CubeSat De-Orbit Point Targeting

using Drag Modulations

a project presented to
The Faculty of the Department of Aerospace Engineering
San José State University

in partial fulfillment of the requirements for the degree

Master of Science in Aerospace Engineering

by

Sebastian Smith
December 2017

approved by

Dr. Marcus Murbach

Faculty Advisor
 CubeSat De-Orbit Point Targeting using Drag
Modulations
Sebastian Smith ∗ Marcus Murbach †

Abstract

Spacecraft de-orbit point targeting can be used for quick returns from low Earth orbits
by relying solely on a passive drag device. Using a series of drag modulations, a small
satellite can target a pre-selected terminal location provided that the targets’ latitude is
less than the orbit inclination and that initial altitude meets specific criteria. Utilizing
an Exo-Brake, a passive drag device capable of manipulating its reference area, orbital
decay rates can be compared to analyze its effectiveness at various altitudes. A numerical
propagator using the Jacchia Bowman 2008 atmospheric model was developed in MAT-
LAB to predict the trajectory and the total amount of time until re-entry. The initial
trajectory can then be used as a reference point to determine the proper ballistic profile
necessary for reaching the desired target. In this paper, multiple target sites and initial
condition constraints were used to assess the accuracy of the model. A targeting error
of 100 km was chosen to allow for the final stage, a guided parafoil, to steer the satellite
remains to the desired location. While it is possible to obtain a targeting error under
100 km using drag modulation techniques, other control methods should be considered to
provide a more robust and accurate solution.

Keywords: Orbit-Decay, Aerodynamic Drag, Exo-Brake, Drag Modulation, MATLAB,

TechEdSat, NASA Ames Research Center, Re-entry targeting, De-Orbit Propagator

Nomenclature
ω Argument of perigee (km)

h Angular momentum of the spacecraft

J2 Orbit perturbations due to planet oblatness

∗Masters
Student, Department of Aerospace Engineering, SJSU
†Principal Investigator, NASA Ames Research Center, SJSU Advisor

1
REarth Radius of the Earth (km)

adrag acceleration due to atmospheric drag (skm )

CD Drag coefficient

A Reference Area (m2)

m Mass (kg)

ρ Density ( mkg3 )

vsat Velocity relative to the spacecraft (km

s
)

r Radius from the center of the planet to the center of the satellite (km)

vatm Velocity relative to the atmosphere ( skm )

β Ballistic coefficient (mkg2 )

Tg (N)
Gravity gradient torquekm

I Moment of inertia

λ Mean longitude (deg)

Me Mean anomaly of elliptical or circular orbit (rad)

E Eccentric anomaly (rad)

M Mean motion (rad

s
)

∆θd Difference in total change in true anomaly between the new and old trajectory (rad)

∆t1 Time until swap point in the new trajectory (s)

p Semi-latus rectum (km)

∆t2 Time from swap point until terminal point in the new trajectory

∆t10 Time until swap point in the initial trajectory (s)

∆t20 Time from swap point until terminal point in the initial trajectory (s)

∆θ10 Change in true anomaly from the initial time to the swap time in the initial trajectory
(rad)

∆θt Desired total change in true anomaly of the new trajectory until the terminal location

∆θ20 Change in true anomaly from the swap time to the terminal time in the initial trajectory
(rad)

2
∆θ1 Change in true anomaly from the initial time to the swap time in the new trajectory
(rad)

∆θ2 Change in true anomaly from the swap time to the terminal time in the new trajectory
(rad)

∆tt Total time for the new trajectory to reach the terminal location (s)

∆θd Difference in total change in true anomaly between initial and new trajectories (rad)

e eccentricity of the orbit

θ True anomaly

T Orbital Period (s)

3
m
µ Gravitational Constant (kg s)2

a Semi-major axis (km)

i Orbit inclination (rad)

Ω Right ascension of the ascending node (rad)

1 Introduction
The ability to frequently return payloads from low Earth orbit has become increasingly popu-
lar since the retirement of the Space Shuttle program in 2011. While re-entry devices such as
NASA’s Orion spacecraft and SpaceX’s Dragon capsule have the ability to return large amounts
of cargo, factors such as cost of deployment and infrequent launch dates make them impractica-
ble for smaller payloads onboard the ISS. Under the Small Payload Quick Return (SPQR) and
Sub Orbital Aerodynamic Re-entry EXperiments (SOAREX) projects, NASA proposes to test
a series of small satellites aimed to fill the demand for smaller payloads utilizing an Exo-Brake,
a parachute-like, modulating drag device capable of changing its reference area to produce drag
in exoamtospheric conditions. The TechEdSat series, a line of CubeSat’s ranging from 1U-6U,
is the primary test-bed for the Exo-Brake design and deployment. In addition to quick returns,
the Exo-Brake offers the unique ability to target a re-entry location without any propulsive
devices, so as long as the desired target lies on the given orbit path. Guaranteeing a safe and
controlled re-entry becomes crucial to mission planning as some of the satellite, including the
payload, will eventually have components capable of surviving the re-entry process, posing as
a potential threat to the surrounding inhabitants.
TechEdSat, or Technology Educational Satellite, is a series of low-cost CubeSats designed to
test new technologies in satellite-to-satellite communications and passive de-orbiting. Beginning
as a conjoined project between San Jose State University, Sweden’s ÅAC Microtec, and NASA
Ames Research Center in 2012, the original TES was engineered to evaluate plug-and-play

3
avionics and experiment with orbital communications using Iridium and Orbcomm satellite
phone networks. TechEdSat-3p, launched in November 2013, then became the first flight test
with the Exo-Brake passive de-orbit system, marking the beginning of quick return experiments
designed for ISS and Mars re-entry applications. Taking up roughly 2/3 of the total volume, the
erectable deployment system was engineered as a tension-based structure to help the spacecraft
re-enter in 10 days after being deployed from the space station. Building upon the Exo-Brake
design, TechEdSat 5 added the ability to modulate the surface area to allow more precise
atmospheric entry predictions. Utilizing a single winch system, the Exo-Brake’s canopy can
be extended or retracted, effectively changing the reference area to modulate drag. With
upcoming launches planned for the fall of 2017 and 2018, future TechEdSat’s look to further
expand the development of the Exo-Brake. While the eventual goal is to fully target and control
the spacecraft autonomously, initial tests will control the Exo-Brake via Iridium constellation
to validate the control algorithm and confirm that guided re-entry is possible using only an
Exo-Brake.

Figure 1: TechEdSat 1 (far left) being deployed in 2012 [9]

2 Orbital Mechanics for a Two-Body Problem

In order to understand the optimal trajectory for TechEdSat’s guided de-orbit, basic orbital
mechanics based off Kepler’s laws and their corresponding orbital elements must be first consid-
ered for an initially circular orbit. The classical two-body problem is defined when two bodies’
motion can be determined solely by their own mutual gravitational attraction.

2.1 Kepler’s Laws

The three laws that describe the motion of planets around the Sun are known as Kepler’s laws.

4
 1. The orbit of a planet is an ellipse with the Sun at one of the two foci.

p
r= (1)
1 + ecos(θ)

2. A line segment from the center of the sun to the center of the planet will sweep out in
equal areas over an equal interval of time.

dA 1 dθ (2)
= r2
dt 2 dt

3. The square of a planets orbital period is proportional to the cube of the of the semi-major
axis of its orbit.
T 2 4π2 (3)
=
r3 µ

Kepler’s first law, also known as the law of ellipses, states that all the planets orbiting the
sun follow an elliptical path with the sun at one of the foci. Equation 1 shows the relationship
of radius with the semi-latus rectum, eccentricity, and true anomaly. The second law, or the
law of equal areas, describes the speed of a planet while it orbits the Sun. Because a planets
velocity varies proportionally to its distance from the Sun, the velocity will vary along the
orbits path. Despite the speed of the planet, Kepler’s second law states that in equal times, the
object will sweep equal areas for that orbit. Unlike Kepler’s second law, the third law, or the
law of harmonics, compares the orbital periods and radi of multiple planets. This relationship
between orbital period and radius offer insight to the fundamental principles of motion for both
satellites and planets. [2, 6]

2.2 Keplerian Elements

The basis of orbital mechanics revolve around Kepler’s three laws of planetary motion and
the elements that govern them. Keplerian elements describe the orbit of an object, such as a
satellite, around a large body using the following parameters:

1. a, Semi-Major Axis: The distance between perigee and the center of the orbit

2. e, Eccentricity: The ratio of half the foci separation to the semi-major axis

3. i, Inclination: The angle between the equitorial plane and the orbital plane, measured
counterclockwise at the ascending node

4. Ω, Right Ascension of the Ascending Node: The angle between the vernal equinox to the
ascending node

5. ω, Argument of perigee: The angle measured from the ascending node to the perigee

