 4–9, simplified and classical PCR3BP models
show that the group of trajectories with clockwise arrival to Moon is slightly

3.8502 3.0657 0.7845 4.569 −116.491 2.4802 Yes
3.8455 3.0649 0.7806 14.313 12.213 2.4983 Yes
PCR3BP 3.8144 3.0593 0.7551 31.913 232.461 2.6166 Yes
3.8093 3.0584 0.7509 40.745 348.870 2.6359 Yes
3.8010 3.0570 0.7441 58.422 229.235 2.6671 Yes
3.8559 3.0687 0.7872 4.771 −113.716 2.4170 Yes
3.8513 3.0678 0.7834 14.992 21.336 2.4346 Yes
PCR3BP 3.8043 3.0595 0.7448 31.873 232.676 2.6137 Yes
3.8054 3.0597 0.7458 39.665 353.111 2.6094 Yes
3.7559 3.0509 0.7050 58.361 223.913 2.7957 Yes
superior to the group of trajectories with counterclockwise arrival to Moon in terms

of total characteristic velocity and flight time excepting trajectories with flight time
about 58.5 days for maneuvers with flight time larger than 30.0 days.
6 For lunar missions with hLMO 100 km, Figures 11, 12, 13, 16, 17, and, 18 show that
the spacecraft carries out one or two close approaches to the Moon before entering
into the low circular orbit around the Moon. Such maneuvers, known as swing-by
maneuvers, reduce the fuel consumption, although the flight time increases see
results in Table 7.
7 According to the results in Tables 4–9, the first change velocity Δv1 is nearly
independent of the LMO altitude, and, the second change velocity Δv2 decreases as
the LMO altitude increases, for all group of trajectories with similar flight time and
same sense of arrival to Moon.
8 The first change velocity Δv1 is nearly the same for maneuvers with flight time
smaller than 25.5 days the differences in Δv1 are approximately 0.6 m/s.
9 For maneuvers with flight time larger than 30.0 days, the first change velocity Δv1
changes slightly the differences in Δv1 are approximately 15.0 to 20.0 m/s.
6. Conclusion
In this paper, a systematic study of optimal trajectories for Earth-Moon flight of a space
vehicle is presented. The optimization criterion is the total characteristic velocity. The
optimization problem has been formulated using the patched-conic approximation and
two versions of the planar circular restricted three-body problem PCR3BP and has been
solved using a gradient algorithm in conjunction with Newton-Raphson method. Results for
direct ascent maneuvers and for maneuvers with multiple revolutions around the Earth are
presented. For direct ascent maneuvers, all models show that lunar missions with clockwise
5 0.3
4
0.2
3
y (105 km) 2 0.1
y (105 km)
1
0 0
−1
−0.1
−2
−3 −0.2
−4
−0.3
−5
−5 −4 −3 −2 −1 0 1 2 3 4 5 −0.3 −0.2 −0.1 0 0.1 0.2
x (105 km) x (105 km)
frame
−0.45 5
4
−0.5
3
−0.55 2
y (105 km)
η (105 km)
1
−0.6 0
−1
−0.65
−2
−0.7 −3
−4
−0.75
−5
−3.95 −3.9 −3.85 −3.8 −3.75 −3.7 −3.65 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6 0.15
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.2 −0.05
−0.4 −0.1
−0.6 −0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9 3.95
ξ (105 km) ξ (105 km)
Figure 9: Multiple revolution ascent, counterclockwise LMO arrival, Δv1 3.065 km/s, Δv2
0.8082 km/s, T 14.33 days, θEP 0 12.466 deg.
0.3
1
0.5 0.2
0
0.1
−0.5
y (105 km)
y (105 km)
−1 0
−1.5
−2 −0.1
−2.5
−0.2
−3
−3.5 −0.3
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
x (105 km) x (105 km)
frame
5
−2.5
4
−2.55 3
2
−2.6
y (105 km)
η (105 km)
1
0
−2.65
−1
−2.7 −2
−3
−2.75
−4
−5
2.65 2.7 2.75 2.8 2.85 2.9 2.95 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.3 0.15
0.2 0.1
0.1 0.05
η (105 km)
η (105 km)
0 0
−0.1 −0.05
−0.2 −0.1
−0.3 −0.15
−0.3 −0.2 −0.1 0 0.1 0.2 3.65 3.7 3.75 3.8 3.85 3.9
ξ (105 km) ξ (105 km)
0.8118 km/s, T 24.019 days, θEP 0 140.054 deg.
0.6
4
3.5 0.4
3
0.2
2.5
y (105 km)
y (105 km)
2 0
1.5
1 −0.2
0.5
−0.4
0
−0.5 −0.6
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
x (105 km) x (105 km)
frame
5
3.5
4
3.45 3
2
y (105 km)
3.4
η (105 km)
1
0
3.35
−1
3.3 −2
−3
3.25
−4
−5
1.75 1.8 1.85 1.9 1.95 2 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6 0.15
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.2 −0.05
−0.4 −0.1
−0.6 −0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9
ξ (105 km) ξ (105 km)
0.7835 km/s, T 31.91 days, θEP 0 232.464 deg.
2.5
2 0.2
1.5
0.1
y (105 km) 1
y (105 km)
0.5 0
0
−0.5 −0.1
−1
−0.2
−1.5
−2 −0.3
−4 −3 −2 −1 0 −0.3 −0.2 −0.1 0 0.1 0.2 0.3

x (105 km) x (105 km)
frame
5
0.3
4
0.25 3
2
y (105 km)
0.2
η (105 km)
1
0
0.15
−1
0.1 −2
−3
0.05
−4
−5
−4 −3.95 −3.9 −3.85 −3.8 −3.75 −3.7 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6 0.15
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.2 −0.05
−0.4 −0.1
−0.6 −0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9
ξ (105 km) ξ (105 km)
0.7795 km/s, T 40.742 days, θEP 0 348.882 deg.
0.6
3.5
3 0.4
2.5
0.2
y (105 km)
y (105 km)
2
1.5 0
1
0.5 −0.2
0
−0.4
−0.5
−1 −0.6
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
x (105 km) x (105 km)
frame
5
3.1 4
3
3.05 2
y (105 km)
η (105 km)
1
3
0
2.95 −1
−2
2.9
−3
2.85 −4
−5
2.3 2.35 2.4 2.45 2.5 2.55 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6 0.15
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.2 −0.05
−0.4 −0.1
−0.6 −0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9 3.95
ξ (105 km) ξ (105 km)
0.7730 km/s, T 58.415 days, θEP 0 229.239 deg.
0.6
1.5
1 0.4
0.5
y (105 km) 0.2
y (105 km)
0
−0.5 0
−1
−1.5 −0.2
−2
−0.4
−2.5
−3 −0.6
−4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
x (105 km) x (105 km)
frame
5
4
−1.05
3
−1.1 2
y (105 km)
η (105 km)
1
−1.15
0
−1.2 −1
−2
−1.25 −3
−4
−1.3
−5
−3.8 −3.75 −3.7 −3.65 −3.6 −3.55 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6
0.15
0.4
0.1
0.2
η (105 km)
η (105 km)
0.05
0
0
−0.2
−0.05
−0.4
−0.1
−0.6
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9 3.95
ξ (105 km) ξ (105 km)
Figure 14: Multiple revolution ascent, clockwise LMO arrival, Δv1 3.0678 km/s, Δv2 0.8107 km/s,
T 14.97 days, θEP 0 21.069 deg.
0.3
1
0.5 0.2
0
0.1
y (105 km)
−0.5
y (105 km)
−1 0
−1.5
−0.1
−2
−2.5 −0.2
−3
−3.5 −0.3
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
x (105 km) x (105 km)
frame
5
4
−1.75
3
−1.8 2
y (105 km)
η (105 km)
1
−1.85 0
−1
−1.9
−2
−1.95 −3
−4
−2
−5
3.25 3.3 3.35 3.4 3.45 3.5 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.15
0.3
0.1
0.2
0.1 0.05
η (105 km)
η (105 km)
0 0
−0.05
−0.1
−0.2 −0.1
−0.3 −0.15
−0.3 −0.2 −0.1 0 0.1 0.2 3.65 3.7 3.75 3.8 3.85 3.9
ξ (105 km) ξ (105 km)
T 25.103 days, θEP 0 154.315 deg.
0.6
4
3.5 0.4
3
y (105 km)
2.5 0.2
y (105 km)
2
0
1.5
1 −0.2
0.5
−0.4
0
−0.5 −0.6
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
x (105 km) x (105 km)
frame
5
4
3.45
3
3.4 2
y (105 km)
η (105 km)
1
3.35 0
−1
3.3
−2
3.25 −3
−4
3.2
−5
1.75 1.8 1.85 1.9 1.95 2 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6 0.15
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.2 −0.05
−0.4 −0.1
−0.6 −0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9 3.95
ξ (105 km) ξ (105 km)
T 31.872 days, θEP 0 232.674 deg.
2.5 0.6
2
0.4
1.5
1 0.2
y (105 km)
y (105 km)
0.5
0 0
−0.5
−0.2
−1
−1.5 −0.4
−2
−0.6
−4.5 −4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
x (105 km) x (105 km)
frame
4
0.3
3
0.25 2
y (105 km)
η (105 km)
1
0.2
0
0.15 −1
−2
0.1 −3
−4
0.05
−5
−4 −3.95 −3.9 −3.85 −3.8 −3.75 −3.7 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6 0.15
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.2 −0.05
−0.4 −0.1
−0.6 −0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9
ξ (105 km) ξ (105 km)
T 40.713 days, θEP 0 349.322 deg.
0.6
4
3.5
0.4
3
2.5 0.2
y (105 km)
y (105 km)
2
0
1.5
1 −0.2
0.5
−0.4
0
−0.5 −0.6
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
x (105 km) x (105 km)
frame
5
3.1 4
3
3.05
2
y (105 km)
η (105 km)
3 1
0
2.95 −1
−2
2.9
−3
2.85 −4
−5
2.3 2.35 2.4 2.45 2.5 2.55 −5 −4 −3 −2 −1 0 1 2 3 4 5
x (105 km) ξ (105 km)
frame
0.6 0.15
0.4 0.1
0.2 0.05
η (105 km)
η (105 km)
0 0
−0.2 −0.05
−0.4 −0.1
−0.6 −0.15
−0.6 −0.4 −0.2 0 0.2 0.4 3.65 3.7 3.75 3.8 3.85 3.9 3.95
ξ (105 km) ξ (105 km)
T 58.42 days, θEP 0 224.48 deg.
LMO arrival spend more fuel than lunar missions with counterclockwise LMO arrival. For
maneuvers with multiple revolutions, fuel can be saved if the spacecraft accomplishes a
swing-by maneuver with the Moon before the arrival at LMO.
Acknowledgments
This research was supported by CNPq under contract 302949/2009-7 and by FAPESP under
contract 2008/56241-8.
References
1 R. R. Bate, D. D. Mueller, and J. E. White, Fundamentals of Astrodynamics, Dover, 1971.
2 A. E. Roy, Orbital Motion, Taylor & Francis, 4th edition, 2004.
3 V. Szebehely, Theory of Orbits, Academic Press, New York, NY, USA, 1967.
4 W. S. Koon, M. W. Lo, J. E. Marsden, and S. D. Ross, Dynamical Systems, the Three-Body Problem and
Space Mission Design, Springer, 2007.
5 E. A. Belbruno and J. Miller, “Sun-perturbed earth-to-moon transfers with ballistic capture,” Journal
of Guidance, Control, and Dynamics, vol. 16, no. 4, pp. 770–775, 1993.
6 E. A. Belbruno and J. P. Carrico, “Calculation of weak stability boundary ballistic lunar transfer
trajectories,” in Proceedings of the AAS/AIAA Astrodynamics Specialist Conference, AIAA, Denver, Colo,
USA, August 2000.
7 W. S. Koon, M. W. Lo, J. E. Marsden, and S. D. Ross, “Low energy transfer to the moon,” Celestial
Mechanics & Dynamical Astronomy, vol. 81, no. 1-2, pp. 63–73, 2001.
8 K. Yagasaki, “Computation of low energy Earth-to-Moon transfer with moderate flight time,” Physica
D, vol. 197, no. 3-4, pp. 313–331, 2004.
9 K. Yagasaki, “Sun-perturbed Earth-to-Moon transfers with low energy and moderate flight time,”
Celestial Mechanics & Dynamical Astronomy, vol. 90, no. 3-4, pp. 197–212, 2004.
10 E. A. Belbruno, “Lunar capture orbits, a method of constructing earth-moon trajectories and lunar gas
mission,” in Proceedings of 19th AIAA/DGLR/JSASS International Electric Propulsion Conference, AIAA,
Colorado Springs, Colo, USA, May 1987, number 87–1054.
11 E. Belbruno, Capture Dynamics and Chaotic Motions in Celestial Mechanics, Princeton University Press,
Princeton, NJ, USA, 2004.
12 A. Miele and S. Mancuso, “Optimal trajectories for Earth-Moon-Earth flight,” Acta Astronautica, vol.
49, pp. 59–71, 2001.
13 S. A. Fazelzadeh and G. A. Varzandian, “Minimum-time Earth-Moon and Moon-Earth orbital
maneuvers using time-domain finite element method,” Acta Astronautica, vol. 66, pp. 528–538, 2010.
14 J. P. Marec, Optimal Space Trajectories, Elsivier, 1979.
15 A. Miele, H. Y. Huang, and J. C. Heideman, “Sequential gradient-restoration algorithm for the
minimization of constrained functions—ordinary and conjugate gradient versions,” Journal of
Optimization Theory and Applications, vol. 4, no. 4, pp. 213–243, 1969.
16 J. Stoer and R. Bulirsch, Introduction to Numerical Analysis, vol. 12 of Texts in Applied Mathematics,
Springer, New York, NY, USA, 3rd edition, 2002.
17 H. Pollard, Mathematical Introduction to Celestial Mechanics, Prentice-Hall, Englewood Cliffs, NJ, USA,
1966.
18 G. E. Forsythe, M. A. Malcolm, and C. B. Moler, Computer Methods for Mathematical Computations,
Prentice-Hall, Englewood Cliffs, NJ, USA, 1977.
doi:10.1155/2012/768973
Research Article
Comparison between Two Methods to Calculate the
Transition Matrix of Orbit Motion
Ana Paula Marins Chiaradia,1 Hélio Koiti Kuga,2

