SSC03-X-7

Lightweight, Low-Power Coarse Star Tracker

Ray Zenick
AeroAstro, Inc.
P.O. Box 502
Solana Beach, CA 92075
858.481.3785
[email protected]

Thomas J. McGuire
MIT Space Systems Laboratory
77 Massachusetts Avenue, Room 37-350
Cambridge, MA 02139
617.253.8541
[email protected]

Abstract. General industry research on spacecraft attitude determination and control components is focused on
increasing accuracy to the exclusion of all else. This is contrary to the clear requirement for decreased complexity,
mass, and power consumption to fit the needs of small, maneuverable, low-cost spacecraft. To address this need,
AeroAstro and the Massachusetts Institute of Technology Space Systems Laboratory are developing a coarse star
tracker, which strikes an appropriate balance between accuracy, power consumption, mass and cost.

The coarse star tracker is targeted for small spacecraft, addressing requirements of a wide range of applications,
including orbit transfer and rendezvous missions. This market is presently inadequately served, because higher
accuracy star trackers simply do not fit within the mass, power, and cost constraints of small spacecraft missions.
The design described herein is a 300-gram unit that consumes less than 1 Watt of power. With a pointing accuracy
of better than 100 arc-seconds, the star tracker will enable a cost-effective three-axis stabilized spacecraft to attain
pointing accuracies to better than 0.25 degree, more than adequate for most low-earth orbit (LEO) missions. The
team is exploiting acquisition and tracking algorithm simplification and is developing ways to further reduce mass,
power consumption, complexity and cost.

Introduction During this investigation, it quickly became apparent to

AeroAstro that there was a market for a coarse star
Several years ago, AeroAstro began investigating tracker – something less precise than the high-end star
microsatellite attitude determination components for trackers, yet more capable than a sun sensor. A low-
missions with coarse pointing requirements. The small end star tracker, even with greatly reduced update rates
satellites investigated required extremely lightweight and roll rates, could be a viable product but must
attitude determination components, with minimal power provide attitude determination in an x,y,z earth-centered
requirements, such that they can rely solely upon body- inertial (ECI) frame and interface easily with typical
mounted solar panels or primary batteries for its microsatellite command and data handling subsystems.
mission. In addition to these constraints, most satellites
required components capable of determining attitude It also became evident that there was a growing trend
from a cold-start, providing information to allow the within the U.S. Department of Defense (DoD) toward
spacecraft to perform tasked adjust maneuvers. smaller and more capable space vehicle applications,
which suffered the same lack of suitable small satellite
Sun sensors, earth limb sensors, and horizon crossing components. The lack of low-mass, low-power
indicators are capable of determining attitude from a components was limiting the development of small
cold-start and can achieve combined accuracies of spacecraft. This problem was further compounded by
slightly better than 0.25 degree. However, several the fact that innovations in component technologies
sensors in combination are required to attain this typically need flight heritage prior to their use on such
accuracy in three axes. This leads to increased systems – something that is usually time consuming and
complexity, mass and power consumption. Greater or expensive to achieve.
similar accuracy from a single sensor currently requires
the use of a heavy, power-hungry star tracker that is out Many small satellite missions require attitude sensors
of the reach of most small satellite budgets. that have an accuracy midway between the low-end

Zenick, Ray 1 17th Annual AIAA/USU Conference on Small Satellites

 Price vs. Accuracy Power vs. Accuracy

10000 10000
Typical Medium Sun Sensors
Typical Sun Sensors
1000
1000

Accuracy
(arc-sec)
Proposed Coarse Star Tracker
Accuracy
(arc-sec)

Proposed Coarse Star Tracker 100

100 Caltrac

Caltrac 10
10 Terma HE-5AS
Terma HE-5AS Ball CT-63x Ball CT-63x
1
1
0 5 10 15 20
$- $200K $400K $600K $800K
Price Power Consumption (watts)

Mass vs. Accuracy

10000
Typical Medium Sun Sensors

1000
Accuracy
(arc-sec)

Proposed Coarse Star Tracker

100
Caltrac
10
Terma HE-5AS Ball CT-63x

1
0.0 1.0 2.0 3.0 4.0
Mass (kg)

Figure 1. Comparison of Star Trackers on Price, Power Consumption, and Mass.

