Energy Budgeting for CubeSats with an Integrated FPGA

Scott Sterling Arnold, Ryan Nuzzaci, and Ann Gordon-Ross

Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL 32611
Also with the NSF Center for High Performance Reconfigurable Computing (CHREC) at the University of Florida
http://www.ann.ece.ufl.edu; [email protected], [email protected], [email protected]

4.   Power Consumption Evaluation and Case

Abstract—CubeSats are a simple, low-cost option for Studies ................................................................. 5  
developing quickly-deployable satellites, however, the tradeoff
for these benefits is a small physical size, which restricts the
4-1.   Experimental Setup ............................... 5  
CubeSat’s solar panels’ size and thus the available power 4-2.   Power Results......................................... 6  
budget and stored energy reserves. These power/energy 4-3.   Case Studies ........................................... 6  
limitations restrict the CubeSat’s functionality and data
processing capabilities, which makes leveraging CubeSats for
5.   Energy Reserve Budget ............................... 8  
compute-intensive missions challenging. Additionally, 5-1.   Energy Reserve Budget Development . 8  
increasing sensor capabilities due to technological advances 5-2.   Single-Payload 1U Case Study ............. 9  
further compounds this functionality limitation, enabling
sensors to gather significantly more data than a satellite’s
5-3.   Triple-Payload 3U Case Study ........... 10  
limited downlink bandwidth can accommodate. The influx in 6.   Conclusions ................................................ 13  
sensed data, which is particularly high for image-processing 7.   Acknowledgments ...................................... 13  
applications, introduces a pressing need for high-performance
on-board data processing, which preprocesses and/or
8.   References .................................................. 13  
compresses the data before transmission. FPGAs have been 9.   Biography ................................................... 14  
incorporated into state-of-the-art satellites to provide high-
performance on-board data processing, while simultaneously
reducing the satellites’ data processing energy consumption. 1. INTRODUCTION
However, even though FPGAs can provide these capabilities in
full-scale satellites, a CubeSat’s limited power budget makes CubeSats—or cube satellites—are small satellites
integration of FPGAs into CubeSats a challenging task. For categorized by a standard size and weight [21] and are
example, the commonly used Virtex4QV Radiation Tolerant divided into three different unit sizes based on physical
FPGA family’s average power consumption ranges from 1.25 volume: the 1U (one unit) CubeSat is 10x10x10 cm, the 2U
to 12.5 Watts, whereas the CubeSat’s power budget ranges (two unit) CubeSat is 10x10x20 cm, and the 3U (three unit)
from 2 to 8 Watts, with the smallest, cheapest CubeSat systems CubeSat is 10x10x30 cm. This size restricts the CubeSat’s
at the lower end of this range. Therefore, in order to
upper weight limit at 1 kg per unit-size (e.g., 3 kg for a 3U
successfully integrate FPGAs into CubeSats, the components’
power consumptions must be clearly budgeted with respect to CubeSat). These size and weight restrictions limit the
the CubeSat’s specific functionalities and orbital pattern, available payload for batteries and external solar panels,
which dictates the available power and stored energy reserves. which limits the CubeSat’s power consumption and
In this paper, we present two detailed energy reserve therefore imposes a strict power budget. A CubeSat’s power
budgeting case studies for FPGA-based CubeSats with respect budget defines the maximum combined power consumption
to stored energy reserves for image compression and of all of the CubeSat’s constituting payload components
processing using a Canny edge detector. CubeSat designers can based on the total power production of the CubeSat’s power
leverage this energy reserve budget with the application- producing components (e.g., solar panels). For a CubeSat’s
specific components’ power consumptions for applications
mission to be successful, designers must consider and
such as hyper-spectral imaging (HSI), ground motion target
indication (GMTI), and star tracking to quickly determine adhere to the power budget during the CubeSat’s design and
maximum payload operational time with respect to specific construction process.
orbital patterns and mission requirements.
To assist designers in CubeSat mission design, CubeSat
TABLE OF CONTENTS construction guidelines [20] impose SWAP (size, weight,
and power), shape, and other restrictions that, when adhered
1.   Introduction ................................................. 1   to, offer several benefits over larger satellites in terms of
2.   Background And Related Work ................ 3   cost, development time, and payload modularity. These
2-1.   CubeSat Missions .................................. 3   benefits make CubeSats more amenable to academia, small
companies, and countries without fully-funded space
2-2.   Space-based FPGA Image Processing . 4   programs. A CubeSat’s small size reduces launch costs [10]
3.   Orbital Patterns ........................................... 4   and a standardized shape reduces the development cost and
time. Since a standardized shape imposes inherent design
restrictions, which restrict complexity and reduce design
options, CubeSats have short development times (the time

