SSC09X6

BoeingsCubeSatTestBed1AttitudeDeterminationDesignandOnOrbit
Experience

MichaelTaraba
PrimaryAuthor,FormerEmployeeoftheBoeingCompany

ChristianRayburn;SeniorEmbeddedSoftwareEngineer
AlbertTsuda;SystemsEngineer
CharlesScottMacGillivray;ProgramManager
5301BolsaAve.MCH013B322,HuntingtonBeach,CA92647
christian.g.rayburn@boeing.com
albert.s.tsuda@boeing.com
charles.s.macgillivray@boeing.com

ABSTRACT

TheCubeSatstandardhasprovidedspaceaccesstorapidlyacceleratethematurityofhardwarecomponentsand
software algorithms for extremely small satellites. The Boeing CubeSat TestBed 1 (CSTB1) onorbit experiment,
launched April 17,2007, validated a highly integrated and multifunctional approach for attitudedetermination.
ThispapercoverstheconstraintsanddesignconceptofaCubeSatattitudedeterminationsystemusingmultiple
integrated sensors. The onorbit data collected from five twoaxis commercialofftheshelf MEMS magnetome
ters,andfoursuitesofsunsensorswasprocessedandanalyzedtodeterminetheattitudeofCSTB1.Theattitude
determinationwasverifiedviaanimagefromalowpowerCMOScameraandsolarcellmeasurements.Lastly,this
paperaddresseshowourattitudedeterminationsolutionwasusedtohelprefinevehicleoperations.

INTRODUCTION exist to provide a threeaxis attitude solution for

BoeingsCubeSatTestBed1(CSTB1)waslaunchedinto
a 750 km sunsynchronous orbit on April 17, 2007 by
an ISC Kosmotras Dnepr rocket from the Baikonur
CosmodromeinKazakhstan(Reference1).Theprimary
mission goal of CSTB1 was to accelerate the maturity
of CubeSat related components and subsystems, as
well as accelerate the general infrastructure and
operationsofthisclassofspacecraft.
CubeSats.TheCSTB1attitudedeterminationapproach
utilizes multifunctional elements and commercialoff
theshelf electronics as a means of providing course
attitude knowledge, while addressing the ultralow
size,weightandpowerconstraints.
Thispaperisstructuredasfollows:
- CubeSatTestBed1GeneralDescription

The CubeSat standard (Reference 2) has enabled low - SimpleAttitudeDeterminationasPartofMul

cost access to space by utilizing rideshare opportuni tifunctionalElements
ties; however, the one kilogram and 10 cm cube
- SimulationResults
definitionshaveintroducednewchallengesinsatellite
design.Mosttypical,flightqualified,satellitehardware - OnOrbitDataAnalysis
are not compatible due to their mass, power and/or
volume requirements; therefore, few attitude sensors - Conclusions