and Antonio Fernando Bertachini de Almeida Prado2
1
Grupo de Dinâmica Orbital e Planetologia, Departament of Mathematics, FEG/UNESP, CEP 12516-410,
Guaratinguetá, SP, Brazil
2
DEM-INPE, Avenida dos Astronautas, 1758, CEP 12227-010, São José dos Campos, SP, Brazil
Correspondence should be addressed to Ana Paula Marins Chiaradia, anachiaradia@feg.unesp.br
Received 10 June 2011; Accepted 6 September 2011
Copyright q 2012 Ana Paula Marins Chiaradia et al. This is an open access article distributed
Two methods to evaluate the state transition matrix are implemented and analyzed to verify the
computational cost and the accuracy of both methods. This evaluation represents one of the highest
computational costs on the artificial satellite orbit determination task. The first method is an
approximation of the Keplerian motion, providing an analytical solution which is then calculated
numerically by solving Kepler’s equation. The second one is a local numerical approximation
that includes the effect of J2 . The analysis is performed comparing these two methods with a
reference generated by a numerical integrator. For small intervals of time 1 to 10 s and when one
needs more accuracy, it is recommended to use the second method, since the CPU time does not
excessively overload the computer during the orbit determination procedure. For larger intervals
of time and when one expects more stability on the calculation, it is recommended to use the first
method.
1. Introduction
The orbit determination process consists of obtaining values of the parameters which com-
pletely specify the motion of an orbiting body, like artificial satellites, based on a set of
observations of the body. It involves nonlinear dynamical and nonlinear measurement
systems, which depends on the tracking system, and estimation technique e.g., the Kalman
filtering or least squares 1–6. The dynamical system model consists of the description for
the dynamics of the orbital motion of a satellite, measurement models, Earth’s rotation effects,
and perturbation models. According to Montenbruck and Gill 7, these models depend on,
besides the state variables that define the initial conditions, a variety of parameters that
either affect the dynamical motion or the measurement process. Due to the complexity of
the applied models, it is hardly possible to have a direct solution for any of these parameters
from a given set of observations. To linearize the relation between the observables and the
independent parameters, one should obtain simplified expressions that can be handled more
easily. A linearization of the trajectory and measurement model requires a large number of
partial derivatives, among them, the state transition matrix 7–9.
The function of the transition matrix is to relate the rectangular coordinate variations
for the times tk and tk 1 . The evaluation of the state transition matrix presents one of the
highest computational costs on the artificial satellite orbit determination procedure, because
it requires the evaluation of the Jacobian matrix and the integration of the current variational
equations. This matrix can pose cumbersome analytical expressions when using a complex
force model 10. Binning 11 suggests one method to avoid the problem of the high
computational cost and extended analytical expressions of the transition matrix. This method
consists of propagating the state vector using complete force model and, then, to compute the
transition matrix using a simplified force model. For sure, this simplification provides some
impact on the accuracy of the orbit determination. The analytical calculation of the transition
matrix of the Keplerian motion is a reasonable approximation when only short time intervals
of the observations and reference instant are involved. On the other hand, the inclusion of
the Earth-flattening J2 effect in the transition matrix can be performed adopting Markley’s
method 12.
In Chiaradia 13, Chiaradia et al. 14, and Gomes et al. 15, simplified and
compact algorithms with low computational cost are developed for artificial satellite orbit
determination in real time and onboard, using the global positioning system GPS. The
state vector, composed of the position, velocity, bias, drift, and drift rate of the GPS receiver
clock, has been estimated by the extended Kalman filter in all three works. The fourth-
order Runge-Kutta numerical integrator has been used to integrate the state vector. The
equations of motion have considered only the perturbations due to the geopotential. The
state error covariance matrix has been propagated through the transition matrix, which
has been calculated considering the pure Keplerian motion. To improve the accuracy of
those algorithms, two methods have been compared for calculating the transition matrix
considering circular 1000 km of altitude and elliptical Molniya orbits. Those methods
considered the pure Keplerian motion and perturbed only by the flattening effect of the Earth.
Markley’s method is used to include the flattening of the Earth in the transition matrix. It is a
method that allows the inclusion of more perturbations in a simple way.
In the present work, the spherical harmonic coefficients of degree and order up
to 50 and the drag effect are included in the reference orbit provided by the RK78
numerical integrator Runge-Kutta with Fehlberg coefficients of order 7-8. For this study,
the atmospheric density is considered constant. For the simulations made here, circular and
elliptical low up to 300 km orbits were used. To analyze the results, the reference orbit is
compared with the two methods implemented in this work.
2. State Transition Matrix

The differential equation for the Keplerian motion is expressed by
μr
r̈ − , 2.1
r3
with r x, y, z,

r x2 y2 z2 , 2.2
where r is the magnitude of the satellite position vector and μ is the Earth gravitational
constant. The equation that relates the state deviations in different times is given by

δr δr0
Φt, t0 , 2.3
δv δv0
where r and v are the position and velocity vectors at the time t, respectively, r0 and v0 are the
initial position and velocity vectors at the time t0 , respectively, and Φ is the transition matrix
given by
⎛ ∂r ∂r ⎞

Φ11 Φ12 ⎜ ∂r0 ∂v0 ⎟
Φ ⎜ ⎝ ∂v ∂v ⎠.
⎟ 2.4
Φ21 Φ22
∂r0 ∂v0
The submatrices Φ11 , Φ12 , Φ21 , and Φ22 are calculated in accordance with the two methods
that will be shown in the next topics.
3. First Method: The Analytical Transition Matrix Solution

Considering the Pure Keplerian Motion
Goodyear 16 published a method for the analytical calculation of a transition matrix for
the two-body problem. This method is valid for any kind of orbit. Kuga 10 implemented
this method using the same elegant and adequate formulation optimized for the Keplerian
elliptical orbit problem. He performed some simplifications in this method to increase its
numerical efficiency processing time, memory, and accuracy. However, this method is used
to propagate the position and velocity of the satellite and can be used in any kind of two-body
orbits. Such equations are simple and easy to be developed.
According to Goodyear 16 and Shepperd 17, the four submatrices 3×3 are obtained
developing 2.4. Then, they are written as