6. θ, True Anomaly: The angle measured from perigee to the satellites location

5
Each Keplerian element is designed to describe a unique aspect of the objects motion. The semi-
major axis is used to describe the orbits size while the eccentricity and inclination elements
portray the orbits shape and plane tilt respectively. Ω, ω, and θ each describe the orbital planes
rotation about the Earth, the orbits orientation in the orbital plane, and the satellites location
within the orbit respectively. Keplerian elements can then be expressed in a set called a TLE,
or two-line element. TLE’s compact orbital elements of an Earth-orbiting object for a given
time and represent the trajectory of the spacecraft in two lines of 80-column ASCII text. [2, 6]

Figure 2: Keplerian Elements using a Geocentric Equatorial Frame [2]

2.3 Governing Equations

For determining the motion of two bodies based solely off their mutual gravitational attraction,
the path of one of the masses relative to the other resembles a conic section whose shape is
determined by the eccentricity of the orbit. The equation that describes this two-body orbit is
known as the orbit equation.
2
r= h 1 (4)
µ 1 + ecos(θ)

where,

rh = r × rṙ (5)

For a two-body orbit, or Keplerian orbit, µ, h, and e are all constants. Because the orbit
equation models conic sections, it becomes a mathematical expression of Kepler’s first law.
When the eccentricity is equal to zero, then the orbit is said to be circular. Because the orbit
is a circle, r becomes a constant and therefore the radial velocity, rṙ is zero. For this unique

6
situation, the position and velocity vectors can be shown as follows:
2
r= h (6)
µ
I
µ
v= (7)
r

In addition, the acceleration of a satellite in a circular orbit about the Earth for a basic two
body problem can be expressed by the following equation:
µ
r̈ = r (8)
r3
(9)

where,
/
r = rx + ry + rz and r = x2 + y 2 + z 2 (10)

However, this equation only accounts for the mutual attraction of two objects in a circular
orbit. Other factors that heavily affect a satellites trajectory are known as orbit perturbations.
[2, 6]

3 Orbit Perturbations
Orbit perturbations, such as J2 perturbations, solar radiation, and atmospheric drag, all con-
tribute to the trajectory of objects orbiting around the Earth. Due to these various factors,
the motion of an object orbiting a non-perfect sphere will experience oscillations in its orbital
trajectory.

3.1 J2 Perturbations
Gravitational fields about a true spherical mass have an inverse relation. But because the Earth
is not perfectly circular, a planets oblatness can affect the orbit of satellites. By having slightly
flatter poles and a wider equator, a difference in forces causes a perturbation acceleration known
as J2. This aspherical nature of the Earth leads to a gravitational attraction that is no longer
directed specifically towards the center of mass. The J2 perturbation is given by the following
equation:
2
3J2 µR 2Earth ((5 rx r2
r̈ = − 1)(rx + ry) + rz (5 x−
3)) (11)
2R5 R2 R2

Here, the x, y, and z position vectors are represented in the Earth centered inertial frame
(ECI). While the J2 perturbation is small, its effect significantly deviates the trajectory away
from the pure inverse square motion. For a nearly circular orbit, the J2 perturbation accelera-
tion equations can be added to the Clohessy-Wiltshire equations.

7
 3J2n2R2 rx rx
ẍ = Earth
(1 − (1 − 3sin 2
(nt)sin 2
(i)) + (9cos2(nt) + 9cos2(i)sin2(nt) − 6)
2R0 R0 r R0 rz (12)
+ (8sin(nt)cos(nt)cos2(nt)) − (8sin(nt)sin(i)cos(i)))
y
R0 R0
3J2n2R2 7rx ry
ÿ = Earth
((1 − )(sin(2nt)sin2(i) + (7cos2(nt) + 5cos2(i)−
2R0 R0 R0 (13)
rz
4 7cos (i))2
(2cos(nt)sin(i)cos(i))
− − R0
7r x rx
3J2n2 R2 ((1 ) ( ) (2 ) + (6 ( ) ( ) ( ))
z̈ = Earth cos i sin i sin nt
2R0 − sin nt sin i
R 0 R0 (14)
ry rz
+ (2cos(nt)sin(i)cos(i)) − (7cos2(i) + 5cos2(nt) − 4 − 5cos2(nt)cos2(i))
R0 R0

This can be simplified to a purely spherical harmonic expression:

∂V µx 3 a 2 z2

ẍ = = (1 − J2 ( ) (5 2 − 1)) (15)
∂x r3 2 r r

∂V µy 3 a 2 z2

ÿ = = (1 − J2 ( ) (5 2 − 1)) (16)
∂y r3 2 r r

∂V µz 3 a 2 z2

z¨ = = (1 − J2 ( ) (5 2 − 1)) (17)
∂z r3 2 r r

While there are several orders of zonal harmonic perturbations, the second order is the most
dominate. The second zonal harmonic for the Earth is known as the constant:

J2 = 1.082629 x 10−3

Although other harmonics induce perturbations on orbiting objects, their contributions are
overshadowed by the J2 perturbation as no other harmonics are greater an order of 10−6. As
shown below, the effects of J2 perturbations on a 3.5 kg CubeSat oscillate steadily between 300
km and 288 km over a period of 4.5 days. However, once the altitude begins to rapidly drop,
the oscillations become less severe. [2, 12, 5]

8
Figure 3: Influence of J2 perturbations for a 3.5 kg CubeSat with an orbit inclination of 51.7◦

3.2 Atmospheric Model

One of the driving factors in orbit perturbations stems from the relationship between density
and altitude. The farther away a satellite travels from the Earth’s atmosphere, the smaller the
density becomes. While there are various methods for estimating Earth’s atmospheric density
model, one of the more accurate models is the Jacchia-Bowman 2008 (JB 2008). Originally
developed as a revision to the 2006 model, JB 2008 builds upon Jacchia’s diffusion equations
by adding extreme solar irradiances, exospheric temperature equations, semiannual density
equations based on multiple 81-day averaged solar indices, and geomagnetic storm effects. De-
termining variations in the ultraviolet solar radiation depend on two components: the solar
rotational modulation of active region emission and the long-term evolution of the solar mag-
netic field. Jacchia utilized the following daily indices as key parameters for both his models:

• F10: The 10.7-cm solar radio flux is a flux density measurement of the total emission at a
wavelength of 10.7 cm from all sources captured by the solar disk. The F10 has physical
units of W m−2Hz−1 but is often referred to in solar flux units (sfu). An 81-day centered
average of F10 is expressed as F¯10

• S10: The S10 variable is measured from the 26-34 nm solar EUV emission. It is then
normalized and converted to sfu by means of linear regression using F10. Most of the
irradiances are dominated by the chromospheric He II line as well as other chromospheric
and coronal lines, which comes from solar active regions. An 81-day centered average of
the S10 can also be expressed as S¯10 .

• M10: The M10 variable accounts for the chromospheric Mg II h and k lines as well as the
photospheric wings. Its purpose is to measure the chromospheric and photospheric solar
activity without the influence of instrument sensitivity. Once a linear regression with F10
has been achieved, the M10 is scaled to an index of sfu. An 81-day centered average of
the M10 can also be expressed as M¯10 .

9
 • Y10:As photons arrive at Earth, they are primarily absorbed into the mesosphere and lower
thermosphere by molecular oxygen and nitrogen to create the ionspheric D-region, also
known as the X10 variable. While the 0.1-0.8 nm X-rays account for a major energy source
during high solar activity in this atmospheric region, hydrogen (H) Lyman-α emissions
dominate during moderate and low activity. Lyman-α emissions are formed in solar active
regions when the photons are absorbed and dissociate nitric oxide (NO) and become
involved in water (H20) activity. In order to account for both solar emissions, Y10 was
created as a mixed solar index to represent mostly X10 during solar maximum and Lyman-
α during moderate and low solar activity. A normalized value of F¯10 is used as the
weighting function.

Until the creation of JB 2006, empirical atmospheric density models typically yielded errors
between 15% - 20% due to the lack of modeling two important components: the semiannual
density variation and an incomplete thermospheric heating parameter. The semiannual density
variation was first discovered in 1961 through the analysis of satellite drag data. Scientists
observed a 6-month periodicity maximum occurring in April and October and minimum’s in
January and July. Despite being a worldwide effect, the semiannual period was found to be only
approximate as each max and min occurred at varying times from year to year. Jacchia found
that the amplitude of the semiannual density variation was strongly dependent on altitude as
shown in the following equation:

∆SAlog10ρ = F (z)G(t) (18)

where G(t) represents the average density variation as a function of time and F(z) is the relation
between the amplitude and height z. The average density variation consists of a Fourier series
with nine coefficients representing a quad-annual variation over a 28-day smoothed density
difference. Using the F¯10 , S¯10 , and M¯10 variables, a new solar index representing the long term
EUV and FUV heating can be determined.

FS̄ M = 1.00F¯10 − .75S¯10 − .37M¯10 (19)

Once the new solar index is calculated, the yearly semiannual phase variations can be used to
model the average density variation.