Rayburn 1 23rdAnnualAIAA/USU
DoDRef:09S0772 ConferenceonSmallSatellites

CUBESATTESTBED1GENERALDESCRIPTION
The primary mission goals of CSTB1 were to mature
andevaluatecommerciallowpowerprocessors,CMOS
ultra low power imagers, and associated software
algorithms in the space environment, and to provide
validation of using integrated attitude determination
sensors as part of the multifunctional side panels.
Secondary missions included validation of the mission
operationscenterandrapidprototypingprocessesina
smallteamenvironment.
CSTB1 conforms to the CubeSat standard for a single
1U CubeSat. CSTB1 consists of eight basic functional
subsystems: 1) Command and Data Handling, 2)
AttitudeDeterminationandControl,3)Telemetryand
Command 4) Electrical Power System, 5) Structure, 6)
Mechanisms, 7) Thermal management, and 8) Bus
Flight Software. Many of these functional subsystems
are physically integrated together as multifunctional
elements. For example, the attitude determination
sensorsandsolarcellsareintegratedonasidepanel,
PCB boards, which also act as structural elements for
CSTB1.IntegrationoftheCSTB1elementsintoa10cm
cubeisshowninFigure1.
Figure1:CSTB1subsystemelementintegration
SIMPLE ATTITUDE DETERMINATION AS PART OF
MULTIFUNCTIONALELEMENTS
The attitude determination concept for CSTB1 con
sisted of four sun sensor suites and five twoaxis
Rayburn
DoDRef:09S0772
magneticsensorsintegratedintothesidepanelsofthe
satellite.Thegoalwastoprovideenoughsimplesensor
measurements so that a reasonable coarse attitude
solution could be derived. Given that four of the six
sides of the satellite contain at least two axes of
attitude information; course attitude information can
beextractedformanyattitudes.Allsensordatacanbe
directlydownlinkedtothegroundforpostprocessing.
The remainder of this section addresses the sensor
configuration and attitude determination algorithm
usedonthegroundforpostprocessing.
CSTB1CoordinateFrames
ThecoordinateframedefinitioninFigure2isprovided
to give a reference when discussing the configuration
of the attitude sensors later in this section. The side
panels,labeledP0throughP5,aretheexposedsidesof
CSTB1, and each panel normal corresponds to a body
axis. Panel 5 (+Z axis) is the payload panel, and the
visiblecolorcameraisalignedwithitspanelnormal.
Figure2:CSTB1Coordinateframedefinitions
SunSensorSuites
There are four sun sensor suites that provide a mea
surement of the sun vector in body coordinates, and
arelocatedonsidepanels,P1,P2,P3andP4.Eachsun
sensor suite consists of four COTS photodiodes inte
gratedontothemultifunctionalsidepanel,seeFigure
3. Each of the four photodiodes is canted 45 degrees
from the normal of the side panel board and rotated
90degreesapartrelativetoeachothertoprovide2pi
steradian coverage. Their placement on the board is
selectedtominimizeinterfacefromthesolarcellsand
structureofthesatellite.
2 23rdAnnualAIAA/USU
ConferenceonSmallSatellites
 (1)

1.0

1.0

(2)

1.0 (3)
Figure3:SunSensorSuiteintegratedtoSidePanel

Opposingphotodiodepairshaveoverlappingresponse
curves,seeFigure4,andthisallowsadifferenceover
sumsmethodtodefinethelocationofthesuninthat
(4)
axis(Reference3).Forexample,assumeapanelwhose
normal is the Z axis, photodiodes 1 & 2 oppose each
1.0
other(intheXZplane)andphotodiodes3&4oppose
eachother(intheYZplane),thesunvectorcanthen 1.0
be defined by equation 1. On CSTB1 each sun sensor
suitehasadifferentnormalvectorwhichcorresponds (5)
toabodyaxis;thevectorcomputationsareuniquefor
eachsuiteandaredefinedinequation2through5.
TwoAxisMagneticFieldSensor

Five of six side panels have a COTS magnetoresistive

type twoaxis magnetic field sensor. The performance
characteristics for the magnetic field sensor are listed
in Table 1. Table 2 lists the measured axes for each
panelfromthemagneticfieldsensors.

Table1:MagnetoresistiveChipCharacteristics

Characteristic Units Value

NoiseDensity nV/rtHz 50
Resolution MicroGauss 120
Orthogonality Degrees 0.01

Sensitivity mV/V/Gauss 1.0
Figure4:PhotodiodeResponsetoSunAngle

Rayburn 3 23rdAnnualAIAA/USU
DoDRef:09S0772 ConferenceonSmallSatellites

 Table2:MagneticFieldinBodyAxesMeasuredby AttitudeDeterminationAlgorithm
EachSidePanel
Theattitudedeterminationalgorithmwasusedtohelp
SidePanel MeasuredBodyAxes correlatevehicleattitudebyusingthesunandmagnet
+X ic field sensor readings downloaded from CSTB1 with
P0
+Z images taken from the onboard visible camera. This
Y algorithmstaticallycomputeseachattitudequaternion
P1
+Z estimatebasedontheavailabledataatthattimestep.
X The attitude quaternion estimate computations are
P2
+Z basedontheTRIADalgorithm(Reference4),areview
Y belowisprovidedforreference.
P3
X
The computation of the attitude matrix using TRIAD
Y
P4 requires two reference frame unit vectors, and
Z
(in this case Earth Center Inertial is used), and the

corresponding measured unit vectors, and , in

Thedatafromeachpanelisaveragedtogethertoform
the body frame. The attitude matrix, , rotates a
asinglemagneticfieldsensormeasurementproviding
vectordefinedintheECIreferenceframetothebody
threeaxesofinformation,seeequation6.
reference frame and therefore must satisfy equations
7a&7b.
3
(7a)

3 (6) (7b)