devices at 1 to 0.5 degree and technology’s best Table 1. Star Tracker Performance Specifications.
performing star tracker at less than 10 arc-seconds, and
at a cost only slightly greater than the coarse accuracy Specification Performance
systems, somewhere in the $50k to $75K range. The
Field of View ~30 degrees (conical)
coarse star tracker now under development at
AeroAstro and the Massachusetts Institute of Accuracy < 100 arc-seconds
Technology (MIT) Space Systems Laboratory (SSL),
funded by the Missile Defense Agency (MDA) through Roll Rate (Minimum with 0.3 degree/sec
the DoD Small Business Technology Transfer (STTR) Above Accuracy)
program, fills this gap, as shown in Figure 1, and Update Rate 1 Hz
supports MDA and other DoD organizations’ small
satellite programs. Mass 300 grams
Power < 1 Watt @ 5.5 VDC
Currently, AeroAstro and the MIT SSL are continuing
development of the coarse star tracker, moving to refine Dimensions 5.1cm x 7.6cm x 7.6cm
the design through flight prototype hardware
Volume 300 cm3
development, integration, and testing. The performance
specifications for the coarse star tracker are shown in Output x, y, z ECI frame
Table 1, and a simplified functional block diagram of
the star tracker is shown in Figure 2. Limiting Star Magnitude 4th
Star Pairs Tracked Maximum of 4
The mass and volume constraints imposed on the coarse Simultaneously
star tracker are a direct result of the low mass and
volume requirements of small satellite systems Interface RS-422
themselves. In an effort to significantly reduce the
mass and complexity of the coarse star tracker, one of The coarse star tracker being developed is a shift away
the key innovations investigated is the use of a pinhole from the industry trend toward increasing star tracker
lens as an alternative to traditional, heavy lenses. attitude determination accuracy to the exclusion of all

Zenick, Ray 2 17th Annual AIAA/USU Conference on Small Satellites

 accuracy of 70 to 100 arc-seconds, compared to higher-
Pinhole Lens end star trackers with a pointing accuracy of 10 arc-
seconds or less.

Image By relying only on magnitude 4 stars and brighter, the

30° exposure times of the imager can be shortened while
Field of Active Pattern still maintaining an adequate signal-to-noise ratio
View Pixel Recognition Processor between imaged stars and the imager background.
CMOS Software Thus, a lower imager frame rate can be maintained,
Imager further minimizing power consumption, while meeting
the attitude solution rate goal of once per second.

Star Catalog A useful item regarding the accuracy of the star tracker
was found in a paper by Birnbaum.1 It gives a rule of
Roll, Pitch, and thumb for the accuracy improvement in using more
Yaw Position than two stars to perform a least squares fit to obtain an
(x, y, z)
attitude determination. The error of the attitude
Figure 2. Simplified Functional Block Diagram. estimate improves with the number of stars used,

Error of 1 star measurement

other considerations. Current star tracker development Optical axis attitude error = (1)
is focused on rapid pattern recognition, widening fields Number of stars matched
of view to process many stars and star patterns
simultaneously, and expanding internal star catalogs to This is significant, in that even if the star tracker pixel
well over 20,000 stars ranging from 0th to † 6th accuracy is on the order of 100 arc-seconds, the attitude
magnitude. While these advances make for very determination will be on average much better than this
capable systems, the penalty to the satellite is a due to multiple stars being imaged. Thus, the decrease
substantial increase in size, mass, power, complexity, in pixel accuracy from increasing the FOV to 30
and expense. For small satellites, some of which are degrees to guarantee more stars in the FOV would be
designed to simply provide on-orbit demonstration of countered by the benefit of identifying more stars.
space technologies, this is contrary to the requirement
for decreases in these metrics. The coarse star tracker Further background research has uncovered an
is, very simply, designed to provide acceptable additional promising identification method – a gridded,
accuracy at an affordable price, not as competition for pattern recognition approach. In contrast to a star pair
the higher end star tracker market. or star triangle algorithm, the identification is done by
what is essentially an image comparison. Using a
Star Identification Algorithms clever method to simplify the star field into a coarse bit
map, it compares the pattern of stars observed to a
As seen in Figure 2, the operation of the coarse star catalog of stored patterns. The advantages, as detailed
tracker is similar to larger, more expensive star trackers. in Padgett, are a decrease in the star identification time
The algorithms that are employed, however, are much of two orders of magnitude and an increase in the
leaner in terms of computation requirements. The star reliability of the algorithm to inaccuracies and false
pair match group algorithm used in the coarse star stars.2
tracker actively seeks to reduce the quantity of stars
considered in a wide field of view (FOV) image for an The grid algorithm requires enough stars in the FOV to
attitude solution relying primarily on about five stars. produce a unique pattern. There is some question as to
This, in turn, limits the catalog size to the brightest stars the effectiveness of this approach when the number of
in any given region of the sky and considers only the stars in the FOV is limited, as for very short exposures.
brighter stars for the solution. The applicability of this gridded technique is being
evaluated as an alternative for the coarse star tracker.
By stripping away unneeded operations, our approach
vastly reduces the operations the microprocessor unit Roll Rate Limitations
(MPU) has to step through in order to achieve a quick
attitude solution. Although using a wide-FOV The roll rate limitations for the coarse star tracker have
identification method limits the overall accuracy of our been resolved and show that substantial spacecraft roll
design, it improves acquisition speed and reduces can be tolerated, as much as 3 degrees per second for a
power consumption while still achieving a pointing 30 degree FOV star tracker. Furthermore, it appears