G(t) = C1 + C2 sin(ω) + C3 cos(ω) + C4 sin(2ω) + C5 cos(2ω) + FS̄ M (C6 + C7 sin(ω)

(20)
+C8cos(ω) + C9sin(2ω) + C10cos(2ω))

where coefficient values with standard deviations are determined from best fit results ob-
tained from satellite data.

10
Figure 4: Average density variation as a function of time and F(z) for three different years [1]

The semiannual height function, F(z), was determined on a yearly and satellite-by-satellite
basis by a nine term Fourier series. The F(z) value was then computed for each fit as the
difference of minimum and maximum values for that given year. Similar to the G(t) function,
a new solar index representing the semiannual variation can be calculated using the F¯10 , S¯10 ,
and M¯10 variables.

FS̄ M = 1.00F¯10 − 0.70S¯10 − 0.04M¯10 (21)

The FS̄ M is then used to determine the new height index as follows:

2
F (z) = B1 + B2FS̄ M + B3 zFS̄ M + B4 z 2FS̄ M + B5 zF¯10 (22)

11
 Figure 5: Amplitude function F(z) for three different years with semiannual amplitudes
plotted vs height[1]

Modeling geomagnetic storms form an important part of correctly calculating atmospheric

density. Geomagnetic storm modeling is monitored through the Disturbance Storm Time (Dst)
index and is primarily used to determine the strength of the storm within the inner magneto-
sphere. During the primary phase of the storm, the storm-time ring becomes highly energized
and generates a southward-directed magnetic field perturbations at lower altitudes. When com-
pared with data from satellites, the JB 2008 model is extremely consistent with orbit altitudes
of 400 and 500 km. [1]

Figure 6: Major 2004 storms with Dst and density ratios displayed [1]

The relationship between density and altitude for a specific day and latitude/longitude pair
can be expressed by an exponential correlation. While an exponential density model provides

12
a simple relationship between density and altitude, the accuracy decreases the closer it gets to
the Karmin line. As shown in the following figure, the JB 2008 model provides a much lower
density distribution for a given longitude and latitude.

Figure 7: JB 2008 altitude plot versus exponential density estimation

Another atmospheric model commonly used to reproduce temperature and density from
ground to space is NRLMSISE-00, an empirical, global model of Earth’s atmosphere based on
actual satellite drag data. Developed by the US Naval Research Laboratory (NRL), MSIS,
or mass spectrometer and incoherent scatter radar, has become the standard for international
space research in predicting satellite decay due to atmospheric drag. However, MSIS accuracy
often suffers both before and after storm periods as a result of not correctly modeling the solar
EUV data. The following figure shows the comparison of MSIS and JB 2008 on a given day
as a function of altitude. Both models portray accurate density readings at lower altitudes,
but tend to differ farther away from the Karman altitude due to the increasing effect of solar
events.

13
 Figure 8: JB 2008 altitude plot versus MSIS on April 10, 2017

After inputting the Earth orientation parameters, space weather data, and geomagnetic
storm values, the density can be calculated as functions of altitude, longitude, and latitude for
any given day. The following plots show the density variation across the globe at 300 km for
different seasons of the year.

Figure 9: Density distribution as a function of longitude and altitude at 300 km for the winter
and spring of 2016 respectively

14
 Figure 10: Density distribution as a function of longitude and altitude at 300 km for the
summer and fall of 2016 respectively

As shown, the density is greatest in the spring, causing the orbital lifetime to decrease when
compared to the other seasons. Results from the Canadian Center of Science and Education
indicate that there is a higher probability of producing an intense geomagnetic storm at the
equinoxes that at the solstices, heating up and expanding the Earth’s upper atmosphere. As the
heat rises, the density above the Karman line increases significantly, thus creating more drag
and decreasing the mission time of the satellite when these storms occur. While geomagnetic
storms can occur in any month, the greater density of storms are produced around the solar
maximum solar cycle and during the descending phase of the solar cycle. In addition to having
a correlation to the season, density is also affected by its location around the globe. Due to
various effects like geomagnetic storms and increased solar ultraviolet emission, Earth’s upper
atmosphere around the lower to mid latitudes tend to be more dense then either of the poles,
producing increased drag in these locations [8].

3.3 Atmospheric Drag

Atmospheric drag affects all satellites orbiting the Earth and is the most significant source of
error in modeling their trajectory. At any orbital altitude, atmospheric drag occurs when gas
molecules frequently collide with the satellite and accounts for the primary cause for orbital
decay in low Earth orbit (LEO). This momentum transfer between the molecules in the at-
mosphere and the spacecraft form the underlying cause of atmospheric drag. Like a positive
feedback effect, a lower altitude corresponds with a faster orbital decay rate due to an increase
in atmospheric density, aerodynamic drag, and heating. Determining the atmospheric drag
force becomes one of the most important parameters when propagating a satellites trajectory.
The most widely accepted equation for atmospheric drag is as follows:
1 A
ra = − C ( )ρV 2 rvrel (23)
drag D rel
2 M |rvrel|
where, rvrel = rvsat + rvatmo (24)

15
Here, CD is the drag coefficient, A is the reference area, m is the mass, ρ is the atmospheric
density, and vrrel is the velocity relative to the spacecraft. Aerodynamic forces and moments
are the reaction of an object doing work on the medium it is traveling in. The momentum and
energy exchanged between the object and the medium depends on various factors such as mass
distribution of the medium, geometry of the object, and characteristics of the objects surface.
The duration of the interaction between the spacecraft and atmosphere plays an important
role. There must be enough time for collisions among gas particles to equilibrate as well as
fluid particles to redistribute energy and momentum. The common approximation for the drag
coefficient is as follows:
2mradrag |rvrel | (25)
cD =
ρv 2rel Arvrel

This approximation is only appropriate when the mean free path between collisions among
gas particles is much greater then the length scale of the problem or when the interaction time
between the spacecraft and fluid is much higher than the time between particle collisions.

Figure 11: Influence of Drag for a 1000 kg satellite with a 100 m2 drag area and a CD = 2.2
using a Harris-Preister atmospheric model [14]

The difficulty in determining the atmospheric drag force is that none of the above variables
are exactly known. For example, the drag coefficient, while unique for each satellite, depends
on the type of scattering between the surface of the structure and the neutral particles in
the atmosphere. Because there are fundamental unknowns in the physics of scattering, it is
impossible to reliably predict how particles in free motion behave under that specific condition.
As shown in the previous section, the density can greatly vary depending on the time of year,
the solar radiation from the sun of that specific day, geomagnetic storms, or position around the
Earth, depending on which atmospheric model is chosen. However, an increase in density will

16
lead to an increase in drag acceleration, effectively speeding up the orbital decay rate. Using
the Jacchia Bowman 2008 atmospheric model, the influence of atmospheric drag on a CubeSat
with a drag device can be shown as follows. [4, 15, 6, 14, 2]

Figure 12: Influence of Drag for a 3.5 kg CubeSat with a 0.2027 m2 drag area and a CD = 2.2
using a JB 2008 atmospheric model

3.4 Gauss Variational Equations

When a perturbing acceleration acts on a system in orbit, the orbital elements will experience
oscillations over time and drift away from their initial values. If a perturbing acceleration
expressed in the inertial XYZ frame is

pr = pxri + pyrj + pzrk (26)

and expressed in the non-inertial rsw frame as

pr = pX r + pY rs + pZ w
r (27)

then the transformation between these two frames can be written as

   
pr pX
 ps  = [Q] Xr  pY  (28)
   
pw pZ

Here, each row of [Q]Xr is the direction cosines of the unit vectors for r, rs, and w
r to relative
the XYZ axes:
 
−sin Ω cos i sin u + cos Ω cos u cos Ω cos i sin u + sin Ω cos u sin i sin u

−sin Ω cos i sin u − cos Ω sin u cos Ω cos i cos u + sinΩ cos u sin i sin u (29)
 
sin Ω sin i −cos Ω sin i cos i

where Ω is the right ascension of the ascending node, i is the inclination, and u is the
oscillating element. The transformation from XY Z to rsw is a rotation about the normal w r

17
through the true anomaly, θ. With this mapping, the time deviates for the oscillating orbital
elements h, e, θ, Ω, i, andω can be derived to form the Gauss planetary equations when looking
at the perturbations due to J2. As stated earlier, the time derivative of angular momentum
is the cross product between the position and velocity vectors. Using the definition of the
magnitude of a vector, the time rate of change for angular momentum can be expressed as

drh rh drh drh

/ (
dh d r r 1 1
= h·h= / 2rh · = · r·
=w (30)
dt dt 2 r r dt h dt dt
h·h

Substituting the cross product for the time rate of change of the angular momentum vector
and utilizing the vector identity,
dh
= (w
r × r) · pr (31)
dt

r and r will be in the rs

From the relationships derived earlier, the cross product between w
direction and r = rr̂, the expression can be simplified to
dh
= rp s = rpr · rs (32)
dt