The algorithm calls for the creation of two sets of

4 columnvectors,whicharethereferencesetofvectors
given in equations 8 and the measurement set of
Prior to the launch of CSTB1, the magnetic field sen vectorsgiveninequations9.
sors were calibrated to the Earths magnetic field.
(8a)
These sensors were calibrated individually and at the
vehicle systems level. In order to compute the offset
| |
(8b)
resulting from integration into the vehicle, magnetic
fieldmeasurementsweretakenintwoorientationsfor
each panel. Since the largest component of the mag | |
(8c)
netic field at the location of the test was in a vertical
direction, the measurements were taken about this (9a)
axis. Ideally, the measurements should produce the
samemagnitudesincewearetakingmeasurementsof | |
(9b)
the same field in the opposite directions. The differ
ences in magnitudes were then evaluated to derive | |
(9c)
sensor offsets and incorporated into the sensor con
versions calculations. These sensor conversions were Theattitudematrixisdefinedinequation10.
applied prior to executing the attitude determination
processingontheground. (10)

Rayburn 4 23rdAnnualAIAA/USU
DoDRef:09S0772 ConferenceonSmallSatellites

Sincethissolutionisnotsymmetricinindices1and2,
part of the second measurement vector is discarded,
making the attitude estimated more heavily weighted
on the first measurement vector. To increase the
accuracy of the attitude estimation matrix, the TRIAD
solution was augmented with the optimized TRIAD
method presented in Reference 5. The optimized
TRIAD method computes two attitude matrices: the
first, ,usesthemagnetometerasthefirstmeasure
ment vector, and the second, , uses the sun vector
as the first measurement vector. The two attitude
estimates are combined based on their respective
standardofdeviations, and ,giveninequation11.
(11)
The attitude determination algorithm is broken into 4
parts: 1) magnetic field sensor processing, 2) sun
sensorprocessing,3)TRIADwithvalidsunandmagnet
ic field measurements and 4) attitude determination
withonlymagneticfieldmeasurements.
The magnetic field sensor processing is the simplest
sectionwhichonlycomputesthemagneticfieldvector
based on the sensor readings using equation 6. The
sunsensorprocessingsectioncomputesthesunvector
foreachsidepanel(P1throughP4)usingequations2
5.Givenfoursunvectors,asetofdeterministicrulesis
usedtodeterminewhichsunvectoristhemostaccu
rateandwillbeusedforattitudedetermination.1)The
side panel with the greatest number of photodiodes
above the minimum threshold is selected as the sun
vector.2)Ifthesidepanelwiththegreatestnumberof
photodiodesonlyhas1activephotodiodethenthesun
vector is set to invalid. 3) If none of the photodiodes
areactivethesunvectorissettoinvalid.
A 4th part of the attitude determination was created
because the satellite has an eclipse period and does
nothave4pisteradiancoveragewiththesunsensors.
Therefore, at times, the attitude estimate will not be
computable with the standard TRIAD algorithm. This
willbediscussedlater.
Rayburn
DoDRef:09S0772
The attitude determination uses the optimized TRIAD
algorithm when the sun vector is valid. Two attitude
matrices are computed, one using the magnetic field
vectormeasurementasthesourceforthefirstvector,
, and the other using the sun vector measurement
asthefirstvector, .Theattitudematrixisthenfully
computed using information from both attitude
matrices according to equation 11. The standard of
deviation ratios used are, 0.6and
0.4, when a sun vector measurement consists of 4
valid photodiodes. For cases when the sun vector
measurement consists of less than 4 photodiodes
measurements, then the attitude matrix, , is set to
.
The 4th part of the attitude determination processing
was created because the satellite does not have 4pi
steradian coverage with sun sensors during eclipse
periodsandwillresultinperiodsofinvalidsunvector
measurements.Inthesecases,theattitudedetermina
tion method uses only the valid magnetic field vector
measurement. This is done by computing the angular
rotation between the current magnetic field mea
surement and the previous magnetic field measure
mentinequation12.
cos (12)
Next, the direction of the magnetic field rotation is
computedinequation13.
(13)
The delta quaternion to the current attitude estimate
isthencomputedinequation14.
sin 2
(14)
cos 2
The delta quaternion estimate is combined with the
previous estimate to compute the current attitude
estimateinequation15.
(15)
5 23rdAnnualAIAA/USU
ConferenceonSmallSatellites
However, this solution is not accurate as it will be examined in the simulation. For each scenario pro
missinginformationinoneaxisandwillbesusceptible videdinTable3,thesatelliterotatesaboutavectorat
to significant attitude drift. For the intended purpose varying tumble rates with correlating sun vector
ofCSTB1,thissolutionwillsuffice. measurementunavailability.