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 1

 Table 1: CP1 power budget estimation [27].
from mission inception to CubeSat construction), typically
ranging from six to twelve months [21]. Additionally, Subsystem Power Usage
CubeSats leverage commercial-off-the-shelf (COTS) ADCS (sun-sensor, no control) 0
components, which are low cost and decrease development
CMOS camera payload 70 mW
time by eliminating the need for lengthy custom component
C&DH (ARM7 core) 110 mW
design and fabrication. However, the inherent risks of using
COTS components offset these low-cost benefits are Color camera payload 80 mW
somewhat offset by the, such as high susceptibility to upset- Communication 267 mw
inducting radiation (e.g., single event upsets (SEUs) [5]) and Power Profile
a shorter lifespan (usually orbit-dependent and on the order Total power consumed 527 mW
of 6 to 18 months). Solar panel-generated power 726 mW
Power differential +190 mW
Since mission design often begins before the payloads’
components are determined and CubeSat construction with large data sets. Without data processing, the
begins, payload modularity enables mission design to communication subsystem must provide sufficient resources
commence relatively independent from CubeSat to transmit the complete, unprocessed data. Data processing,
construction, allowing payloads to be easily integrated later using components such as microprocessors, reduces the data
in the construction process. However, payload modularity set size, which reduces the total amount of data transmitted
introduces power budgeting challenges since the payload’s and reduces communication requirements. Satellites
constituting components may not be determined until late in typically employ microprocessors for this data processing,
the mission design process, which may restrict the payload’s but microprocessors may not provide sufficient processing
power and data processing capabilities. speed for highly compute-intensive missions, such as
image-processing missions that pre-
A CubeSat’s power budget is composed of several key process/compress/coalesce multiple, large data images
subsystems, these subsystem’s constituting components, and before transmitting the complete image.
the payload power. The CubeSat’s subsystems include
attitude determination and control (ADCS or ACS), Large satellites often leverage specialized processors, such
electrical power and supply (EPS), communication, and as field-programmable gate arrays (FPGAs), for compute-
control and data handling (C&DH). The ADCS can vary intensive applications such as hyperspectral imaging (HSI)
from passive magnetic control, which requires little to no [15] and synthetic aperture radar (SAR) [14]. For HSI, a
power, to high power methods such as gyroscopic control multispectral camera delivers image data cubes ranging in
based on horizon sensors and mechanical actuators using size from hundreds of Megabytes to Gigabytes [15]. Jacobs
star-tracking. The ADCS choice depends on the mission’s et al. [15] show that a Virtex-4 SX35 FPGA achieves a 15x
functionality requirements, and the ADCS influences the speedup as compared to a PowerPC 7455 for an HSI data
payload options based on these requirements (e.g., camera processing algorithm. For SAR, Jacobs et al. [14] show that
directionality, antennae position, etc.). a Virtex-4 SX55 FPGA achieves a 6x speedup for azimuthal
data processing as compared to a G4 PowerPC.
The EPS directly dictates the solar panels’ generated and
stored power, which defines the CubeSat’s maximum power Even though FPGAs provide high performance data
budget. Complex, foldout solar panels and higher quality processing, typical FPGA power consumption ranges from 5
batteries can increase the maximum power budget, to 10 Watts for Virtex and Stratix FPGAs [2], which may
However, physical CubeSat size restricts the area available exceed a CubeSat’s power budget. Table 1 depicts the CP1
for solar panels. Typically, the 1U, 2U, and 3U CubeSats’ CubeSat’s—Cal Poly’s first developed CubeSat—power
maximum power budgets range from 1 to 2.5 Watts, 2 to 5 budget estimation for each subsystem and the power profile,
Watts, and 7 to 20 Watts, respectively [10]. which includes the total power consumed by the
subsystems, the power generated by the solar panels, and the
The communication subsystem, one of the largest power power differential, which is the difference between the
consuming subsystems, is critical for establishing and power consumed and the power generated. The CP1 had no
maintaining both uplink and downlink speeds and attitude control, contained two payloads (a sun-sensor and a
communication reliability. The communication bandwidth color picture camera), and had a low-power communication
and reliability levels are directly proportional to the subsystem [27]. The positive power differential shows that
communication subsystem’s allocated power. However, due while in sunlight, the solar panels captured more power than
to limited power, CubeSats typically have low the CP1 required, thus the excess power could be stored in
communication bandwidth and reliability. the batteries for use during eclipse time. Even though
demand for high-performance embedded computing has
These communication limitations make the C&DH, which is
realized lower-power FPGAs that are more amenable to
responsible for on-board data processing and handling,
CubeSats, such as the Xilinx Spartan [8], Altera Cyclone
critical for supporting compute-intensive mission
[29], and Xilinx Artix families [19] with power
functionality, such as image processing and/or missions
consumptions ranging from 1 to 2.5 Watts, these FPGAs