It becomes clear that the variation of angular momentum depends solely on the transverse
perturbation components, meaning that only the local horizon component of a perturbing
acceleration will change the angular momentum over time.
The change in eccentricity over time can be derived from differentiating the scalar form of
e,
h2 ( 2µ
e= 1+ v 2
− (33)
µ2 r

with the velocity vector and the relations ∂v2/∂rv = 2rv and ∂h2/∂rv = 2rh × r, the change in
eccentricity over time equates to
( (
∂e 1 2µ
= 2 h2v rr + hr v2 − + h2v srs (34)
∂rv µe r

Knowing that vs is the same as v⊥, and rearranging the orbit equation, the new expression
becomes
∂e h 1
= sin θrr + (h2 + µr)cos θ + µer rs (35)
∂rv µ µh

After substituting for r and pr, the final form can be shown as
de h 1
= sin θp r + (h2 + µr)cos θ + µer p s (36)
drv µ µh

Unlike the angular momentum, the eccentricity is affected by perturbations that lie only in the
orbit plane.
Variation of the true anomaly can be first expressed as

18
 dθ h
= 2 + ∂θ · pr (37)
dt r ∂rv

To solve for the partial derivative expression, the orbit equation combined with the radial
speed formula can be rearranged to obtain

∂θ ∂(rr · rv) ∂h
[(h2 − µr)cos θ + h(rr · rv )sin θ] =h cos θ + [(rr · rv)cos θ − 2hsin θ] (38)
∂rv ∂rv ∂rv

Simplifying the above equation and utilizing ∂(rr · rv)/∂v and ∂h/∂rv = (rh × r)/h leaves
(
∂θ h 1 h2 sin θ
= cos θrr − +r rs (39)
∂rv µe e µ h

Recalling that rs = wr × r and substituting the above equation into the original equation
yields the time variation of true anomaly due to a perturbing acceleration. Like eccentricity,
true anomaly is only affected by perturbations within the orbit plane.
( 2
dθ h 1 h2 h
= 2+ cos θpr − + r sin θps (40)
dt r eh µ µ

The right ascension of the ascending node can be found as the angle between the nodal line,
N, and the z-component in the inertial frame.
N
cos Ω = √ · ri (41)
N·N
It can also be represented in terms of the angular momentum vector and its components as
rh · ri
tan Ω = (42)
−
rh · rj

Taking the time derivative of the above equation and simplifying some algebra leaves

dΩ drh hxrj − hyri (43)

= cos2 Ω ·
h
dt dt 2
y

Because the angular moment vector is in the same direction as w

r , utilizing the transforma-
tion of w
r to the XY Z coordinate system leaves

dΩ drh h sin i(cos Ωri + sin Ωrj) 1 drh (44)

= · = N̂ ·
dt dt (hsin i)2 hsin i dt

where N̂ is the unit vector along the line of nodes. Substituting the cross product of r and
pr for the time derivative of rh and utilizing the vector identity between a cross and dot product,
the equation simplifies to
dΩ rsin u rsin u (45)
= r · pr =
w pw
dt hsin i hsin i
An orbits inclination can be expressed as the angle between the angular momentum vector

19
and the z-axis of the inertial reference frame

20
 rh · rk
cos i = (46)
h

Differentiating with time and using the definition of a dot product yields

di r
r cos i − rk) · (rr × pr)
sin i = (w (47)
dt h

Inserting the transformation of w

r to the XY Z coordinate system and interchanging the dot
and cross products leaves

di r rr
= sin Ω cos iri − cos Ω cos irj − sin irk × r · pr (48)
dt h

After taking the cross and dot products, the time variation of inclination simplifies to
di r
= cos u p (49)
dt h w

Looking at the relationship between the argument of periapsis and latitude,

ω = u− θ (50)

the variation of argument of periapsis can be expressed as

dω du
= − dθdt (51)
dt dt

Because the variation of true anomaly was solved earlier, computing the expression for
dω/dt relies on first solving for du/dt. Upon rearranging the radial unit vector r from the
transformation matrix found previously, the time rate of change for u can be expressed as

∂u ∂Ω
= −cos i (52)
∂t ∂rv

Substituting for ∂Ω/∂t and dotting with pr, the expression can be simplified to

∂u rsin u
=− w
r (53)
∂rv htan i
Inputting the results into the original equation and substituting the variation of true
anomaly, the variation of argument of periapsis can be expressed as
(
dω 1 h2 h2 rsin(ω + θ)
=− cos θpr − r+ sin θps − pw (54)
dt eh µ µ htan i

Unlike the other orbital elements, the argument of periapsis is affected by all three com-
ponents of the perturbing acceleration. This analysis is allows for the observation of different
perturbing forces and their effects on an orbits geometry. Whether a J2 perturbation, atmo-
spheric drag, or a tangential thrust, the perturbing acceleration variable can be modified to
simulate each case and show how orbital elements will drift over time due to outside forces.
When looking at the effects of J2 gravitational perturbations in particular, the perturbation

21
accelerations pr, ps, and pw in the Gauss variation equations can be substituted to form the
Gauss planetary equations.
dh 3 J2µR2 2
=
− i sin 2u (55)
dt 2 r3 sin
(
de 3 J2µR2 h2
= sin θ (3sin 2 i sin2 u − 1) − sin 2u sin 2 [(2i + ecos θ)cos θ + e] (56)
dt 2 hr 3 µr
dθ h 3 J2µR2 h2
= 2+ cos θ (3sin 2 i sin2 u − 1) + (2 + ecos θ)sin 2u sin 2 i sin θ (57)
dt r 2 hr3 µr
dΩ J2µR2 2
= −3 sin u cos i (58)
dt hr3
di 3 J2µR 2
=
−
dt 4 hr3 sin 2u sin2i (59)
dω 3 J2µR2 h2
= cos θ (1 − 3sin 2 i sin2 u) − (2 + ecos θ)sin 2u sin 2 i sin θ + 2e cos2 i sin2 u
dt 2 ehr3 µr
(60)

When looking at a satellite orbiting around earth with a radius of perigee of 6678 km,
radius of apogee of 9440 km, an inclination of 28 ◦, Ω of 45 ◦, and a ω of 30 ◦, the variation
of orbital elements due to J2 perturbations looks as shown in the following figure. While the
inclination and eccentricity oscillate slightly, the argument of perigee and right ascension have
a somewhat linear relationship as time increases. Based off these six orbital elements, other
elements such as seimmajor axis, eccentric anomaly, and mean anomaly can all be calculated
from these variation equations.

22
 Figure 13: Variation of orbital elements due to J2 perturbations over time [2]

4 De-Orbit Maneuver
Without thrusters to keep the spacecraft in orbit, CubeSats jettisoned from their deployers
follow a de-orbit trajectory. A de-orbit, or re-entry trajectory, is usually a slow orbit decay
from which the spacecraft will continue to lose altitude over time and eventually re-enter into
the atmosphere. Factors such as atmospheric conditions, ballistic coefficient, and propulsive
devices all contribute to the duration of the de-orbit. While the majority of cubesats will dis-
integrate during re-entry, some components containing heavy metals or heat tolerant materials
will not vaporize in this time period. Larger satellites perform precise burn to safely guide the
remaining debris away from people on the ground. However, a spacecraft with no propulsive
device offers no ability to target a location, posing as a potential danger. Because propulsive
devices are somewhat difficult to integrate with the limited storage capacity of CubeSats, it
becomes a challenge to steer hazardous debris away from the mainland. One option becoming
increasingly attractive is drag modulation. [11, 3, 10]

23
4.1 Drag Modulation
Drag modulation aims at varying the drag of a spacecraft in order to increase or decrease
the duration of the mission. This can be done by changing the surface area exposed to the
atmospheric drag, thus increasing or decreasing its air resistance. The ballistic coefficient, or
a measure of an objects ability to overcome air resistance, can be used as a benchmark for
modulating the drag.
β= m (61)
cDA

The ballistic coefficient is inversely proportional to the negative acceleration, meaning that
a high number represents a lower drag on the body. Increasing the effective surface area
will then decrease the ballistic coefficient, effectively decreasing the duration of the mission.
TechEdSat’s solution to drag modulation is with an Exo-Brake. Similar to a parachute, the
Exo-Brake is a drag device capable of changing surface area through a single winch design.
By retracting the winch, two of the sides pull in to decrease the surface area and increase the
ballistic coefficient. As shown in the following figure, precise winch control allows for various
surface area configurations.

Figure 14: Potential Exo-Brake configurations through a winch system [3]

Using the Program to Optimize Simulated Trajectories II (POST2), a team at NASA Lan-
gley was able to simulate TechEdSat 4’s end-to-end, or ISS to Earth surface, trajectory using
various Exo-Brake deployment configurations. POST2 is a 6-degree-of-freedom, generalized
point mass, rigid body, discrete parameter targeting and trajectory optimization simulator ca-
pable of simulating up to 20 independent or connected rigid bodies. However, for their analysis,

24
the attitude was constrained to an angle of attack and slideslip angles of 0◦, essentially modeling
the satellite as a 3-degree-of-freedom vehicle. By limiting the movement, the only aerodynamic
force acting on the body can be assumed to be the axial force.