SIMULATIONRESULTS Table3:CSTB1AttitudeDeterminationSimulations

TheinitialvalidationoftheCSTB1attitudedetermina SatelliteRate %SunSensorUnavai

Case
tion algorithms were simulated using MATLAB. The [deg/sec] lability
simulationincludedkeyelementsnecessarytovalidate 1 0.5 0
the attitude determination: orbital and spacecraft 2 0.5 30
dynamics models, a magnetic field model,math mod 3 0.5 75
els of the magnetic field, and sun sensors including 4 0.25 0
noisetermsandmisalignments. 5 0.25 30
6 0.25 75
Multiplesimulationcaseswereevaluatedtodetermine
7 0.1 0
an average expected performance of the attitude
8 0.1 30
determination algorithm based on CSTB1s configura
9 0.1 75
tion. Nine specific scenarios of performance were

Attitude Determination Errors With Continuously Valid Sun Sensor Measurements

10
Case 1
Case 4
5
Case 7

0
4000 4500 5000 5500 6000 6500 7000 7500
Attitude Determination Error for 0.5 Deg/sec Tumble Rate
50 Case 1
Attitude Error [degrees]

Case 2
Case 3

0
4000 4500 5000 5500 6000 6500 7000 7500
Attitude Determination Error for 0.25 Deg/sec Tumble Rate
50 Case 4
Case 5
Case 6

0
4000 4500 5000 5500 6000 6500 7000 7500
Attitude Determination Error for 0.1 Deg/sec Tumble Rate
50 Case 7
Case 8
Case 9

0
4000 4500 5000 5500 6000 6500 7000 7500
Time [seconds]

Figure5:SimulationAttitudeErrorforEvaluatedCases

Rayburn 6 23rdAnnualAIAA/USU
DoDRef:09S0772 ConferenceonSmallSatellites

ThesimulationresultsshowninFigure5demonstrate side panel X sensor and side panel Z sensor,
thevaryingaccuracyandsensitivityofthealgorithmto which were not used in the attitude determination.
sensor characteristics and satellite tumble rates. The Theinformationalsoindicatesthatthevehicleattitude
attitudeaccuracyisimpactedbytwoelements:invalid isoscillatingrelativetothemagneticfieldoftheEarth.
sun vector measurements and satellite tumble rate. It is suspected that the significant magnetic moment
The attitude accuracy is less than five degrees during createdbythewavedipoleantennahadalignedwith
periods of continuously valid sun vector measure theearthsmagneticfieldandthesatellitewasoscillat
ments. The increasing magnitude of the tumble rate ing about it. The fact that the satellite had a strong
increasesthemagnitudeoftheattitudeerrorwhenthe momentthatalignedwiththemagneticfield,allowed
tumble rate causes outages in the sun vector mea operation of CSTB1 to coarsely predict where the
surement. payload panel would be facing for operations of the
visiblecamera.
ONORBITDATAANALYSIS
4
x 10 Magnetic Field Sensor Measurements [nTelsa]

CSTB1waslaunchonApril17,2007intoa745kmsun 4

synchronous orbit. Onorbit operational life has re 3

Magnetic Field Measurement [nTelsa]

sulted in over one million various sensor data points 2

and 50 images. The flight data presented here 1

representsa90minutesnapshotofoneorbitofCSTB1 0

on June 17, 2007. This sequence of data is used to -1

draw valuable information regarding the attitude -2

determination system, and how the satellite is tum -3

bling. The measurements from the sun sensors on -4

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

panel 1 (Figure 6) show that the sensors saturate

Time, [sec]

quicklywhenexposedtothesun.Asaresult,muchof Figure7:CalibratedMagnetometerMeasurements
the valuable angular data that can be extracted from
thesunvectorwaslost. The attitude determination algorithm was run against
the flight data over the 90minute period shown in
1
PD 1
PD 2
Figure8.
0.9
PD 3
PD 4 1
0.8
Normalized Sun Sensor Measurements

q1
0.8 q2
0.7 q3
q4
0.6

0.6
0.4
Quaternion [unitless]

0.5 0.2

0.4 0

-0.2
Eclipse

0.3
-0.4
0.2
-0.6

0.1
-0.8

0 -1
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time, [sec] Time, [sec]