Zenick, Ray 3 17th Annual AIAA/USU Conference on Small Satellites

that larger images (3-10 pixels across) help increase the a measure of maximum tolerable roll rate as a function
roll tolerance of the sensor. The pinhole places more of the exposure time (or update rate) and the image size
severe limitations on roll rate tolerance in comparison (diameter), where the image size is the region of the
to the use of lenses. The basic relationship for roll can bright central peak.
be found by considering the image spot from a lens or
pinhole being smeared by a spacecraft roll. Roll rate, r˙ =
image size, i
= Update rate (Hz) • i
time, t (2)
A star tracker experiencing no roll would image stars as
quasi-circular patterns. Some distortion occurs at the To find the update rate, the imager characteristics must
edges of the array, so stars would appear elliptical. † be considered. Many Complementary Metal-Oxide-
Pinhole optics operating in the geometrical regime Semiconductor (CMOS) chips are very similar in size
would produce images of approximately uniform and capability, with notable exceptions being the
intensity, while a lens and focused pinhole would logarithmic response chips, which have continuous
produce an image with a more intense central region. output capability and enhanced range, as opposed to a
As roll increases, the circles become stretched into standard integrating-type chip, which collects a signal
blobs, as shown in Figure 3. Now the left and right over a given time period. Especially if a pinhole optic
portions of the image become less intense, since the is used, the integrating chips are the correct choice for
pixels in those regions are not exposed for the full the low-intensity condition of star imaging.
exposure.
The baseline chip for the coarse star tracker is the IBIS4
If, due to a low signal-to-noie ratio, a star is on the from Fillfactory.3 It is 1280 x 1024 pixels, each 7
verge of being detected as a circle, a pixel would have micrometers square, features on-board analog-to-digital
to collect light for the full exposure to be considered as conversion, quantum efficiency of greater than 50%,
part of a star. Thus, as the roll increases, a limiting and a high fill factor of approximately 60%. Analytical
condition occurs when the image circles at the predictions of the signal-to-noise ratio find that at 1 Hz,
beginning and the end of an exposure are still touching, magnitude 4 stars should be visible in an IBIS4-type
as shown in Figure 3. This limiting condition provides CMOS imager. This is backed up experimentally,

Figure 3. Affect of Roll on Image Produced on CMOS Array.

Zenick, Ray 4 17th Annual AIAA/USU Conference on Small Satellites

 reduction eventually would result in that star becoming
14 so dim that none of the streak’s pixels would have a
signal-to-noise ratio greater than 1. The intensity
12 reduction is given by:
Largest Update rate (Hz) S/N>1

10
I rolling iU
= (3)
8 I no roll r˙

6
Figure 6 shows how a bright star can be resolved at
4
higher roll rates
† as a streak in comparison to dimmer
stars at given update rates. This is important, since the
2 update rate will be somewhat fixed at slower update
rates in order to image dim stars, and bright stars will
0
show up as streaks. In the figure, for a given update
3 3.5 4 4.5 5 5.5 6
rate, the brighter stars can handle higher roll rates.
Limiting Magnitude

1.5
Figure 4. Largest Acceptable Update Rate vs.
Limiting Stellar Magnitude.

Limiting Roll Rate (deg/sec)