Figure 15: Potential Exo-Brake configurations through winch system [3]

As expected, the highest axial force is generated when the Exo-Brake is fully deployed and
is lowest at 60% (0.76 of maximum surface area) deployed, the maximum retractable state of
the drag device. At higher altitudes, the axial forces become relatively constant due to the
exponential decay of density values.

Figure 16: Days since ISS departure for both 60% and 100% deployments [3]

25
 In addition to targeting re-entry positions, drag modulation can be used to speed up the
duration of the mission. As shown in the figure above, the Exo-Brake can alter the time until
re-entry by 7-8 days depending on the deployment configuration and atmospheric conditions.
This indicates that modulating the drag can provide a reasonable window of time for a guidance
scheme without the use of a propulsion device, given that a typical orbit of previous TechEdSat’s
were around 90 minutes.
The research team at NASA Langley used a predictor-corrector type of algorithm for an
end-to-end trajectory in the POST2 software to simulate Exo-Brake de-orbit targeting. This
algorithm is based on the cost function of the run, a projected-gradient based optimizer that
modifies the deployment time and level until the optimum trajectory is achieved. A basic
predictor-corrector algorithm is designated to integrate ordinary differential equations to find
an unknown function that satisfies a given differential equation. Each algorithm can be devised
into two main steps:

1. The initial, ”prediction” step begins from a function fitted to the function-values and
derivative-values with initial conditions and extrapolates, or anticipates, the said function
at subsequent points.

2. The next, ”corrector” step refines the initial approximation from the predicted value of
the function with an additional method to interpolate the unknown functions’ value at
identical subsequent points.

While the exact algorithm used in the POST2 software was not provided, the following
figure shows the result of a 1200 case Monte Carlo simulation with minimum drag modulations
aiming for a 50 km circle around the target. As shown, 87% of the cases landed within the
desired range with only 7 total drag deployment sequences. However, increased accuracy can
be achieved by allowing more modulations at the cost of an increased chance for failure.[3]

Figure 17: Landing points from a Monte Carlo simulation [3]

26
4.2 Gravity Gradient and Aerodynamic Torque
At higher altitudes, the atmospheric torque generated by the Exo-Brake will be insufficient to
overcome the torque due to the gravity gradient. The gravity gradient torque results when
portions of a non-symmetric satellite are subjected to a slightly larger forces the closer it is
to the central body. This force imbalance creates a tendency for the satellite to orient with
its long axis pointing towards the center of the central body. The worst case gravity gradient
torque can be modeled as follows:
3µ
Tg = |Iz − Iy|cos(2θ), (62)
2R3

where Iz and Iy represent the maximum and minimum moments of inertia, R is the radius
from the center of the Earth to the center of gravity of the spacecraft, and θ is the maximum
deviation of the z-axis from the local vertical with a worst case scenario of 45 ◦. The gravity
gradient torque acting on T5 (3U) and T7 (2U) varies with altitude as shown:

Figure 18: Gravity gradient torque acting on T5 and T7 [9]

In order for the drag device to provide any stability or control capabilities, the aerodynamic
torque must be able to overcome the gravity gradient torque. The aerodynamic drag force can
be universally defined as:
1
F = ρC AV 2 (63)
drag
2 D

The torque can then be defined as the the aerodynamic forces times the lever arm, which is
the distance between the satellites center of mass and the centroid of the drag force. Five sce-
narios were tested to simulate various Exo-Brake configurations and orientations at -90◦angle
of attack: a normal inflated ”flat-top”, vane strip surface area, full hemisphere, T5 surface
area, and a normal T7 respectively. An AoA of -90◦was chosen as the worst case orientation
for controllabilty due to the smallest possible reference surface area collecting drag. The corre-
sponding drag forces and aerodynamic torques were computed as a function of altitude using
the JB 2008 atmospheric model.

27
 Figure 19: Five Exo-Brake scenarios for various surface areas and configurations [9]

Figure 20: Exo-Brake aerodynamic drag force and torque for various surface areas and
configurations

As expected, the Exo-Brake with the highest effective surface area produces the most drag
and torque. In order to broadcast via Iridium’s forward pointing patch antenna, the satellite
must be pointed in the direction of the velocity vector, a meta-stable orientation for which
the satellite was designed for. However, if the satellite gets caught in another meta-stable
orientation, communication with the satellite can be lost, resulting in a loss of mission. Thus,
the Exo-Brake must have the ability to provide enough torque to overcome the gravity gradient
and reorient the cubesat.

28
 Figure 21: Intended meta-stable orientation for TechEdSat’s forward pointing antenna [9]

Figure 22: Gravity gradient plotted over aerodynamic torque for various surface areas and
configurations

A lower ratio indicates that the aerodynamic torque produced from the Exo-Brake is able
to overcome the gravity gradient and reorient the spacecraft while a high ratio suggests that
the gravity gradient will dominate and render the Exo-Brake ineffective. Despite the ”flat-top”
Exo-Brake being ineffective until around 300 km, the other configurations provide significant
aerodynamic torque past 400 km. This indicates that at higher altitudes, it may not be possible
to perform a handshake, an automated process between the on-board modem and the satellite
network that dynamically sets parameters to establish a communication channel. While a sec-
ondary antenna mounted on the side of the cubesat would allow additional broadcasting, there
is no current method of controlling the cubesats roll orientation, leaving this as an unreliable,
yet attractive secondary option if both are integrated into the system. Because TechEdSat is
focused on both quick payload returns and controlled de-orbits, a secondary patch antenna is
mounted on the side to increase the chances of successful handshakes per orbit.

29
4.3 De-Orbit Propagator
Before re-entry targeting can be achieved, a reference trajectory of a fixed Exo-Brake surface
area can be used to better understand the dynamics of the de-orbit process. Building upon the
governing equations of motion found the previous sections, a system of first order differential
equations can be presented as
∂V µx 3 RE 2 z2

ẍ = = (1 − J2 ( ) (5 − 1)) − adrag,x (64)

∂x r3 2 r r2

∂V µy 3 RE 2 z2

ÿ = = (1 − J2 ( ) (5 2 − 1)) − adrag,y (65)

∂y r3 2 r r

∂V µz 3 RE 2 z2

z¨= = (1 − J2 ( ) (5 − 1)) − adrag,z (66)

∂z r3 2 r r2

where

CDAρ
adrag,x = − (x˙ + ω ex)2 (67)
m
CDAρ
adrag,y = − (y˙ − ωe y)2 (68)
m
CD Aρ 2
adrag,z = − ż (69)
m

In order to decrease the computational expense, only three degrees of freedom (3DoF) were
utilized compared to the usual six. While fixing the vehicles attitude to an angle of attack and
slideslip angle of 0◦ further constrains the problem, the results are expected to yield accurate
estimates due to the theorized lack of large-scale attitude dynamics during an actual flight.
Using Matlab’s ODE45, a simple ordinary differential equation solver, the position and velocity
vectors of the spacecraft can be propagated until an altitude of 150 km is reached. After each
timestep, the orbital parameters are recalculated to provide an updated initial condition for
the next integration. While other perturbations such as solar radiation pressure, oceanic tides,
and additional celestial bodies all contribute to the orbital lifetime of the satellite, the J2 and
atmospheric drag accelerations are the most significant. As previously mentioned, the Jacchia
Bowman 2008 atmospheric model was used to generate the density at specified longitudes,
latitudes, and altitudes for the given day and year of the mission. Using the beginnings of a
source code from the Mathworks website, a function was created inside the differential equation
to continuously calculate the appropriate density at the current latitude, longitude, and altitude
[7].
The accuracy of this propagator can be validated by recreating test data from TechEdSat
4. Deployed in March 2015, TES4 had a pre-flight estimated ballistic coefficient of 8m2kg in the
desired orientation while taking just over 29 days to de-orbit. However, because the satellite

30
will oscillate around a meta-stable orientation and not remain constant, the ballistic coefficient
will vary along with the effective reference area exposed to the flight path. Thus, in order to

31
accurately capture the satellite’s de-orbit trajectory, a ballistic coefficient range of 7 ±m22 kg will
be used to duplicate the TES4 test data. Despite not having on board inertial measurement
units or Global Positioning Systems (GPS), all flight data was retrieved via TLE’s and based
on ground range estimates taken every couple of days.