M21 M22 T
Φ11 fI r v r0 v0 ,
M31 M32

M22 M23 T
Φ12 gI r v r0 v0 ,
M32 M33
3.1
M11 M12 T
˙ − r
Φ21 fI v r0 v0 ,
M21 M22

M12 M13 T
Φ22 ġI − r v r0 v0 ,
M22 M23

where I is the identity matrix 3 × 3, r v is matrix 3 × 2, and the Mi,j , i, j 1, 2, 3, are the
˙ ġ are calculated in the
components of a 3 × 3 matrix M, which will be shown later, and f, g, f,
next topic 8, 9.
˙ ġ
3.1. Calculating the Functions f, g, f,
Given r0 x0 , y0 , z0 , v0 ẋ0 , ẏ0 , ż0 , and the propagation interval Δt t − t0 ,
r0 x02 y02 z20 ,
h0 x0 ẋ0 y0 ẏ0 z0 ż0 ,

v0 ẋ02 ẏ02 ż20 ,

3.2
2μ
α v02 − ,
r0
1 α
− ,
a μ
where a is the semimajor axis. The eccentric anomaly E0 and the eccentricity e for the initial
orbit are calculated by
h0
e sin E0 √ ,
μa
3.3
r0
e cos E0 1 − ,
a
with E0 reduced to the interval 0 to 2π. The mean anomalies for the initial and propagated
orbits, M0 and M, are given by
M0 E0 − e sin E0 ,
M
μ
n , 3.4
a3
M nΔt M0 ,
with M0 and M reduced to the interval 0 to 2π. The eccentric anomaly for the propagated
orbit is calculated by Kepler’s equation. Kepler’s equation is solved from an initial guess
based on an approximation series, then iterated by Newton-Raphson’s method, until
convergence to the level of 10−12 is achieved. The variation of the eccentric anomaly is
calculated and reduced to the interval 0 to 2π:
ΔE E − E0 . 3.5
The transcendental functions s0 , s1 , s2 for the elliptical orbit, according to Goodyear 16, are
calculated by
s0 cos ΔE,

a
s1 sin ΔE, 3.6
μ
a
s2 1 − s0 .
μ
˙ ġ are valid only for elliptical orbits and are calculated by

Therefore, the functions f, g, f,
r r0 s0 h0 s1 μs2 , 3.7
μs2
f 1− , 3.8
r0
g r0 s1 h0 s2 , 3.9
μs1
f˙ − , 3.10
rr0
μs2
ġ 1 − . 3.11
r
The propagated vectors r e v are given by
r r0 f v0 g,
3.12
v r0 f˙ v0 ġ.
There is no singularity problem and Kepler’s equation is solved through Newton Raphson’s
method in double precision. The classical parameters a, e, i, Ω, ω, are constant in the Keplerian
motion, and, therefore, there is no different subscript for them.
3.2. The Evaluation of the Matrix M3× 3

First of all, it is necessary to calculate the secular component U including the effect of multi-
revolutions in the case where the orbit propagation time, Δt, is larger than one orbital period:

n
ΔE ΔE INT Δt 2π,
2π
s4 cos ΔE − 1,
3.13
s5 sin ΔE − ΔE,

a 5
U s2 Δt ΔEs4 − 3s5 ,
μ
where INTx provides the truncated integer number of the real argument x, and the variables
s4 and s5 are related with the transcendental functions s4 and s5 Goodyear 16. Therefore,
the components of the matrix M are related to the partial derivatives of 3.8–3.11 with
respect to the r0 and v0 , given by

s0 1 1 U
M11 f˙ − μ2 ,
rr0 r02 r 2 r 3 r03

fs˙ 1 ġ − 1
M12 ,
r r2

ġ − 1 s1 U
M13 − μ 3,
r r

fs˙ 1 f −1
M21 − ,
r0 r02 3.14
M22 −fs ˙ 2,

M23 − ġ − 1 s2 ,

f − 1 s1 U
M31 − μ 3,
r0 r0

M32 f − 1 s2 ,
M33 gs2 − U.
3.3. The Property of the Transition Matrix

Sometimes the inverse matrix is required, such as in backward filters. It is also easily accom-
plished as follows. The inverse matrix Φ−1 , which propagates deviations backward from t to
t0 , is given by

−1
ΦT22 −ΦT12
Φ , 3.15
−ΦT21 ΦT11
which results from the canonic nature of the original equations 18, 19. This also applies
to the second method Markley’s if the perturbations are derived from a potential e.g., J2
effect.
4. Second Method: Markley’s Method

Markley’s method uses two states, one at the tk−1 time and the other at the tk time, and
calculates the transition matrix between them by using μ, J2 , Δt, the radius of the Earth, and
the two states. In this case, the effect of the Earth’s flattening is the most influent factor in the
process. Markley’s method consists of making one approximation for the transition matrix
of the state vector based on the Taylor series expansion for short intervals of propagation,
Δt. This method can be used by any kind of orbit and the equations are simple and easily
implemented, as shown next. The state transition’s differential equation is defined by

dΦt, t0 0 I
A1 tΦt, t0 Φt, t0 , 4.1
dt Gt 0
where Φt0 , t0 ≡ I is the initial condition, r x y zT and v ẋ ẏ ż T are the Cartesian
state at the instant t, r0 and v0 are the Cartesian state at the instant t0 , 0 ≡ the matrix 3 × 3 of
zeros, I ≡ the identity matrix 3 × 3, Gt ≡ ∂ar, t/∂r ≡ the gradient matrix, and ar, t the
accelerations of the satellite.
Performing successive derivatives of 4.1, followed by substitutions, gives the deriva-
tive of the transition matrix:
di Φ
Ai tΦt, t0 , 4.2
dti
where
Ai t Ȧi−1 t Ai−1 tA1 t. 4.3
The dot represents the derivative with respect to the time. Developing Φt, t0 in Taylor’s
series at t t0 , using the matrices Ai t0 for i 1, . . . , 4 and the initial condition Φt0 , t0 ≡ I,
the transition matrix of the position and velocity obtained after some simplifications is given
by 12

Φrr Φrv
Φt, t0 ≈ , 4.4
Φvr Φvv 6×6
where
Δt2
Φrr ≡ I 2G0 G ,
6
Δt3
Φrv ≡ IΔt G0 G ,
12
Δt 4.5
Φvr ≡ G0 G ,
2
Δt2
Φvv ≡ I G0 2G ,
6
Δt ≡ t − t0 , G0 ≡ Gt0 .
The calculations of these matrices pose no problems, since the gradient matrix G in the
end of the propagating interval is a function of the final Cartesian state which should be
calculated during the data processing and is available at no extra cost. The G and, therefore,
Φrr , Φrv , Φvr , Φvv are symmetric if the perturbation derives from a potential. The G gradient
matrix including only the central force and the J2 is given by
⎡ ∂a ∂ax ∂ax ⎤
x
⎢ ∂x ∂y ∂z ⎥
⎢ ⎥
∂ar, t ⎢ ∂ay ∂ay ∂ay ⎥
Gt ⎢
⎢ ∂x
⎥. 4.6
∂r ⎢ ∂y ∂z ⎥ ⎥
⎣ ∂az ∂az ∂az ⎦
∂x ∂y ∂z
The accelerations due the central force and the Earth’s flattening are given by

−μx 3 J2 re2 5z2
ax 3 1 1− 2 ,
r 2 r2 r
y
ay ax , 4.7
x

−μz 3 J2 re2 5z2
az 3 1 3− 2 .
r 2 r2 r
The partial derivatives are 20

∂ax μ 3 2 15 J2 re2 2 105 J2 r 2
e 2 2
5 3x − r − J2 re
2 2
x z −
2
x z ,
∂x r 2 2 r2 2 r4

∂ax 3μxy 5 J2 re2 35 J2 re2 2
1 − z ,
∂y r5 2 r2 2 r4

∂ax 3μxz 15 J2 re2 35 J2 re2 2
1 − z ,
∂z r5 2 r2 2 r4
∂ay ∂ax
,
∂x ∂y
∂ay y ∂ax ax
,
∂y x ∂y x
∂ay y ∂ax
,
∂z x ∂z
∂az ∂ax
,
∂x ∂z
∂az ∂ay
,
∂y ∂z

∂az μ 3 2 J2 re2 2 35 J2 re2 4
5 −r 3 z − J2 re 15 2 z −
2 2
z .
∂z r 2 r 2 r4
4.8
5. Analyses of the Methods

First of all, the reference orbit is integrated using the numerical integrator Runge-Kutta of
eighth order with automatic step size control RK78. This integrator is implemented to
integrate simultaneously the stated vectors considering the spherical harmonic coefficients
up to 50th order and degree and the transition matrix considering the Earth’s flattening and
the atmospheric drag effects.
The transition matrix generated by the numerical integrator is used as the reference
to compare the transition matrix generated by the two methods. To compare the accuracy of
them, let us define the global relative error as 10
% %%
%
& 6 %
6 & Φ − Φij %
1 % ij
%
εGlobal % %, 5.1
%
36 i1 j1 % Φij %
%
where Φij is the component i, j of the transition matrix calculated by the RK78 considered
as reference and Φij is the component i, j of the transition matrix calculated by one of the
methods. It gives a rough idea of the number of the common significant figures retained after
the computations.
The first method considers the pure Keplerian motion and the second one considers
the Keplerian motion and the J2 effect. A whole day of integration is divided in intervals of 1
to 60 seconds, which produces from 86,400 to 1,440 steps of integration, respectively, as shown
in Table 1. The total processing time of each method is also shown in Table 1, although it
depends on the code, the computer, and the programmer skills the evaluation used a regular
PC Pentium II processor, 512 MB, running FORTRAN codes. However, it illustrates roughly
the comparative expected performance. One can note that the CPU time difference at any
step of integration and for any method is very small. The first method is slightly faster than
the second one for small steps of time, like 1 and 10 seconds.
To analyze the accuracy of the methods, six comparisons are done, using two kinds
of orbits: one circular and one elliptical. The transition matrix generated by each of the
methods is compared with the reference generated by the RK78. The circular orbit is from
the Topex/Poseidon satellite 21 and the elliptical one is from the Molniya satellite 12.
All those tests are performed for a period of 24 hours. The data of these orbits are shown in
Table 2.
Besides, the RK4 Runge-Kutta of fixed 4th order numerical method for computing
the transition matrix is also implemented and the transition matrix generated by the RK4 is
also included in the comparison. The RK4 used the same dynamical model of the reference
RK78. The results of those three comparisons for the circular orbit are shown in Table 3, and
the results for the elliptical orbit are shown in Table 4. The errors are shown in terms of per
Table 1: Processing time for analytical calculation of the transition matrix.
CPU TIMEs
Δt s Steps
First method Second method
1 86,400 19.8 26.2
10 8,640 2.1 2.7
30 2,880 0.7 0.9
60 1,440 0.4 0.4
Table 2: Keplerian parameters of the test orbits.
Parameter Topex/Poseidon Molniya