Figure6:NormalizedSunSensorSuitemeasurements Figure8:ComputedAttitudedeterminationQuater
fromPanel1 nion

The magnetic field sensors measurements in Figure 7 The image in Figure 9 was scheduled and captured
are the calibrated measurement values. Two of the from CSTB1 and was used as a source to validate the
field measurements provided incorrect readings, the attitude determination estimate. The image clearly
Rayburn 7 23rdAnnualAIAA/USU
DoDRef:09S0772 ConferenceonSmallSatellites

shows several significant Earth landmarks, which
include lakes, the Himalayan mountain range and
MountEverest.Inordertodeterminewhichlandmarks
are in the image and the accuracy of the attitude
determination solution, we computed where the
camerawaspointedatthetimetheimagewastaken.
The latitude and longitude of the subsatellite point
(46.9N, 94.5E) and the unit vector of the Z axis ac
quired from the attitude determination are used to
determinethelatitudeandlongitudeoftheimage.The
computed angle between the Z axis and nadir is 16.7
degrees which results in a latitude and longitude for
the image of 33.4 degrees North latitude and 81.8
degrees East longitude. The USGS Global Visualization
Viewer(Reference6)wasusedtoidentifythelakeand
mountain range landmarks from a LandSat 7 image
from April 2007, as shown in Figure 10. The green
outline in the LandSat 7 imagery represents the area
imagedbyCSTB1.Thelandmarkslocatedat29degrees
Figure10:USGSComparisonImage,Greenoutline
Northlatitudeand85.6degreesEastlongitudearealso
representstheCSTB1Image,Redoutlinesarethe
outlined in the image for clarity. Based on the image correspondingidentifyinglandmarks
latitude and longitude from the USGS, the pointing
angle off nadir was determined to be 18.9 degrees. CONCLUSIONS
This demonstrated a 2.2 degree error in the attitude
This paper demonstrated how simple, highly inte
determinationsolutionfromCSTB1.
grated, commercialofftheshelf components can be
used for CubeSat attitude determination. The opti
mized TRIAD algorithm implemented on the ground
usedthesunsensorsuitesandmagneticfieldsensors
datatodeterminethesatelliteattitudeandverifythe
location of an earth image taken by CSTB1s camera.
The magnetic field sensors proved to be the most
usefulofallthesensorsbecauseoftheirhighavailabili
ty. Attitude estimates derived for CSTB1 sensor data
was used to determine the impact of the magnetic
dipolecreatedbyCSTB1antenna.Thisallowedground
operatorstocoarselypredictthepointingdirectionof
the onboard camera and schedule collections of

imageryoverareasofinterest.
Figure9:ImagecapturedbyLowPowerImageron17
June2007(redoutlinesarekeylandmarksidentified BoeingsCSTB1CubeSathassuccessfullycompletedall
onthecomparisonimage) primary and numerous secondary mission objectives,
andhasprovidedextremelyusefuldatathroughoutits
20+monthsonorbit.

Rayburn 8 23rdAnnualAIAA/USU
DoDRef:09S0772 ConferenceonSmallSatellites

ACKNOWLEDGMENTS

Theauthorswouldliketothankallthevariouspeople
who contributed to the success of CSTB1. Specifically,
we would like to thank Arthur Frank Cooper, Chris
Day, Dan Minear, Phil Reid, Thanh Tang, and Doug
Yarbroughforalltheirdedicationthistoproject.

REFERENCES

1. CubeSat Community Website,

http://cubesat.atl.calpoly.edu/pages/missions/d
neprlaunch2.php

2. CubeSat Design Specification REV. 11,

http://cubesat.atl.calpoly.edu/media/CDS_rev1
1.pdf

3. Larson, W.J. and J.R. Wertz, Space Mission

Analysis and Design, 3rd Ed. Microcosm Press,
October1999.

4. Shuster,M.D.andS.D.Oh,ThreeAxisAttitude
DeterminationfromVectorObservations.AIAA
JournalGuidanceandControl,Vol.4,No.1,Jan
uaryFebruary1981,pp.7077.

5. Itzhack,B.Y.andR.R.Harman,OptimizedTRIAD
algorithm for attitude determination.
AIAA/AAS Astrodynamics Conference, San Di
ego,CA,July1996.pp.422427.

6. USGS Global Visualization Viewer,

http://glovis.usgs.gov

Rayburn 9 23rdAnnualAIAA/USU
DoDRef:09S0772 ConferenceonSmallSatellites