Figure 23: TES4 (2015) trajectory reconstruction comparing POST2 [3] (right) and the
current MATLAB propagator

After reconstructing the 2015 TechEdSat 4 mission, a ballistic coefficient of 7m2kg was found
to be accurate to within a day of the the actual re-entry time. De-orbiting in just under 30
days, the propagator closely matched the flight data of 29 plus days. Additional tests were run
to compensate for the potential ballistic coefficient ranges and give a set of bounds where the
satellite would re-enter based off slight angular oscillations.
The possible effectiveness of the Exo-Brake can be tested by running two simulations with a
minimum and maximum drag configuration. Under the maximum reference area, the ballistic
coefficient approaches 8mkg2 compared to a 60 % deployed ballistic coefficient of 13
m2
kg
. Starting
on April 10, 2014 at an altitude of 400 km, the fully deployed Exo-Brake re-entered almost 13
days before the partially deployed satellite. Given that the satellite has an orbital period of
roughly 90 minutes per orbit, the partially deployed Exo-Brake CubeSat will experience over
200 more orbits, giving the satellite a generous amount of time to target any location within
the orbits path.

32
Figure 24: Difference in days to de-orbit for a minimum and maximum Exo-Brake deployment

4.4 De-Orbit Targeting Via Aerodynamic Drag

Performing a series of drag modulation maneuvers makes targeting a specific latitude and
longitude possible through various optimization algorithms, provided the desired location lies
on the initial orbit path. However, due to the limited accuracy of drag modulation, a tolerance
of a 100 km radius will be used as as success criteria for point targeting. In addition, the total
number of modulations should be minimized as much as possible to lower the risk of mechanical
failure. A terminal altitude of 150 km was used to provide additional time for the final stage,
a guided parafoil, to safely steer the satellite remains to the target location.

4.4.1 Analytically Calculating Re-Entry Location Based on Applied Controls

Using the approach suggested by Sanny Omar and Riccardo Bevilacqua [14], a targeting algo-
rithm of a satellite can be computed by first propagating an initial trajectory and analyzing
perturbations from a set of initial conditions. By mapping an initial trajectory to the final im-
pact location, rapid calculation and testing of modulating control parameters can be evaluated
until a desired latitude and longitude are obtained. However, before the control algorithm can
be discussed, an outline of the effects of orbit perturbations must first be analyzed. Starting
from the Gaussian Variation of Parameters equations from Vallado’s book[13], the change in
semi-major axis can be approximated as follows:
da 2 p
= √ [esinθFr + adrag] (70)
dt n 1 − e2 r

where the mean motion,

I µ
n= (71)
a3

33
 By Simplifying to a circular orbit around a spherical Earth, the new change in semi-major
axis with respect to time becomes √
da 2adrag a3
=
(72)
√
dt µ

Substituting the acceleration due to drag, the velocity of a circular orbit, and rearranging to
solve for the time derivative results in the following equation:
β da
− √ = dt (73)
2 µa ρ

Integrating equation (73) from an initial semi-major axis, a0, calculates the total time
required for a spacecraft in a circular orbit to reach a final semi-major axis, af .
f af
β da
∆t = − √ (74)
a0 2 µa ρ

If the time required to reach af with a ballistic coefficient of β1, then the relationship for
the time required to reach the same final semi-major axis can be expressed as
β2∆t1
∆t2 = (75)
β1

In addition, the mean motion for a circular orbit can be expressed as the time rate of change,
or true anomaly, θ.
dθ
n= (76)
dt

Multiplying by equation (73) and substituting n with equation (71) results in the following:
β da
dθ = − (77)
2a2ρ

Integrating with respect to an initial and final semi-major axis will result in the change in
true anomaly over the provided interval.
f af
β da
∆θ = − (78)
a0 2a2ρ

The average orbital angular velocity can then be expressed as the ratio between equation
(78) and (74).
∆θ
ωavg = (79)
∆t

Because the effects of zonal harmonics are to be considered, the average rate of change in
precession, assuming a circular orbit, can be determined by the following equation:
√
Ω̇ = 3 µJ2R2 () (80)
e
avg − cos i
2 (1 − e2)a7/2

34
 If the trajectory has previously been numerically propagated, then Ω̇ avg can be calculated by
dividing the total change in right ascension by the total change in time. If given Ω̇ avg , then the
total change in right ascension can be approximated by

∆Ω = Ω̇avg ∆t (81)

The initial trajectory is propagated using initial conditions with a predetermined ballistic
coefficient, β10 until ts,old, the initial time until swapping to β20, and is propagated until the
final semi-major axis is reached. This is called the terminal point, which can be characterized
by the change in time, ∆tterm, the change in true anomaly, ∆θterm, and the change in right
ascension, ∆Ωterm. The new terminal point location is then analytically determined for each
set of control parameters at this point to estimate the new de-orbit location based off the
updated ∆θterm and ∆Ωterm variables. The location of the terminal point required to target a
specified de-orbit location can then be uniquely determined based off the approximations that
the orbits inclination does not change significantly due to environmental perturbations and
that ∆θterm, ∆Ωterm, and ∆t are approximately the same for each new trajectory. The
purpose of the targeting algorithm then becomes to define a new trajectory which passes
through the terminal point that guarantees a re-entry at the desired latitude and longitude.
During the initial propagation, the time, position, velocity, and average orbital angular
velocity at each step are recorded from the initial time, t0, to the current time, t, and from
the current time to the terminal time, tterm.However, one previous assumption made was
that the argument of perigee would not change between the initial and final orbits. While
this is not particularly true, the argument of perigee can be approximated by utilizing the
mean longitude, the point at which an orbiting body could be founds if its orbit were
circular and perturbation free.

λ = Ω + ω + Me (82)

where,

Me = E − esinE (83)
1−e
( (
θ
E = 2tan−1 tan (84)
1 +e 2

Mathematically, the mean longitude is the angular distance the body would have from the
reference direction if it were to move at a uniform speed. Because the mean longitude increases
uniformly with time, the mean longitude at future point can be described as

λ = λ0 + n(t − t0) (85)

where λ0 is the mean longitude at epoch, n is the rate of motion in mean longitude, and t0
is the time at epoch. Similarly, the mean motion at a future point can be equated as

35
 - − t0 )
M = M0 + n(t (86)
where M0 = λ0 − (Ω0 + ω0) (87)
- = n − (Ω − ω)
n (88)

The terminal location of the spacecraft can be calculated from equations (75), (79), and
(81) using the same initial conditions and varying ballistic coefficients, β1 and β2, and the time
until these ballistic coefficients are swapped (ts). Determining the terminal point can then be
broken down into three different phases, each with respect to an initial and final semi-major
axis with the same ballistic coefficients between a0 and af . By keeping the ballistic coefficients
constant for the given semi-major axis intervals, the change in orbital elements for each new
trajectory can be observed for each phase. The following figure represents both the old and
new trajectories for varying ts, or the time when β1 swaps to β2.

Figure 25: Semi-major axis over time for both initial and new trajectories [14]

In this particular case, tequiv is less than ts,old, effectively lowering the time until af is
reached. Assuming that the only orbital elements changing are θ, a, and Ω, the total time and
orbital elements at de-orbit can be calculated and converted to the ECI and LLA (latitude,
longitude, altitude) frames for the final de-orbit latitude and longitude. Once this analytically
relationship between the initial conditions and the final latitude and longitude is established,
the effects of drag modulation can be analyzed for point targeting.
Using the analytical mapping between the control parameters (β1, β2, and ts) and the de-
orbit latitude and longitude, the desired trajectory can be analytically calculated and decoupled
from the corresponding drag modulations required to achieve point targeting. Assuming that
the desired target is below the orbits inclination and that the ballistic coefficient is modulated

36
(β1 and β2 are not equal), any location is possible to target given sufficient initial altitude and
ExoBrake surface area.
The change in true anomaly required for latitude targeting can be calculated by taking the
modulus of the difference between the initial and final angles.

∆θd = mod(φd − φi, 2π) (89)

where φi is the initial angle and φd is the final, desired angle. Analytically calculating ∆θd
to get an increase in swap time necessary to produce a desired ∆θd can be done by analyzing
the latitude controllability. Controllability is the ability to bring an initial state to a desired
final state in a finite amount of time from a given set of control parameters. If the system is
not controllable, then the satellite will be unable to reach the targeted latitude or longitude.
Parameters such as an insufficient initial altitude or too small of a drag configuration variation
can lead to inaccurate control of the spacecraft. Latitudinal controllability can be achieved by
first looking at the effects of only varying ts from an initial trajectory. Considering scenario
in which the time of swapping ballistic coefficients is increased, the second phase of the new
trajectory will have a different change in time and true anomaly from the old as shown in the
following figure.