Semiaxis major 7714423.46 m 26563000.0 m
Eccentricity 1.13458 × 10−4 0.75
Inclination 66◦ .039 63◦ .435
Longitude of node 236◦ .72 0◦ .0
Argument of perigee 102◦ .83 270◦ .0
Mean anomaly 153◦ .54 0◦ .0
Table 3: Comparison for a circular orbit Topex.
Global error
Δt s RK4
1 3.7 × 10−3 ± 6.0 × 10−3 1.2 × 10−6 ± 1.3 × 10−5 6.3 × 10−8 ± 8.9 × 10−7
10 3.8 × 10−3 ± 8.3 × 10−3 1.0 × 10−4 ± 5.4 × 10−4 5.0 × 10−6 ± 4.0 × 10−5
30 3.9 × 10−3 ± 9.5 × 10−3 7.6 × 10−4 ± 2.1 × 10−3 3.5 × 10−5 ± 1.4 × 10−4
60 3.6 × 10−3 ± 4.9 × 10−3 2.7 × 10−3 ± 5.5 × 10−3 1.3 × 10−4 ± 4.0 × 10−4
300 3.2 × 10−3 ± 1.7 × 10−3 6.6 × 10−2 ± 1.1 × 10−1 3.8 × 10−3 ± 8.2 × 10−3
600 3.1 × 10−3 ± 1.4 × 10−3 2.3 × 10−1 ± 2.3 × 10−1 2.1 × 10−2 ± 2.5 × 10−2
Table 4: Comparison for an elliptical orbit Molniya.
Global error
Δt s RK4
1 3.7 × 10−4 ± 9.8 × 10−4 8.8 × 10−7 ± 2.3 × 10−5 6.2 × 10−8 ± 1.7 × 10−6
10 3.6 × 10−4 ± 8.2 × 10−4 5.7 × 10−5 ± 1.1 × 10−3 4.0 × 10−6 ± 7.6 × 10−6
30 3.5 × 10−4 ± 7.4 × 10−4 2.4 × 10−4 ± 2.7 × 10−3 1.7 × 10−5 ± 1.9 × 10−4
60 3.4 × 10−4 ± 6.9 × 10−4 5.1 × 10−4 ± 3.1 × 10−3 3.6 × 10−5 ± 2.3 × 10−4
300 2.7 × 10−4 ± 3.8 × 10−4 1.5 × 10−2 ± 2.3 × 10−1 1.6 × 10−3 ± 1.3 × 10−2
600 3.5 × 10−4 ± 6.7 × 10−4 4.3 × 10−2 ± 1.9 × 10−1 4.6 × 10−3 ± 1.9 × 10−2
step mean and standard deviations considering the whole day sample. The comparison is
done for only these two kinds of orbits, where the Keplerian approximation for the transition
matrix provides already reasonable accuracy. For large time interval, the second method is
always disadvantageous because it is a local numerical approximation, as depicted in Tables
3 and 4.
6. Conclusions
The goal of this research was to compare and choose the most suited method to calculate the
transition matrix used to propagate the covariance matrix of the position and velocity of the
state estimator, e.g., Kalman filtering or least squares, within the procedure of the artificial
satellite orbit determination. The methods were evaluated according to accuracy, processing
time, and handling complexity of the equations for two kinds of orbits: circular and elliptical.
The processing time of the second method is around 30% larger than the first one; however,
this difference is not considered enough to harm the computer load in the orbit determination
tasks. The second method is more accurate for short intervals, as Δt 1 and 10 seconds, for
the orbits considered. For other intervals of propagation, the first method shows to be more
stable in the sense of keeping the same accuracy regardless of the step size. The equations
of the second method are easier to be handled, which means that it is possible to include
more perturbations easily. However, the first method is a closed analytical solution for the
two-body problem only. The second method has no singularity and no restriction about the
type of orbit; the first one is optimized for elliptical orbits not optimized for parabolic and
hyperbolic orbits. The second method performs an approximation in Taylor’s series, whereas
in the first there is an analytical solution for the Keplerian motion. Therefore, for small
intervals of time 1 to 10 seconds and when one expects more accuracy, it is recommended
to use the second method, since the CPU time does not overload excessively the computer in
the orbit determination procedure. For larger intervals of time and when one expects more
stability on the calculation, it is recommended to use the first method.
Acknowledgments
The authors wish to express their appreciation for the support provided by UNESP
Universidade Estadual Paulista “Júlio de Mesquita Filho” of Brazil and INPE Brazilian
Institute for Space Research.
References
1 G. J. Bierman and C. L. Thornton, “Numerical comparison of Kalman filter algorithms: orbit determi-
nation case study,” Automatica, vol. 13, no. 1, pp. 23–55, 1977.
2 R. G. Brown and P. Y. C. Hwang, “A Kalman filter approachto precision GPS geodesy,” Navigation:
Global Positioning System, vol. 2, pp. 155–166, 1984.
3 A. Gelb, J. F. Kasper Jr, R. A. Nash Jr, C. F. Price, and A. A. Sutherland Jr, Applied Optimal Estimation,
M.I.T. Press, Londres, Inglaterra, 1974.
4 P. S. Maybeck, Stochastic Models, Estimation, and Control, vol. 1, Academic Press, New York, NY, USA,
1979.
5 J. R. Raol and N. K. Sinha, “On the orbit determination problem,” IEEE Transactions on Aerospace and
Electronic Systems, vol. AES-21, no. 3, pp. 274–290, 1985.
6 T. P. Yunck, “Orbit determination,” in Global Positioning System: Theory and Applications, vol. 2, chapter
21, pp. 559–592, American Institute of Aeronautics and Astronautics, Long Beach, Calif, USA, 1996.
7 O. Montenbruck and E. Gill, Satellite Orbits: Models, Methods, and Applications, Springer-Verlag, New
York, NY, USA, 2001.
8 G. J. Der, “An elegant state transition matrix,” Journal of the Astronautical Sciences, vol. 45, no. 4, pp.
371–390, 1997.
9 J. M. A. Danby, Fundamentals of Celestial Mechanics, Willmann-Bell, Richmond, Va, USA, 2nd edition,
1988.
10 H. K. Kuga, Transition Matrix of the Elliptical Keplerian Motion, Instituto Nacional de Pesquisas
Espaciais, São Paulo, Brazil, 1986.
11 P. W. Binning, Absolute and relative satellite to satellite navigation using GPS, Ph.D. thesis, University of
Colorado, Denver, Colorado, USA, 1997.
12 F. L. Markley, “Approximate cartesian state transition matrix,” Journal of the Astronautical Sciences, vol.
34, no. 2, pp. 161–169, 1986.
13 A. P. M. Chiaradia, Autonomous artificial satellite orbit determination and maneuver in real-time using single
frequency GPS measurements, Ph.D. thesis, Instituto Nacional de Pesquisas Espaciais, São Paulo, Brazil,
2000.
14 A. P. M. Chiaradia, H. K. Kuga, and A. F. B. A. Prado, “Single frequency GPS measurements in real-
time artificial satellite orbit determination,” Acta Astronautica, vol. 53, no. 2, pp. 123–133, 2003.
15 V. M. Gomes, H. K. Kuga, and A. P. M. Chiaradia, “Real time orbit determination using GPS
navigation solution,” Journal of the Brazilian Society of Mechanical Sciences and Engineering, vol. 29, no.
3, pp. 274–278, 2007.
16 W. H. Goodyear, “Completely general closed-formsolution for coordinates and partial derivatives of
the two-body problem,” The Astronomical Journal, vol. 70, no. 3, pp. 189–192, 1965.
17 S. W. Shepperd, “Universal Keplerian state transition matrix,” Celestial Mechanics, vol. 35, no. 2, pp.
129–144, 1985.
18 J. M. A. Danby, “The matrizant of Keplerian motion,” American Institute of Aeronautics and Astronautics
Journal, vol. 2, no. 1, pp. 16–19, 1964.
19 R. H. Battin, Atronautical Guidance, Academic Press, New York, NY, USA, 1964.
20 H. K. Kuga, Adaptive orbit estimation applied to low altitude satellites, M.S. thesis, Instituto Nacional de
Pesquisas Espaciais, São Paulo, Brazil, 1982.
21 R. S. Bhat, B. E. Shapiro, R. B. Frauenholz, and R. K. Leavitt, “TOPEX/Poseidon orbit maintenance
for the first five years,” Advances in the Astronautical Sciences, vol. 100, no. 2, pp. 973–988, 1998.
doi:10.1155/2012/123138
Research Article
Higher-Order Analytical Attitude Propagation of an
Oblate Rigid Body under Gravity-Gradient Torque
Juan F. San-Juan,1 Luis M. López,2 and Rosario López1