0.1
where images show that magnitude 4 stars can be 1 0.5

resolved adequately with a 1 Hz update rate and a 30 1

2
degree FOV on an IBIS4-type CMOS imager, which
3
should allow a reliable star pair algorithm to succeed. 0.5 4
optimal update rate
Using the IBIS4 characteristics and a limiting signal-to-
noise ratio of 1, the fastest possible update rate to image 0
a star of a particular magnitude can be found, as shown 3 4 5 6 7
in Figure 4. For a given image size of 1000 arc- Limiting Magnitude
seconds, corresponding to 10 pixels, the limiting roll
rate can be calculated for a given magnitude star by
Figure 6. Roll Rate Increases from Use of Bright,
using the maximum update rate as given in Figure 5.
Streaked Stars with Fixed Update Rates.
Further, if a bright star is allowed to move so much Already, one can see that the coarse star tracker design
across the imager as to produce a streak, as was shown provides a considerable roll capability. The effect of a
in Figure 3, the intensity of the affected pixels will be lens over a pinhole would be to increase the intensity of
less than would be expected for a full exposure. The light on the pixels, which would greatly increase the
reduction in exposure time means that bright stars fastest allowable update rate. Thus, a lens would
would be imaged as less intense streaks. The intensity tolerate faster roll rates by operating at these faster
update rates.
4
Another interesting possibility is that the star tracker
3.5
could be used as a crude roll sensor. The simple
Roll Rate (deg/sec)

3
approach is to look at two subsequent attitude fixes
2.5 finding the roll rate. An intriguing option would be to
2 intentionally allow the star images to become streaks,
1.5 and measure those streaks in order to find the roll. This
1 technique has been implemented in the Italian FAST
0.5 project.4 This is an interesting possibility, but not
0
within the specifications for the initial star tracker being
3 3.5 4 4.5 5 5.5 6 developed by AeroAstro and the MIT SSL.
Limiting Magnitude
Instrument Definition

Figure 5. Limiting Roll Rate for a Corresponding Work to date suggests that the power goal of less than 1
Stellar Magnitude. Watt is achievable. Even though at this time, a firm

Zenick, Ray 5 17th Annual AIAA/USU Conference on Small Satellites

estimate of the amount of imager/MPU logic that will Table 2. Component Breakdown of Mass & Power.
be required is not known, judging from the interface
presently supplied on the IBIS4 evaluation board and Component Mass Power
that supplied for high-speed interface on most of the (grams) (Watts)
SH7065 commercial off-the-shelf developmental
boards presently available on the market, we feel MPU 40 0.285
confident the quantity or power requirements for the Imager 80 0.430
amount of logic that would be required will not exceed
our goal and the power consumption of the final design Housing 120 0
will remain at approximately 0.7 Watts. Footprint
growth of imager/MPU logic is still well within Pinhole Aperture 35 0
expectation for the coarse star tracker. At the present
Total 295 0.715
time, the only growth not precisely accounted for is
providing for the large pad count both on the MPU and
imager, but this too has been roughly taken into High-level processor requirements were examined. In
account. Thus, based on the imager and MPU order to enumerate the processing requirements for the
developmental boards and the application of five-layer algorithm as it is developed to date, the running time in
printed circuit boards, it is anticipated that the desired the Matlab programming environment and the floating
dimensions can be achieved. point operation count were analyzed for each task in the
star pair algorithm operating an a 330K pixel test image
Table 2 shows the breakdown of mass and power by obtained from CMOS ground observations. The results
component for the baseline star tracker design. are shown below in Figure 7.

Task Breakdown, Total Computation Time = 36 seconds, 3.7 Million Flops

load

noise
Sequential Tasks

sift

centroid

match

group

attitude

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70

Percentage of Total Time
attitude group match centroid sift noise load
flops (% of 3.7 Million 0.00 0.00 0.63 0.00 0.00 0.37 0.00
Flops)
time (% of 36 sec) 0.03 0.03 0.58 0.00 0.08 0.22 0.06

Figure 7. Computation Requirements of each Task in Time and Flops.