Figure 26: The effects of an increase in swap time with respect to time in days [14]

If the time required for the second phase is ∆t20, then the time required for phase two in
the new trajectory can be calculated as
β1∆t20
∆t2 = (90)
β2

given that the semi-major axis interval and the ballistic coefficients do not vary. The total
increase in time caused from the increase in ts is equal to the difference in change of swap time

37
and the time required for phase two of the initial trajectory. Combined with equation (90), the
difference in orbital lifetime between new and old trajectories can be rearranged as
β2
∆t = ∆t (1 − ) (91)
d s
β1

If β1 is greater than β2, then the orbital lifetime of the spacecraft will decrease given an
increase in ts due to the decrease in drag. Inversely, if the ballistic coefficient decreases after a
decrease in ts, then the change in swap time would be equal to the negative time required for
phase two of the initial trajectory. The difference in total true anomaly change can then be
calculated once the change in orbital lifetime is determined as

∆θd = ω2,avg ∆td (92)

where ω2,avg is the average angular velocity calculated during the second phase from the
initial trajectory. This is valid because all of the orbital lifetime variations and changes in
true anomaly occur in this phase. However, there is a maximum time for which the change
in ballistic coefficients can take place. If ts is set to its maximum value, then the ballistic
coefficient will not change until the terminal point is reached. The maximum time can then be
written as
β1
ts,max = ts,old + (tterm,old − ts,old ) (93)
β2

While the latitude targeting algorithm must select a ts value less than the maximum, it is
important to enforce additional constraints due to longitudinal controllability being limited by
times farther away from the middle of the attainable range. Therefore, only ts values between
25% and 75% of the maximum will be used as valid swap times.
When calculating the optimal swap time for latitude targeting, a sweep of possible swap
times can be performed to better understand its effects on the terminal location. As shown in
the following figure, a wide range of locations can be targeted just from varying the swap time.

Figure 27: Possible terminal locations from varying only the swap time

In order to determine the optimal swap time, a for refined search of the latitude error can
be conducted. Every time the latitude error, actual latitude minus the target latitude, has a

38
sign change, somewhere between the two points yields the correct latitude its corresponding
swap time. A simple algorithm was constructed to look at all possible points within a ± of
10◦that underwent a sign change.

Figure 28: Sign Changes in swap times used to refine latitude targeting

From these potential swap times, and expanded search can be conducted to determine the
time the latitude error becomes zero. However, because the onboard hardware will not be able
to effectively implement winch controls in factions of seconds, the minimum latitude error was
selected from one second intervals.

Figure 29: An expanded latitude search to determine the minimal latitude targeting error

When choosing the optimal swap time, the value that yields the minimum positive longitude
error must be chosen. A positive longitude error indicates that the satellites orbit must last
slightly longer with the same total change in true anomaly to hit the desired target longitude
while a negative error would have to last slightly shorter. Positive longitude errors are desir-
able because the initial control parameters are designed to make the orbit last longer without
changing the total change in true anomaly.
Once the best possible ts time is determined, the initial and final ballistic coefficients must be
varied to eliminate the longitudinal error without disrupting the total change in true anomaly.

39
The effects of modulating ballistic coefficients on impact location, assuming the drag configu-
rations are swapped at the same semi-major axis in both the new and old configurations, can
be calculated as

∆θ1 + ∆θ2 = ∆θt (94)

∆t1 + ∆t2 = ∆tt (95)
∆ β
∆θ1 = 10 1 (96)
β1 0
∆θ20β2
∆θ2 = (97)
β20
∆t10β1
∆t1 = (98)
β20
t20β2
∆t2 = (99)
β20

where ∆tt is the total time the new trajectory takes to reach the terminal point, ∆ti is the
time until swap point in the new respective trajectory, and ∆θt is the desired total change in
true anomaly of the new trajectory until the terminal point is reached. Using equations (94) -
(99), beta1 and β2 can be rearranged to analytically calculate the desired ∆θt and ∆tt.
∆θ = ∆θ + ∆θ = ∆θ10β1 + ∆θ20β2 (100)
t 1 2 β10 β20
∆θ10
β1 = ∆θ (101)
β10β2( β2t − ∆θβ20 )
20

∆t10β1 ∆t20β2 (102)

∆t t= β10 + β20
β20(∆tt∆θ10 − ∆t10∆θt
β2 = (103)
∆t20∆θ10 − ∆t10∆θ20
Here, ∆θt will be the same as it is in the ts trajectory calculated from the latitude targeting
while the time to de-orbit and total change in true anomaly of the updated trajectory after the
terminal point will be identical to the initial trajectory after the drag profile is manipulated by
the targeting algorithm. Assuming the the swap times occur at the same semi-major axis for
the new and initial trajectories, it becomes important to update the new swap time, ts,new

ts,old β1
ts,new = (104)
β10

When optimizing the longitude error, a similar search to the latitude targeting algorithm
can be performed. By varying the total time the new trajectory takes to reach the terminal
point, the possible terminal locations can be shown as follows.

40
 Figure 30: Possible terminal locations from varying ∆tt

As in the latitude targeting algorithm, the optimal ∆tt can be determined by looking at the
sign changes in longitude error and expanding the search at one second intervals.

Figure 31: Expanded search for possible terminal locations by varying ∆tt

Given a specific β2, β1 must be calculated to ensure the difference in total change in true
anomaly between the initial and final trajectories. If, for example, the ballistic coefficient is in-
creased at the swap time by decreasing the reference area, then the orbital lifetime will decrease
due to the increase in drag. With the change in true anomaly calculated for the different phases
for the new trajectory, the true anomaly at the terminal location can be calculated using the
mean motion of the spacecraft and the change in true anomaly of the initial trajectory. Using
the relation between true anomaly, angular velocity, and time, the average mean motion of the
initial trajectories second phase can be calculated:
(
θ − θ10 rad
n2,avg = 20 (105)
t2 s

Because the spacecraft will have the same ballistic coefficients for phase two of the initial
trajectory, and the final phase of the new trajectory, the only differences will be the amount of
time and the true anomaly at the terminal location.
(
θ3 − θ2 rad
n2,avg = (106)
t3 s

Rearranging the equation solves for theta, which can then be used as on of the final orbital
elements to compute the position and velocity vector. Using the known variables at both

41
epoch and the terminal location, the final argument of perigee can be calculated. From here,
calculating the position and velocity vectors become trivial once all of the orbital elements are
known. Using algorithm 4.2 from Curtis’s book, the following pseudo code can be used to
compute the position and velocity vectors at the terminal location.

1. Calculate the distance

√
r= r·r (107)

2. Calculate the speed

√
v= rv · rv (108)

3. Calculate the radial velocity

rv
vr = r · (109)
r
4. Calculate the specific angular momentum:

rh = r × rv (110)

5. Calculate the magnitude of the specific angular momentum

/
h= rh · rh (111)

6. Calculate the inclination (

hz
i = cos−1 (112)
h
7. Calculate the node line
r = [0 0 1] × rh
N (113)
r
8. Calculate the magnitude of N
/
r ·N
r (114)
N= N

9. Calculate the right ascension of the ascending node

 (
Nx
cos
 − 1
N
(N ≥
y 0)
Ω= (
< 0) (115)
2π − cos−1 Nx
(N
 N y

10. Calculate the eccentricity vector

1 r 1 r
re = rv × rh − µ = rv2 − rv(rr · rv) − µ (116)
µ r µ r

11. Calculate the eccentricity

√
e= re · re (117)
42
 12. Calculate the argument of perigee
 (
 −1 N e
cos · (e ≥ 0)
N e

π ( z
ω = 2 − −1 N · 0) (118)
e (
 π
cos N e ez <

13. Calculate the true anomaly

 (
 · r ( (vr ≥ 0
cos−1 e )
θ = 2 − −1 e · r ( 0) (119)
e r
 π
cos e r vr <

Once the position and velocity vectors are known, they can be converted from the ECI
reference frame to a latitude, longitude, and altitude (LLA) coordinate system by first rotating
to the earth-centered inertial frame (ECEF). The earth-centered inertial frame is a geographical
coordinate system with its center located at the center of mass of the earth. Its axis are fixed
with respect to the surface of the earth and does not rotate about the z-axis like the earth-
centered internal frame. From the ECEF frame, the latitude, longitude, and altitude of a
given position vector can be computed with paired with a time stamp in the universal time
coordinated (UTC) format.
This targeting algorithm aims to find the optimal set of control parameters resulting in
a minimized latitude and longitude targeting error. After the optimal control parameters
are found, the initial trajectory can then be run with the new ts, β1, and β2 values. While
there may be some error between the propagated trajectory and the analytical solution due to
assumptions and unmodeled parameters, both will converge within a certain tolerance after a
few iterations. However, one of the biggest flaws in the analytical calculations was using an
exponential atmospheric model to generate density values. Despite decreasing complexity, the
accuracy of analytical perturbations suffer from this approximation. In addition, the MATLAB
propagation simulation took around 15 minutes to run, making it unattractive to run on-board
a CubeSat without first converting to C and optimizing. However, if run every few orbits to
continuously update the numerical propagation and coupled with a feedback loop, the targeting
algorithm can be used to guide the satellite to within 100 km of the target.