1
Departamento de Matemáticas y Computación, Universidad de La Rioja, 26004 Logroño, Spain
2
Departamento de Ingenierı́a Mecánica, Universidad de La Rioja, 26004 Logroño, Spain
Correspondence should be addressed to Juan F. San-Juan, juanfelix.sanjuan@unirioja.es
Received 3 March 2011; Revised 25 June 2011; Accepted 3 August 2011
Copyright q 2012 Juan F. San-Juan et al. This is an open access article distributed under the
A higher-order perturbation theory for the rotation of a uniaxial satellite under gravity-gradient
torque demonstrates that known special configurations of the attitude dynamics at which the
satellite rotates, on average, as in a torque-free state, are only the result of an early truncation
of the secular frequencies of motion. In addition to providing a deeper insight into the dynamics,
the higher order of the analytical solution makes it competitive when compared with the long-term
numerical integration of the equations of motion.
1. Introduction
The torque-free rotation of artificial satellites may be perturbed by a variety of effects 1.
Due to the complexity of the force models to integrate, the problem of attitude propagation
is commonly approached numerically, with a variety of available algorithms 2, 3. Nev-
ertheless, the problem can also be approached analytically, in which the analytical alternative
is usually based on perturbation methods: the Euler-Poinsot problem is taken as the un-
perturbed part of the problem, and the other effects are perturbations of the torque-free
rotation.
Among the perturbing torques that drive the rotational motion of an artificial satellite,
a research topic of current interest is the study of the effects produced by external torques
due to the satellite’s interaction with the Earth’s magnetic field 4, 5. On the other hand,
the gravity-gradient torque is often identified as one of the more important perturbations
affecting the torque free rotation 6, 7. Therefore, the model of a free rigid body perturbed
by gravity-gradient torque appears in this literature on this matter as one of the basic,
nonintegrable models used in the study of attitude propagation of artificial satellites 8, 9.
The usefulness of this model is not restricted to the case of artificial satellites and also fits the
description of the rotational motion of natural satellites 10, 11.
Whereas bodies whose attitude propagation may be of interest are triaxial in general,
many celestial bodies have symmetry or near-symmetry rotation, which justifies the use of
the uniaxial model in understanding the perturbed dynamics in this case. The uniaxial model
is not limited to natural bodies, and this has also attracted the attention of aerospace engineers
in the study of the attitude of artificial satellites 12.
The uniaxial model still remains of interest in the study of the attitude propagation,
either due to its direct application to actual problems 13 or because it can be taken as a
truncation of a whole triaxial model as long as the departure from axisymmetry is small. In
this last case, the uniaxial model is considered to be the zeroth order part in a perturbation
approach in which the small triaxiality plays the role of a perturbation 14, 15.
In the case of natural celestial bodies, the work of external moments derived from the
gravitational attraction of other bodies is normally negligible when compared with the kinetic
energy of their torque-free rotation. This is why only first-order effects of the gravity-gradient
torque are usually taken into account in the study of the rotation of most solar system bodies.
However, even when the perturbation model only includes first order effects, extending the
perturbation solution to higher orders provides a clearer insight into the dynamics involved.
Thus, the secular terms of a first-order approach reveal special configurations of the satellite
13, 16 in which the satellite’s attitude under gravity-gradient torque evolves, on average, as
in the torque-free state, but with a slightly modified angular momentum. However, we will
demonstrate that these special configurations do not survive when considering higher-order
terms in calculating the solution. In addition, higher-order solutions would be definitely
useful in increasing computational accuracy, thus extending the validity of an analytical
approach for much longer intervals. For artificial satellites, the gravity-gradient effect may
be much more important than in the case of natural celestial bodies, and the inclusion of
second-order effects would be imperative when investigating the attitude propagation in the
long term. In these case,s when the gravity-gradient second-order effects are nonnegligible,
other effects such as a small triaxiality of the rigid body, or the orbit eccentricity or other
external torques, may introduce observable frequencies in the rotation solution.
We will deal with the problem of the attitude propagation of a satellite under gravity-
gradient torque. In order to get insight into the contribution of higher-order terms in cal-
culating the solution, we will make some simplifying assumptions and focus on the specific
case of a uniaxial satellite under the disturbing gravitation of a perturber in circular orbit. The
problem is solved by perturbation theory up to higher order effects in the gravity-gradient,
thus extending the applicability of analytical results of 13. The successive approximation
method used by 13 now becomes too intricate for computing an analytical solution in which
higher-order effects are involved. Therefore, from the beginning, we resort to the method
of averaging using Lie transforms 17–19 to solve the nonintegrable perturbed problem. It
deserves to be mentioned that approaching attitude dynamics problems with Lie transforms
is not new and has been used from long ago in finding either analytical or semianalytical
solutions of perturbed rotational motion see, for instance 6, 7, 20, 21.
We use Deprit’s method 18, which is specifically designed for automatic machine
computation which allows the perturbation theory to be easily computed to higher orders,
although it is enough to compute the second-order effects introduced by the gravity-
gradient torque to show how the secular behavior changes with respect to that arising from
lower-order truncation theories. Specifically, we will show that special configurations of the
attitude dynamics at which the satellite has been established as rotating as in a torque-
free state except for periodic terms are only the result of an early truncation of the secular
frequencies. The new solution not only provides the required insight into the long-term dy-
namics, but also reveals it is highly competitive when compared with the numerical in-
tegration of the equations of motion.
2. Hamiltonian Formulation
The Hamiltonian of the rigid-body rotation is
H T V, 2.1
where T is the kinetic energy of rotation of a rigid body around its center of mass
1
T Aω12 Bω22 Cω32 , 2.2
2
ω1 , ω2 , and ω3 are the components of the instantaneous rotation vector in the frame of the
principal axis of the body, and A ≤ B ≤ C denote the principal moment of inertia. In the case
which concerns us, the gravity-gradient torque, within the accuracy of higher-order terms in
the ratio of the linear dimensions of the satellite to those of the orbit, the potential energy V
of the external torques is taken from MacCullagh’s approximation 22,

Gm1 m A B C − 3D
V − 1 , 2.3
r 2mr 2
where m is the mass of the rotating body, m1 is the mass of the disturbing body, r is the
distance between the centers of mass of both bodies, G is the gravitational constant, and
D Aα2 Bβ2 Cγ 2 is the moment of inertia of the rigid body with respect to an axis in
the direction of the line joining its center of mass with the disturbing body’s, with direction
cosines α, β, and γ, where the constraint α2 β2 γ 2 1 is applied.
In order to get insight into the contribution of higher-order terms of the analytical
solution, we make some simplifying assumptions. First, we assume that the rotation does
not affect the orbital motion; hence, we neglect the Keplerian term in the potential. Then, we
assume circular orbital motion with r a constant. Finally, we assume that the rotating body
is oblate; hence, A B. Therefore,
1+ 2 , Gm
1

H A ω1 ω22 Cω32 − C − A 1 − 3γ 2
. 2.4
2 2a3
In the Hamiltonian setting, the components of the instantaneous rotation ω1 , ω2 , ω3 ,

and the direction cosine γ in 2.4 are assumed to be state functions of any specific set of
canonical variables. Although Euler angles provide an immediate description of the attitude,
the Hamiltonian of the free rigid body H T does not reveal all the symmetries of the Euler-
Poinsot problem when using Euler canonical variables. In contrast, the integrable character
of the free rigid body problem is evident when using Andoyer variables 23, 24, which
take advantage of the angular momentum integral so as to use the plane perpendicular to
it, leaving aside the center of mass of the rigid body. The invariant plane is defined by the
canonical variables that link the inertial and rotating frames, in which is the latter defined by
the principal axes of the rotating body with the x and y axes which define its equatorial plane.
Using Andoyer variables λ, μ, ν, Λ, M, N, the kinetic energy takes the form

1 sin2 ν cos2 ν N2
T M2 − N 2 , 2.5
2 A B 2C
where M is the modulus of the angular momentum vector, N is its projection on the z axis of
the body frame, and ν is the angle encompassed by the x axis of the body frame and the axis
defined by the intersection of the equatorial plane of the body and the invariant plane. The
variables μ, the angle encompassed by the intersections of the invariant plane with both the
inertial plane and the equatorial plane of the rigid body, λ, the angle encompassed by the x
axis of the inertial plane and the intersection of the inertial and invariant planes, and Λ, the
projection of the angular momentum vector on the axis perpendicular to the inertial plane,
are cyclic in T showing that the torque-free rotation is a problem of one degree of freedom
and, therefore, integrable. Furthermore, in the case of axisymmetric bodies A B, the kinetic
energy is simply
M2 − N 2 N 2
T , 2.6
2A 2C
which is a trivial Hamiltonian integration problem.

In order to express the disturbing function introduced by the gravity-gradient in
Andoyer variables, it is convenient to refer the inertial motion to the orbital plane. Then, after
the usual rotations that match the body and inertial frames, we obtain the direction cosine γ
in Andoyer variables,

γ sin J cosλ − nt sin μ cos J sin I sin J cos I cos μ sinλ − nt, 2.7
where t is time, n is the mean orbital motion, J arccosN/M is the inclination angle
between the equatorial and invariant planes, and I arccosΛ/M is the inclination angle
between the invariant and inertial planes Figure 1.
The explicit appearance of time in the Hamiltonian is easily avoided by moving to a
rotating frame at the same rotation rate of the orbital motion. Thus, we introduce the new
variable
λ − nt; then,
d
dλ ∂H ∂
−n −n H − nΛ, 2.8
dt dt ∂Λ ∂Λ
shows that this change of the reference frame will also require the introduction of the Coriolis
term −nΛ in the Hamiltonian. As a result, we now deal with the conservative Hamiltonian
Kμ, ν,
, M, N, L Hμ, ν,
, M, N, L − nL, where L Λ is the conjugate momentum to the
new variable
.
M Equatorial plane
ν
Plane perpendicular to
J
the angular m om entum
N
μ
λ
I
Inertial plane
Figure 1: Relations between the Andoyer angles and the inclination angles I and J.
Therefore, the Hamiltonian of the rotation of the oblate rigid body in the Andoyer
variables in the rotating frame takes the form
a1 2 a1 − a3 2
K M − N − nL
2 2
κ
2 − 6 cos2 J 1 − 3 cos2 I − 3 sin2 I cos 2
8
> ?
− 6 sin I sin 2J 2 cos I cos μ − 1 cos I cos 2
μ 1 − cos I cos 2
− μ
+ ,
3 sin2 J 2 sin2 I cos 2μ 1 cos I2 cos 2
2μ 1 − cos I2 cos 2
− 2μ ,
2.9
where we use the notation κ −Gm1 C − A/2a3 , a1 1/A, a3 1/C. Note that ν is cyclic,
and, therefore, N remains constant in the perturbed problem.
The equations of motion of the perturbed problem are obtained from the Hamilton
equations

d
, ν, μ ∂K dL, M, N ∂K
, − . 2.10
dt ∂L, M, N dt ∂
, ν, μ
Therefore,
dM 3 2 + ,
κ sin J 2sin2 I sin 2μ 1 cos I2 sin 2
2μ − 1 − cos I2 sin 2
− 2μ
dt 4
> ?
− sin 2J sin 2I sin μ − 1 cos I sin I sin 2
μ − 1 − cos I sin I sin 2
− μ ,
dL 3 > ?
κ 2 sin I sin 2J 1 cos I sin 2
μ − 1 − cos I sin 2
− μ − 2 − 6 cos2 J
dt 4 + ,
× sin2 I sin 2
sin2 J 1 cos I2 sin 2
2μ 1 − cos I2 sin 2
− 2μ ,
dμ
a1 M
dt
3κ + ,
3 cos2 J 1 − 6 cos2 J cos2 I 2 − 2 cos 2
− 4 cos2 J
4M
− cot2 J cot2 I − 2 sin 2I sin 2J cos μ 2 cos2 Jsin2 I cos2 Isin2 J cos 2μ
+ ,
− sin I sin 2J 1 − cos I cot2 J cot2 I − 1 2 cos I cos 2
− μ
+ ,
sin I sin 2J 1 cos I cot2 J cot2 I − 1 − 2 cos I cos 2
μ