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 2

may still require a large percentage of a CubeSat’s power Canny edge detector (Canny filter) to approximate how
budget. much power the FPGA requires for a multistage image-
processing filter. Based on the energy reserve budgets from
In order to aid designers in leveraging low-power FPGAs in these case studies, the components’ power consumptions,
CubeSat missions, we propose an energy reserve budget, and the specific orbital pattern and elevation,, results
which equates the total energy produced by the EPS to the analysis shows that low-power FPGAs can be incorporated
total energy consumed by the CubeSat during a single orbit into a CubeSat’s C&DH, and the energy reserve budget
around the Earth. The energy reserve budget leverages assists designers in effectively leveraging power resources
multiple power budgets, which are associated with different to maximize the payload processing capabilities .
modes of operation, referred to as power modes. All
missions should have at least two basic power modes: a 2. BACKGROUND AND RELATED WORK
power-storing mode and an overpower mode. In the power-
storing mode, the total power consumption is less than the In this section, we summarize the goals and motivations
power produced by the solar panels so that the batteries can behind CubeSat missions and discuss several past CubeSat
store reserved energy. In the overpower mode, total power missions. We discuss how CubeSats have become
consumption can exceed the solar panels’ generated power increasingly popular for demonstrating novel technologies
by leveraging the batteries’ stored energy. in space and how CubeSats have provided low-cost
opportunities for academia, small businesses, and
Systems with large payload power consumption, such as governments to perform space-based experiments. Since
those including FPGAs, may have several variations of CubeSats are a relatively new platform—the first proposed
these basic power modes. For example, a system with both CubeSat was in 1999—we outline notable failures and
high power data processing and high communication lessons learned from these failures. Lastly, we discuss
requirements may have three power modes: a power-storing processing speed up and energy savings potential afforded
mode, a processing-overpower mode, and a communication- by FPGAs for image-processing space systems.
overpower mode. More complex CubeSats with multiple
payload operations may require a specific overpower mode 2-1. CubeSat Missions
for each operation. Additionally, due to orbital variations, The California Polytechnic (Cal Poly) University
multiple power modes may be required to compensate for Multidisciplinary Space Technology Laboratory and
varying power availability. Stanford’s Space Systems Development Laboratory initially
conceptualized CubeSats as an effective way to provide
Since the orbital pattern dictates the amount of sunlight a students with hands-on experience with developing,
CubeSat receives and the elevation dictates the amount of launching, and operating a spacecraft [10]. The CubeSat
time, if any, a CubeSat spends in the Earth’s eclipse, these standard was developed to provide basic physical features
factors primarily determine the amount of available power and safety requirements for CubeSat designers. The
and the amount of storable energy. Even though the exact standard specified size, weight, and shape requirements
orbital pattern and elevation may not be known at design (Section 1) and imposed additional restrictions such as no
time (CubeSats are often secondary payloads on to-be- pyrotechnics/explosives and no interference with other
determined launch vehicles), the designer can anticipate the CubeSats being deployed from the same launch vehicle.
orbital pattern to create an approximate energy reserve
budget early in the design process, which enables the Cal Poly also designed a standardized deployment capsule
designer to determine if a launch opportunity is a valid called the P-POD [25], which could hold any combination
candidate based on the mission’s required functionality. of 1U, 2U, or 3U CubeSats totaling up to three units in size
Additionally, the energy reserve budget allows designers to (i.e., three 1Us, one 2U and 1U, or one 3U). P-PODs were
more effectively leverage CubeSat resources with respect to designed to lower the inherent risk of CubeSats by
predicted power availability, thereby maximizing overpower providing protection from debris during launch, ensuring
mode operation and increasing CubeSat efficacy. that early activation of CubeSats did not occur, and ensuring
that antennas and booms were not prematurely deployed.
In this paper, we propose energy reserve budgets for 1U and Alternative deployment capsules, such as the X-POD and T-
3U case studies based on common orbital patterns and POD [18], perform the same function as the P-POD but are
elevations to determine the maximum percentage of orbital less popular due to the X- and T-PODs’ lower carrying
time available for high-power FPGA processing (i.e., time capacities and because Cal Poly, who organized the majority
spent in a processing-overpower mode). This effective of early CubeSat launches, required the use of P-PODs. As a
processing time allows designers to most effectively choose result, 90% of all CubeSats are launched from P-PODs [18].
the maximum processing resources with respect to the
processing resources’ power consumption. The energy CubeSat construction guidelines, though strict with respect
reserve budget also determines what types of compute- to SWAP and shape, are amenable to a wide range of
intensive functionalities are possible on CubeSats of all unit missions including: sensor deployment to retrieve telemetric
sizes. Using Xilinx ISE and ISIM tools, we calculate the data such as the QuakeSat that used a magnetometer to
power consumption for several low-power FPGAs using a measure the Earth’s magnetic field [17]; testing COTS