5 Results
In order to test the effectiveness of the ExoBrake’s ability to target, several simulations were
run at random potential deployment locations from an altitude of 300 km. The intended target,
NASA Wallops, was chosen as a targeting location due to their satellite tracking capabilities
necessary to help validate future TechEdSat experiments. The final swap time’s and total

43
targeting error’s were recorded as follows:

44
 Initial Orbital Parameters
(a (km), e, Ω(deg),ω Final Swap Time (sec) Targeting Error (km)
(deg), θ (deg), i (deg))
(6678.55, 0, 67.529, 58.417,
254832 21.39
0, 40.757)
(6672, .00029, 51.641,
121260 37.233
244.3, 220.82, 139.273)
(6678.58, .0245, 57.69,
103442 102.03
306.615, 314.19, 99.89)
(6678.55, 0.00035, 54.49,
128550 31.1322
201.01, 269.01, 80.99)
(6678.55, 0.01, 56.29,
106920 17.6835
224.35, 278.599, 270.21)

Figure 32: Final terminal locations from 5 randomly selected potential deployments

As shown in the figure about, all but one of the simulations yielded a targeting error under
the desired 100 km. In some cases, the varying of swap time alone was enough to maneuver
the satellite to within 50 km of the targeted terminal location. While these results determine
it is possible to target a location within 100 km, further tests should be run to validate the
accuracy and robustness of the algorithm before future flight use.

6 Limitations
Due to forces such as atmospheric drag, J2 perturbations, and gravitational forces from other
planets or masses, an initially circular orbit around Earth will not remain perfectly circular,
causing the velocity and altitude to increasingly oscillate as the spacecraft de-orbits and the
drag forces increase. The aerodynamic drag force slightly reduces the instantaneous velocity
of the initially circular orbit, causing the current point to become the orbits apogee. At the
perigee, the altitude will be lower while the density is higher, causing a higher velocity and drag
force when compared to the apogee. This then converts the old perigee into the new apogee,
creating a cycle which explains the oscillations in velocity with respect to altitude. As shown

45
in the figure below, the oscillations become greater the closer to Earth the spacecraft gets due
to the increase in drag.

Figure 33: Velocity oscillations over time during an orbit decay

Despite these oscillations being small, the velocity, density, and aerodynamic drag profiles
will differ from what they would be for a circular orbit and accumulate small errors over time
in the analytically perturbation analysis used to predict the new orbit. However, because the
targeting algorithm is based off perturbations form a numerically propagated trajectory using
a sophisticated density model, the drag force deviations from the circular orbit will be visible
in the numerical calculation.
In addition, a single drag modulation maneuver would not be feasible due to the unknown
changes in density or attitude of the spacecraft. One major assumption used in this simulation
stated that the spacecraft would remain in its highest drag form throughout the trajectory.
While this is true for most of the spacecrafts life, the ExoBrake might not be fully effective at
higher altitudes if the aerodynamic torque

7 Future Work
While this targeting method proves that the Exo-Brake can successfully guide the satellite
to a targeted location provided enough time and inclination in the initial orbit, a single drag
modulation will not be enough to account for unknown perturbations when predicting unknown
variables such as density or drag. As all simulations were run with preexisting information on
solar and geomagnetic indices, these unknown perturbations were accounted for, thus making
the simulation provide accurate results with a single maneuver. For this reason, a closed loop

46
solution will be implemented to ensure that the satellite follows the provided trajectory to
ensure the target location is achieved. One such solution is to break the control algorithm into
two distinct stages, thus allowing more optimal trajectory tracking without implementing a
high amount of drag modulations. The first stage would consist of the single swap maneuver
discussed in this paper until a target location at 200 km is reached. This 200 km location will
be determined to guarantee that the satellite will be within a window capable of reaching the
desired impact location while de-orbiting the satellite in a timely manner.

Figure 34: Stage 1 of proposed targeting algorithm

By limiting the number of drag modulations, the risk of failure in the Exo-Brake and
winch system is lowered, thus making the control algorithm safer compared to a continuously
modulating system. However, as stated before, small perturbations in the density will cause this
method to accumulate errors over time without closing the loop. The second stage, beginning
once the satellite reaches the 200 km mark, will close the feedback loop and allow the satellite
to adjust its deployment configuration on the fly.

47
 Figure 35: Stage 2 of proposed targeting algorithm

The starting point of stage two creates a window of opportunity based off the minimum
and maximum ballistic coefficients. Any further drag modulation combinations would create
a new branching point, as shown in the figure above. A closed feedback loop would lie within
the targeting window, allowing for the satellite to adjust its ballistic coefficient on the fly to
ensure the impact location is reached. This adaptive control method will help compensate for
the changes in orbital elements as well as density and drag variables.
Once the second stage of the proposed targeting algorithm is completed, the second stage
of SOAREX, a future testbed for the ExoBrake, will utilize a re-entry vehicle called the TDRV
(Tube Deployed Re-entry Vehicle) to protect the satellites internal components during the re-
entry process. Once re-entry is complete, a guided parafoil will be used to steer the satellite
remains to the desired location.

48
Figure 36: SOAREX Deployment Process proposed to work with the de-orbit targeting algo-
rithm

In order to run the propagator and control algorithm onboard the satellite, the code must
also be optimized and converted to C++ to decrease the run time and computational expense
of the model. Optimizing the numerical propagation will allow the trajectory to be updated
every few days, helping decrease the error accumulation seen in the first stage of the proposed
algorithm. Including higher order accuracy conditions such as lunar gravitational and solar
pressure effects can also be added to increase the numerical calculations accuracy at the cost of
longer convergence times. However, because the de-orbit process is relatively slow, additional
computation time can be tolerated so long as the spacecraft can supply the increase in power
consumption.

8 Conclusions
Using drag modulation, it is possible to guide a CubeSat or small satellite to any location on the
globe, provided there is enough time and inclination of the initial orbit. By first numerically
propagating the trajectory using an advanced atmospheric model, analytical perturbations
can be performed based off simplified assumptions to predict when and where the satellite
will reach its terminal location, or a desired altitude of 150 km. This can be used to safely
guide satellites and small payloads from the ISS to within 100 km of a desired location at an
increased rate without endangering the population. Once the terminal location is reached, the
proposed SOAREX stages will be implemented for the re-entry and guided parafoil portion
of the trajectory. While further investigation of the targeting algorithm will be performed

49
to perfect the maneuver through a multi-stage approach, drag modulation can be seen as an
attractive and inexpensive alternative to propulsion systems for small satellite recovery.

50
References
[1] Bruce R. Bowman and W. Kent Tobiska. “A New Empirical Thermospheric Density
Model JB2008 Using New Solar and Geomagnetic Indices”. In: AIAA (2008).
[2] Howard D. Curtis. Orbital Mechanics for Engineering Students. 225 Wyman Street, Waltham,
02451, USA: Elsevier Ltd., 2014.
[3] Soumyo Dutta et al. “Guidance Scheme for Modulation of Drag Devices to Enable Return
from Low Earth Orbit”. In: AIAA (2016).
[4] E.M. Gaposchkin and A.J. Coster. “Analysis of Satellite Drag”. In: The Lincoln Labora-
tory Journal 1.2 (1988).
[5] E.M. Gaposchkin and A.J. Coster. “Linearized Equations for J2 Perturbed Motion Rela-
tive to an Elliptical Orbit”. In: SJSU Master’s Thesis and Graduation Research Depart-
ment of Mechanical and Aerospace Engineering (2005).
[6] Maj Edward P. Chatters IV and Maj Bryan Eberhardtand Maj Michael S. Warner. “Or-
bital Mechanics”. In: (2010), pp. 89–114.
[7] Meysam Mahooti. Jacchia-Bowman Atmospheric Model. 2016.
[8] Gustavo A. Mansilla. “Solar Cycle and Seasonal Distribution of Geomagnetic Storms with
Sudden Commencement”. In: Earth Science Research 3.1 (2014).
[9] M. Murbach, F. Tanner, and S. Smith. T5 Mission Anomaly Analysis. NASA Ames
Research Center. 2017.
[10] NASA. Guidance and Navigation for Entry Vehicles. National Aeronautics and Space
Administration, 1968.
[11] Sanny R. Omar and Riccardo Bevilacqua. “Spacecraft De-Orbit Point Targeting using
Aerodynamic Drag”. In: AIAA Guidance, Navigation, and Control Conference (2017).
[12] Prasenjit Sengupta. “Satellite Relative Motion Propagation and Control in the Presence
of J2 Perturbations”. In: Office of Graduate Studies of Texas AM University (2003).
[13] D.A Vallado. Fundamentals of Astrodynamics and Applications. Springer-Verlag New
York, 2007.
[14] David A. Vallado and David Finkleman. “A Critical Assessment of Satellite Drag and
Atmospheric Density Modeling”. In: AIAA/AAS Astrodynamics Specialist Conference
and Exhibit (2008).
[15] Josep Virgili, Peter C. Roberts, and Nathan C. Hara. “Atmospheric Interface Reentry
Point Targeting Using Aerodynamic Drag Control”. In: Journal of Guidance, Control,
and Dynamcis 38.3 (2015).

51