1 cos I cos2 J cos I cos 2J cos 2
2μ

1 − cos I cos2 J − cos I cos 2J cos 2
− 2μ ,
dν
−a1 − a3 N
dt
3κ
cos J 4 − 2 sin2 I 6 sin2
cos 2μ − 1 cos I2 cos 2
2μ
4M

− 1 − cos I2 cos 2
− 2μ 2 sin IcotJ − tan J
> ?
× 2 cos I cos μ − 1 cos I cos 2
− μ ,
d
3κ
−n 4 cos I 2 − 3 sin2 J sin2
− 2 sin I 1 − cot2 I sin 2J cos μ
dt 4M
> ?
sin2 J 1 cos I cos 2
2μ − 2 cos I cos 2μ − 1 − cos I cos 2
− 2μ

sin 2Jsin I 1 − cos IcotI cos 2
− μ

sin 2Jsin I − 1 cos IcotI cos 2
μ .
2.11
We must note that Andoyer variables are singular for zero inclination of the in-
termediate plane with respect to either the inertial or the equatorial planes of the body, or
both. These singularities may be avoided when using other sets of variables 8, 25.
3. Perturbation Approach
In general, one can resort to the classical double averaging method to find the secular terms
of the disturbing function see 26, for instance. In our case, the averaging would remove
the fast angles μ and
from 2.9. However useful the classical double averaging may
be in finding the relevant long-term evolution of a dynamical system, in the case of the
Hamiltonian equation 2.9, it would be limited to providing the known first order terms
in the pertinent literature 13, equation 22.
In order to reach the higher orders required in this work, the perturbation solution
is computed analytically by Lie transforms using Deprit’s algorithm 18. This method is
based on solving the homological equation L0 Wn Kn − H Bn , which, in general, is a partial
B
differential equation where the term Hn comes from previous computations, the term Kn of
the new Hamiltonian is chosen at our convenience, and L0 is the Lie derivative, such that
now the term Wn of the generating function can be solved.
We constrain the solution to the case when the spin rate of the satellite is much faster
than its orbital rate, thus precluding the problem of resonances, and then we place the Coriolis
term at the first order. Besides, we assume that the gravity-gradient torque is a second-order
6
effect. Then, the Hamiltonian equation 2.9 is ordered as K m≥0 Km,0 /m!, where
1 1
K0,0 a1 M2 − a1 − a3 N 2 ,
2 2
K1,0 −nL,
κ
K2,0 2! 2 − 6 cos2 J 1 − 3 cos2 I − 3 sin2 I cos 2
8
> ?
− 6 sin I sin 2J 2 cos I cos μ − 1 cos I cos 2
− μ
+ ,
3 sin2 J 2 sin2 I cos 2μ 1 cos I2 cos 2
2μ 1 − cos I2 cos 2
− 2μ ,
Km,0 0, m > 2.
3.1
This order allows us to proceed stepwise to eliminate first the rotation angle μ by
means of a Lie transform and then the angle
by means of a second Lie transform. Moreover,
this splitting of averages has the added advantage that the homological equation of the
method can be solved by quadrature in both canonical transformations.
Note that if the spin and orbital rates were of the same order, the Lie derivative of the
homological equation would be the operator
∂ ∂
L0 ω − a1 M 3.2
∂
∂μ
which involves the solution of a partial differential equation, contrary to a quadrature, for
computing the generating function of the canonical transformation. As shown with the
generating function W just below 22 of 13, this way of proceeding explicitly shows the
denominators that would be small when close to resonances, thus preventing the convergence
of the series solution.
To avoid the explicit appearance of square roots in the automatic evaluation of Poisson
brackets required by the method, it is easier for us to handle circular functions of the
inclination angles I and J as state functions of the Andoyer variables. Their nonvanishing
partial derivatives with respect to the Andoyer variables are
∂cos Q 1 ∂sin Q 1
− cos Q, cot Q cos Q,
∂M M ∂M M
3.3
∂cos Q 1 ∂sin Q 1
, − cot Q,
∂N M ∂N M
either for Q J or Q I.
Averaging of μ
The first canonical transformation μ, ν,
, M, N, L → μ , ν ,
, M , N, L has the effect of
averaging the Hamiltonian over the rotation angle μ . The Lie derivative is now the operator
∂
L0 −a1 M , 3.4
∂μ
and the homological equation can be solved by a simple quadrature to compute the
generating function of the canonical transformation.
6
The Hamiltonian in the new variables is K m≥0 K0,m /m!, which, up to the fourth
order approximation,
1 1
a1 M − a1 − a3 N 2 ,
2
K0,0
2 2
K0,1 −nL ,
κ
K0,2 2! 1 − 3 cos2 J 1 − 3 cos2 I − 3 sin2 I cos 2
,
4
K0,3 0,
a1 M2 3
K0,4 4!
2 64
=
n2 κ 2

3.5
× 32 2 2 1 cos I sin2 J cos 2

a1 M a1 M 2
3κ2 + 2 4

2 4

− 1 26 cos J 5 cos J 6 − 356 cos J 414 cos J cos2 I
a21 M 4

− 5 − 126 cos2 J 153 cos4 J 3 cos4 I sin4 I cos 4

+ ,
− 4 1 26 cos2 J − 27 cos4 J 5 − 126 cos2 J 153 cos4 J cos2 I
F
,
2
× sin I cos 2
,
that only depends on one angle,

. As a consequence, up to the truncation order, the
averaging makes M constant, and, therefore J .
Elimination of

A second canonical transformation μ , ν ,
, M , N, L → μ , ν ,
, M , N, L is now
computed so as to eliminate
. Now, the Lie derivative is
∂
L0 ω , 3.6
∂

and again the homological equation is solved by quadrature. From this second transformation
6
we obtain the double averaged Hamiltonian K m≥0 Km /m! given by
1 1
a1 M − a1 − a3 N 2 ,
2
K0
2 2
K1 −nL ,
κ
K2 2! 1 − 3 cos2 J 1 − 3 cos2 I ,
4
9κ2 2
2 2
K3 −3! cos I sin I 1 − 3 cos J ,
16M n
=
a1 M 2 9 3κ3 3
K4 −4! 1 − 3 cos2 J sin2 I 1 − 5 cos2 I
2 64 a1 M n2 4
κ2 +
1 26 cos2 J 5 cos4 J 6 − 356 cos2 J 414 cos4 J cos2 I
a21 M 4
F
,
2 4 4
−3 5 − 126 cos J 153 cos J cos I ,
3.7
which, up to the truncation order, only depends on momenta and, therefore, is easily in-
tegrated to give the linear motion
μ μ0 nμ t, ν ν0 nν t,

0 n
t ⇒ λ λ0 n
nt, 3.8
in which the secular frequencies of the motion are
3κ + 2 ,
nμ a1 M cos J 1 − 6 cos 2
J cos 2
I
2M
9κ2 + ,
2 2 2 2
cos I 1 − 3 cos J 1 − 9 cos J − 2 1 − 6 cos J cos I
8M 2 n
9κ2 + 2 4

2 4

1 52 cos J 15 cos J 12 1 − 89 cos J 138 cos J cos2 I
64M 3 a1
, 27κ3 2
−9 5 − 168 cos2 J 255 cos4 J cos4 I 1 − 3 cos2 J
+ 64M n2 ,
3
× 1 − 12 cos J − 12 − 90 cos J cos I 15 1 − 6 cos2 J cos4 I ,