Zenick, Ray 6 17th Annual AIAA/USU Conference on Small Satellites

One can see that the noise correction and matching recognition methods explored to date. The gridded
routines account for the bulk of the processing effort. algorithm offers a significantly improved probability of
The total processing effort for 330K pixel image acquisition and significantly reduces the active memory
amounted to 3.7 Million floating point operations. This required of the star catalog. However, this gain comes
is well below the baseline Hitachi processor capability at a cost in algorithm complexity.2 This option has not
of 78 MIPs. Baseline images will be somewhat larger, been fully explored, and during the next phase of work,
on the order of 1 million pixels, but this will not greatly our plan is to implement and model the grid algorithm
affect the match time, although it will affect the noise to determine if this method is preferable to the pattern
correction. In addition, the above simulation was recognition approach.
carried out with a larger catalog (1140 stars) than is
envisioned for the final product (~600). Still, the noise Conclusions and Future Objectives
correction operation is linear with pixel count, and the
total computations should be well within the processor AeroAstro and the MIT SSL are developing a coarse
capabilities. The code currently executes without star tracker that serves a presently unserved attitude
regard for fixed point vs. floating point operations. The determination need of small spacecraft. This star
computation should be easily adapted to complete fixed tracker will have a mass of less than 300 grams,
point operation, allowing for faster operation. dimensions less than 5.1cm x 7.6cm x 7.6cm and
consume 1 Watt or less of power.
Several schemes for the further reduction of the star
tracker’s overall power consumption have been Our work to date has brought the coarse star tracker
considered. These mainly consist of reducing or from concept to a preliminary design. The design has
economizing on the quantity of MPU operations and the been modeled, and its operation and dynamic ability
selection of the MPU itself. Some of these schemes have been verified. The team is moving toward
could contribute further to the reduction in the star implementing the star tracker design in a testbed star
tracker’s power consumption. Among those that will configuration. This will allow further refinement of
receive further consideration in future work include: star tracker hardware design to minimize the circuitry
and power consumption in the imager and MPU. The
ÿ More efficient grid algorithm – This has the testbed will also serve to optimize the star tracking
potential to greatly reduce the processor power algorithms and verify its performance through
draw by a factor of 10. simulated and sky conditions. It is anticipated that
fabrication of a proto-flight model of the coarse star
ÿ Keeping the star catalog as small as possible – tracker will be complete in mid-2004.
This is important because of the non-linear scaling
for match group algorithms processing Acknowledgements
requirements with catalog size. The grid algorithm
scales linearly, however, with catalog size. This material is based upon work supported by the
ÿ Reduced update rate – By reducing the update Dahlgren Division, Naval Surface Warfare Center,
under Contract No. N00178-02-C-3125.
rate from 1 Hz to 0.5 Hz, the average power draw
on the processor could be reduced by a factor of 2.
The authors wish to thank the Missile Defense Agency
However, this may adversely affects the ability of
the star tracker to achieve its mission requirements, and the DoD SBIR/STTR Program for their support of
the Phase I contract under which this work was
so this trade needs to be investigated further.
Another option is to have an update rate tailored to conducted. We especially thank Dr. Erwin Myrick and
Mr. Marc Wigdor, both of the MDA, for their
the available power.
consideration and support.
ÿ Implement a tracking mode – A tracking mode
would help to alleviate the processor load from the References
matching routine, but would increase the software
complexity and logic. 1. Birnbaum, M.M., “Spacecraft Attitude Control
Using Star Field Trackers,” Acta Astronautica,
Most of the required power savings identified to date vol. 39, No. 9-12, 1996.
have been in the form of lower duty cycles on the
imager and processor circuitry. One other solution that 2. Padgett, C., K. Kreutz-Delgado, and S.
may both reduce MPU operation and contribute Udomkesmalee, “Evaluation of Star Identification
significantly to power consumption reduction is the Techniques,” Journal of Guidance, Control &
application of a grid algorithm rather than the pattern Dynamics, vol. 20, No. 2, March – April 1997.

Zenick, Ray 7 17th Annual AIAA/USU Conference on Small Satellites

3. IBIS4 SXGA Image Sensor Datasheet, Fillfactory, Applications.” Proceedings of the 15th Annual
http://www.fillfactory.com, December 2000. AIAA/USU Conference on Small Satellites,
Logan, Utah, August 2001.
4. Ruocchio, C., D. Accardo, G. Rufino, S. Mattei, and
A. Moccia, “Development and Testing of a Fully 6. Liebe, C. C., “Accuracy Performance of Star
Autonomous Star Tracker.” Proceedings of the Trackers – A Tutorial,” IEEE Transactions on
2nd Symposium on Small Satellites for Earth Aerospace and Electronic Systems, vol. 38, No. 2,
Observation, Berlin, Germany, April 1999. April 2002.

5. Shuker, B., “A Ground-Based Prototype of a CMOS

Navigational Star Camera for Small Satellite

Zenick, Ray 8 17th Annual AIAA/USU Conference on Small Satellites