2 2 2
3κ 27κ2
nν −a3 − a1 N −
1 − 3 cos2 I cos J cos I
sin 2
I cos J
1 − 3 cos 2
J
2M 4M 2 n
9κ 2 + ,
2 2 2 2 4
− cos J 13 5 cos J − 178 − 414 cos J cos I 27 7 − 17 cos J cos I
32M 3 a1
243κ3 2
2 2
cos J 1 − 3 cos J 1 − 5 cos I sin2 I ,
64M 3 n2
3κ 9κ2 2
n
−n −
cos I 1 − 3 cos2 J − 1 − 3 cos2 I 1 − 3 cos2 J
2M 16M n 2
9κ 2 + ,
− cos I 3 − 178 cos2 J 207 cos4 J − 3 5 − 126 cos2 J 153 cos4 J cos2 I
32a1 M 3
27κ3 3
3 2 1 − 3 cos2 J 3 − 5 cos2 I cos I .
32M n
3.9
We note that the special configurations mentioned in 13—critical inclinations such

as cos I cos2 J 1/3 or cos I cos J 0 in which the rigid body evolves, on average, as in
2
torque-free rotation, although at a slightly different rotation rate from the unperturbed case—
are exactly the result of the order of approximation used, whose perturbation solution only
considered first-order effects on the gravity-gradient perturbation which are equivalent to
the second-order of our present approach by Lie transforms. These special configurations no
longer exist when truncating the perturbation approach at higher orders. Thus, while special
inclinations such as cos2 I cos2 J 1/3 are preserved up to the third-order truncation of
3.9, where
nμ a1 M , nν −a1 − a3 N, n
n 0, 3.10
and the frequencies of the averaged problem correspond to a torque-free rotation state, this
is no longer true for the case cos I cos J 0, where
9κ2
nμ a1 M , nν −a1 − a3 N, n
n − 3.11
16M 2 n
show that λ is no longer fixed, on average, and suffers from a small precessional motion.
Furthermore, the unperturbed-type state at special configurations cos2 I cos2 J 1/3 is also
destroyed at the fourth order of the theory, where
√ √
3κ2 3κ2 5 3κ2
nμ a1 M − , nν −a1 − a3 N − , n
n . 3.12
2M 3 a1 2M 3 a1 4M 3 a1
These higher-order terms may explain the observed behavior in Figure 6 of 13.
4. Numerical Comparisons
So as to evaluate the performance of the analytical solution, we take a fictitious Earth satellite
with moments of inertia A B 400 kg km2 , C 600 kg km2 , which we assume to be rotating
at a rate of 1 rotation per minute. The satellite is assumed to be in a circular orbit of the MEO
region with a semimajor axis a 13 000 km.
In the first test case, we take the satellite in a high-inclination orbit with I 70 deg
and assume that it is rotating close to the axis of maximum inertia with J 10−3 . Besides,
internal units such that the initial value of the modulus of the momentum is M 1 are
1013 × δM
108 × δM
0.4 0.2
0 −0.2
−0.4 −0.6
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Orbital periods Orbital periods
a b
102 × δL
1010 × δL
0.2
0.2
0
−0.2 −0.2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
c d
λ rad
λ rad
0.3 3
0.1 2
−0.1 1
103 ×
−0.3
109 ×
0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
e μ rad f
μ rad
0.6 0.5
0.2 −0.5
−0.2 −1.5
107 ×
102 ×
−0.6 −2.5
0 5 10 15 20 25 30 0 5 10 15 20 25 30
g h
ν rad
ν rad
0.6 2
0.2 1
−0.2 0
102 ×
107 ×
−0.6 −1
0 5 10 15 20 25 30 0 5 10 15 20 25 30
i j
Figure 2: Errors of the fourth-order analytical solution versus the numerical integration for initial
conditions μ ν 0, λ 1, I 70 deg, J 10−3 rad. Left: only the secular terms of the analytical
solution are considered. Right: the analytical solution includes secular and periodic terms. The notation
δ is used for relative errors, whereas Δ means absolute differences.
chosen for the integration; the other initial conditions are μ ν 0 and
1 radian.
For these initial conditions, the nonaveraged equations of motion, 2.11, are integrated
for 30 orbital periods, a time interval in which the satellite completes more than 11 000
rotation cycles. Results of the numerical integration are then compared with the analytical
attitude propagation. In the latter case, the initial conditions must be transformed to the
double averaged phase space, resulting in μ0 0.005824457466, ν0 −0.005932985721,
0 1.0003166358, M0 1.0000000025, and L0 0.34164643181.

Results of the comparison between the analytical solution and the numerical
integration of the nonaveraged equations are presented in Figure 2, where the left-hand
column presents the differences between the numerical integration and the analytical
propagation of the secular terms of the fourth order truncation defined by the frequencies
in 3.9. The right-hand column of Figure 2 shows the differences obtained when using the
×10−4
4
2
μ−μ
0
−2
−4
0 5 10 15 20 25 30
Orbital periods
a
−4
×10
6
ν−ν
2
−2
−6
0 5 10 15 20 25 30
Orbital periods
b
×10−6
4
2
λ−λ
0
−2
−4
0 5 10 15 20 25 30
Orbital periods
c
Figure 3: Attitude evolution under gravity-gradient torque for the special configuration cos2 I cos2 J
1/3 when compared with the corresponding torque-free rotation μ : μ0 a1 M t, ν: ν0 − a1 − a3 Nt, and
:
λ λ0 .
full fourth-order theory, which, in addition to the secular terms propagation, includes the
third-order transformation equations for recovering the medium period terms related to the
averaging over
, and the fourth-order transformation equations which allow us to recover
the short-period terms related to the averaging of μ.
As shown in the plotting on the left of Figure 2, the periodic errors have a noticeable
amplitude when compared with the secular trend. The amplitude of the errors due to periodic
terms is of about 2 arc minutes for λ, although this increases up to about one degree for μ and
ν. When the full fourth order analytical theory is used right-hand column of Figure 2, the
periodic errors are confined to very small values, revealing a secular trend in the order of
10−10 radians per cycle in angular-variables errors, even though the amplitude of the periodic
errors of μ and ν mask this linear trend to some extent. The computation of higher orders in
the perturbation approach should improve the behavior of the analytical theory for both the
secular and periodic terms.
The second test case is for a satellite of the same D physical characteristics and initial
configuration, except that now we take cos I cos J 1/3, one of the special configurations
of the lower-order theories. The propagation of these initial conditions in the nonaveraged
model shows that this configuration is very close to an unperturbed state. As shown in
Figure 3, the time history of the difference between μ and the unperturbed state μ : μ0 a1 M t
seems to consist of only periodic terms, as well as what happens to the difference between
ν and the unperturbed analog ν: ν0 − a1 − a3 Nt. In fact, there is a small linear trend of
the order of 3.1 · 10−7 t for μ and 7.7 · 10−7 t for ν, that is masked by the amplitude of periodic
105 × δM
1012 × δM
0.2 0.6
0 0.2
−0.2
−0.4 −0.2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
a b
1011 × δL
105 × δL
1
0.2
0 −0.2
−1 −0.6
0 5 10 15 20 25 30 0 5 10 15 20 25 30
c d
λ rad
λ rad
0.4 0
0 −0.5
−1
105 ×
109 ×
−0.4
0 5 10 15 20 25 30 0 5 10 15 20 25 30
e f
μ rad
μ rad
0.4 0
0 −0.4
−0.8
103 ×
−1.2
108 ×
−0.4
0 5 10 15 20 25 30 0 5 10 15 20 25 30
g h
ν rad
ν rad
0.6
0.2 −0.2
−0.2 −0.6
103 ×
109 ×
−0.6 −1
0 5 10 15 20 25 30 0 5 10 15 20 25 30
i j
Figure 4: Errors of the fourth-order analytical

D solution versus the numerical integration for initial con-
ditions μ ν 0, λ 1, cos I cos J 1/3 rad. Left: only the secular terms of the analytical solution
are considered. Right: the analytical solution includes secular and periodic terms. The notation δ is used
for relative errors, whereas Δ means absolute differences.
oscillations. In contrast, the linear trend of the differences, in spite of being of the order of
2.9 · 10−8 t, is better appreciated in the evolution of λ because of the notably smaller amplitude
of the periodic oscillations. To highlight this difference, we superimposed a linear fit to the
differences λ − λ0 to their time history presented in the inferior plotting in Figure 3, thus
revealing a clear departure from this fit represented by the straight dashed white line from
zero value.
The initial state in the double-averaged phase-space is now μ0 0.00030577890713,
ν0 −0.0005329981843,
0 0.9999991669, M0 1.0000016387, and L0 0.577352484,
which are the initial conditions that feed the analytical solution. The comparison between
the numerical integration of the nonaveraged equations, and the perturbation solution is
presented in Figure 4. As in the previous example, we note that the propagation of the
secular terms alone introduces important periodic errors left-hand column of Figure 4. In
contrast, when recovering the periodic terms removed from in averaging process, the errors
are reduced to quite acceptable values. As presented in the right-hand column of Figure 4, the
momenta obtained from the fourth-order analytical theory are mainly affected by very small
periodic errors, whilst the secular ones seem to be negligible. On the other hand, the angular
variables are affected by both periodic and secular errors that are apparent in their time
histories. Nevertheless, the secular errors grow at the very low rates of just a few microarc
seconds during an orbital period for λ and ν, and tens of microarc seconds during an orbital
period in the case of μ.
5. Conclusions
The rotation of an oblate rigid body under gravity-gradient torque is a nonintegrable problem
that may be approached by perturbations. Lower-order approaches to the solution that
only consider first-order effects in the gravity-gradient perturbation are normally considered
enough in some applications. However, analytical solutions that consider the effects of the
gravity-gradient torque up to the second-order may be required in the engineering problem
of attitude propagation. Carrying the perturbation approach up to this higher-order provides
a complete insight into the long term dynamics, and the perturbation solution continues
being competitive when compared with the numerical integration of the equations of motion.
The use of Andoyer variables resulted crucial for the perturbation approach used, because
their use notably facilitates the solution of the partial differential equations that provide the
generating functions of the Lie transforms. Finally, since modern perturbation methods are
designed for automatic computation, the analytical solution may be easily extended to even
higher orders if required.
Acknowledgments
Support is acknowledged from grant Gobierno de La Rioja Fomenta 2010/16. The authors
would like to thank Dr. M. Lara and the two anonymous reviewers for providing us with
constructive comments and suggestions.
References
1 V. V. Beletskii, Motion of an Artificial Satellite About Its Center of Mass, Israel Program for Scientific
Translations, S. Monson, Jerusalem, Palestine, 1966.
2 M. D. Shuster, “A survey of attitude representations,” Journal of the Astronautical Sciences, vol. 41, no.
4, pp. 439–517, 1993.
3 T. Fukushima, “Simple, regular, and efficient numerical integration of the rotational motion,” The
Astronomical Journal, vol. 135, no. 6, pp. 2298–2322, 2008.
4 M. C. Zanardi, I. M.P. Quirelli, and H. K. Kuga, “Analytical attitude prediction of spin stabilized
spacecrafts perturbed by magnetic residual torque,” Advances in Space Research, vol. 36, no. 3, pp.
460–465, 2005.
5 R. V. Garcia, M. C. Zanardi, and H. K. Kuga, “Spin-stabilized spacecrafts: analytical attitude
propagation using magnetic torques,” Mathematical Problems in Engineering, vol. 2009, Article ID
242396, 18 pages, 2009.
6 J. E. Cochran, “Effects of gravity-gradient torque on the rotational motion of A triaxial satellite in a
precessing elliptic orbit,” Celestial Mechanics, vol. 6, no. 2, pp. 127–150, 1972.
7 M. C. Zanardi, “Study of the terms of coupling between rotational and translational motion,” Celestial
8 M. C. Zanardi and L. S. Moreira, “Analytical attitude propagation with non-singular variables and
gravity gradient torque for spin stabilized satellite,” Advances in Space Research, vol. 40, no. 1, pp.
11–17, 2007.
9 C. Hellström and S. Mikkola, “Satellite attitude dynamics and estimation with the implicit midpoint
method,” New Astronomy, vol. 14, no. 5, pp. 467–477, 2009.
10 H. Kinoshita, “First-order perturbations of the two finite body problem,” Publications of the
11 H. Kinoshita, “Theory of the rotation of the rigid earth,” Celestial Mechanics and Dynamical Astronomy,
vol. 15, no. 3, pp. 277–326, 1977.
12 J. W. Crenshaw and P. M. Fitzpatrick, “Gravity effects on the rotational motion of a uniaxial artificial
satellite,” American Institute of Aeronautics and Astronautics Journal, vol. 6, no. 11, pp. 2140–2145, 1968.
13 M. Lara, F. Fukushima, and S. Ferrer, “First-order rotation solution of an oblate rigid body under the
torque of a perturber in circular orbit,” Astronomy and Astrophysics, vol. 519, no. 1, pp. 1–10, 2010.
14 J. Touma and J. Wisdom, “Lie-poisson integrators for rigid body dynamics in the solar system,” The
Astronomical Journal, vol. 107, no. 3, pp. 1189–1202, 1994.
15 J. Getino and J. M. Ferrándiz, “On the effect of the mantle elasticity on the Earth’s rotation,” Celestial
Mechanics and Dynamical Astronomy, vol. 61, no. 2, pp. 117–180, 1995.
16 F. L. Chernousko, “On the motion of a satellite about its center of mass under the action of
gravitational moments,” Journal of Applied Mathematics and Mechanics, vol. 27, pp. 708–722, 1963.
17 G.-I. Hori, “Theory of general perturbation with unspecified canonical variable,” Publications of the
Astronomical Society of Japan, vol. 18, no. 4, pp. 287–296, 1966.
18 A. Deprit, “Canonical transformations depending on a small parameter,” Celestial Mechanics, vol. 1,
no. 1, pp. 12–30, 1969.
19 J. A. Campbell and W. H. Jefferys, “Equivalence of the perturbation theories of Hori and Deprit,”
Celestial Mechanics, vol. 2, no. 4, pp. 467–473, 1970.
20 R. V. de Moraes, “A semi-analytical method to study perturbed rotational motion,” Celestial Mechanics,
vol. 45, no. 1–3, pp. 281–284, 1988.
21 M. C. Zanardi and R. V. de Moraes, “Analytical and semi-analytical analysis of an artificial satellite’s
rotational motion,” Celestial Mechanics, vol. 55, no. 4, pp. 227–250, 1999.
22 J. MacCullagh, “On the rotation of a solid body,” Proceedings of the Royal Irish Academy, vol. 2, pp.
520–545, 1840.
23 M. H. Andoyer, Cours de Mécanique Céleste, vol. 1, Gauthier-Villars et cie, Paris, France, 1923.
24 A. Deprit, “Free rotation of a rigid body studied in the phase plane,” American Journal of Physics, vol.
35, no. 5, pp. 424–428, 1967.
25 T. Fukushima, “New canonical variables for orbital and rotational motions,” Celestial Mechanics and
Dynamical Astronomy, vol. 60, no. 1, pp. 57–68, 1994.
26 R. A. Broucke, “Long-term third-body effects via double averaging,” Journal of Guidance, Control, and
Dynamics, vol. 26, no. 1, pp. 27–32, 2003.

Derivation DF DT

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Derivation DF DT

Hochgeladen von

Copyright:

Verfügbare Formate

Mathematical

Guest Editors: Antonio F. Bertachini A. Prado, Maria Cecilia Zanardi,

Hindawi Publishing Corporation

Mathematical Methods Applied to

A Numerical Study of Low-Thrust Limited Power Trajectories between Coplanar Circular

Application of the Hori Method in the Theory of Nonlinear Oscillations,

Low-Thrust Orbital Transfers in the Two-Body Problem, A. A. Sukhanov and A. F. B. A. Prado

Four-Impulsive Rendezvous Maneuvers for Spacecrafts in Circular Orbits Using Genetic

Semianalytic Integration of High-Altitude Orbits under Lunisolar Eﬀects, Martin Lara,

Analysis of Attitude Determination Methods Using GPS Carrier Phase Measurements,

Numerical Analysis of Constrained, Time-Optimal Satellite Reorientation, Robert G. Melton

Low-Thrust Out-of-Plane Orbital Station-Keeping Maneuvers for Satellites, Vivian M. Gomes

An Adaptive Remeshing Procedure for Proximity Maneuvering Spacecraft Formations,

Modified Chebyshev-Picard Iteration Methods for Station-Keeping of Translunar Halo Orbits,

Study of Stability of Rotational Motion of Spacecraft with Canonical Variables,

The Orbital Dynamics of Synchronous Satellites: Irregular Motions in the 2 : 1 Resonance,

Higher-Order Analytical Attitude Propagation of an Oblate Rigid Body under

Antonio F. Bertachini A. Prado,1 Maria Cecilia Zanardi,2

Correspondence should be addressed to Antonio F. Bertachini A. Prado, prado@dem.inpe.br

Received 9 May 2012; Accepted 9 May 2012

Another paper of this category is “Semianalytic integration of high-altitude orbits under

Another of this category is the paper “Higher-order analytical attitude propagation of an

Sandro da Silva Fernandes,1 Carlos Roberto Silveira Filho,2

Correspondence should be addressed to Sandro da Silva Fernandes, sandro@ita.br

Received 15 November 2011; Accepted 19 January 2012

Academic Editor: Silvia Maria Giuliatti Winter

2. Optimal Low-Thrust Limited Power Trajectories

2.1. Formulation of the Optimization Problem

u0  0 v0  1 r0  1 J0  0, 2.4

to the ﬁnal state at the prescribed ﬁnal time tf :

2.2. Application of the Gradient-Based Algorithm

From the Hamiltonian H, one ﬁnds the adjoint equations:

Hxα  04 × 2 null matrix,

From 2.5 and 2.6, one deﬁnes the functions ψ and Φ:

where Λi , i  1, 2, 3 are Lagrangian multipliers associated to the ﬁnal constraints deﬁned by

Φxx  0. 4 × 4 null matrix.

The results of the gradient-based algorithm to the optimization problem described

2.3. Application of the Neighboring Extremals Algorithms

with the boundary conditions:

2.4. Linear Theory

where α  a/a, h  e cos ω, k  e sin ω, where a is the semimajor

1 the orbits of the planets are circular;

ρ tf − t0 Janal Jgrad JNeigh1 JNeigh2 drel 1 drel 2 drel 3

ρ tf − t0 Jlinear Jgrad JNeigh1 JNeigh2 drel 1 drel 2 drel 3

ρ tf − t0 Jlinear Jgrad JNeigh1 JNeigh2 drel 1 drel 2 drel 3

ρ tf − t0 Jlinear Jgrad JNeigh1 JNeigh2 drel 1 drel 2 drel 3

Table 5: Consumption variable J for Earth-Venus transfers.

ρ tf − t0 Jlinear Jgrad JNeigh1 JNeigh2 drel 1 drel 2 drel 3

Table 6: Consumption variable J for Earth-Mars transfers.

ρ tf − t0 Jlinear Jgrad JNeigh1 JNeigh2 drel 1 drel 2 drel 3

Table 7: Consumption variable J for interplanetary transfers with large-amplitude.

Table 8: Consumption variable J for large transfers.

Thrust acceleration-ρ = 0.975 Thrust acceleration-ρ = 1.025

Thrust acceleration-ρ = 0.950 Thrust acceleration-ρ = 1.050

Thrust acceleration-ρ = 0.727 0.6 Thrust acceleration-ρ = 1.523

Linear theory Linear theory

Figure 1: Thrust acceleration for tf − t0  2.0.

Thrust acceleration-ρ = 0.975 Thrust acceleration-ρ = 1.025

Thrust acceleration-ρ = 0.950 Thrust acceleration-ρ = 1.050

0 Circumferential acceleration 0.02

Thrust acceleration-ρ = 0.727 Thrust acceleration-ρ = 1.523

Linear theory Linear theory

Figure 2: Thrust acceleration for tf − t0  3.0.

u0 0 v0 1 r0 1 J0 0, 2.4

Hxα 04 × 2 null matrix,

where Λi , i 1, 2, 3 are Lagrangian multipliers associated to the ﬁnal constraints deﬁned by

Φxx 0. 4 × 4 null matrix.

where α a/a, h e cos ω, k e sin ω, where a is the semimajor

Figure 1: Thrust acceleration for tf − t0 2.0.

Figure 2: Thrust acceleration for tf − t0 3.0.

Figure 3: Thrust acceleration for tf − t0 30.0.

Figure 4: Thrust acceleration for tf − t0 50.0.

Figure 5: Thrust acceleration for large-amplitude transfers with tf − t0 100.0.

Figure 6: Thrust acceleration for large-amplitude transfers with tf − t0 200.0.

Following Hori 4, the transformation of variables z1 , . . . , zn → ζ1 , . . . , ζn gener-

j 1, . . . , n, where j stands for the generalized Poisson brackets

ζj ζj c1 , . . . , cn , t, j 1, . . . , n, 3.3

ζj ζj c1 , . . . , cn−1 , M, j 1, . . . , n, 3.5

being c1 , . . . , cn−1 arbitrary constants of integration and M t τ, where τ is an additive constant;

ẍ x εfx, ẋ. 4.1

ż1 z2 , ż2 −z1 εfz1 , z2 . 4.2

ζ̇1 ζ2 , ζ̇2 −ζ1 , 4.5

∗m ∂ −Et + Et m ,

Δ1ζ Δ2ζ ΔC , 4.24

ζ̇1 0, ζ̇2 0, 4.31