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 3

components in a space-based environment such as the scientific applications such as experimentation and Earth
CUTE 1.7 APD II that used a personal data assistant (PDA) science and imaging [7]. This altitude range defines LEO,
as the primary C&DH subsystem [11]; and advancing rather than the specific orbital pattern, which defines the
CubeSat technology through experimentation such as the particular path a satellite follows during orbit. Above LEO
ION that was equipped with an electrical propulsion is medium Earth orbit (MEO), which extends outward to
subsystem [13]. 35,786 Km, and is followed by high Earth orbit (HEO) for
altitudes greater than 35,786 Km.
Due to the CubeSat’s small size, light weight, and COTS
components, CubeSats are often a risky platform, with only Orbital patterns are defined by the orbit’s inclination and
two-fifths of CubeSat launches successfully meet the eccentricity. The inclination establishes the angle at which a
mission’s goals [10]. CubeSat failures can occur due to satellite orbits the Earth with respect to the equator, is
launch vehicle failures, such as the DNEPR rocket dictated by the CubeSat’s deployment from the launch
explosion that destroyed fourteen CubeSats [9], and vehicle, and dictates the satellite’s Earth geographical
subsystem malfunctioning, such as the AAU CubeSat that position at a given point in time. Figure 1 shows the orbit of
experienced a short-circuited radio shortly after launch that a satellite at a 75o inclination, which is why the ground-
drained the entire battery [1]. However, since the majority tracking view shows the satellite taking an apparent north
of CubeSat failures cause the CubeSat to never establish and south motion as it traverses around the globe. Satellite
radio contact, the failure reason cannot be diagnosed. inclination is always quantified between 0o and 180o, with a
0o heading indicating eastward movement in orbit and a
2-2. Space-based FPGA Image Processing 180o heading indicating westward movement.
The FPGA’s ability to perform computations and operations
The eccentricity establishes the deviation an orbit takes
in circuit-emulating hardware as compared to processors
from a perfect circle around the Earth based on a two-
performing computations in software is essential to
dimensional plane. Orbit eccentricity divides orbits into two
understanding the speedups that are afforded from executing
categories: closed and open. Closed orbits, including
parallelizable applications on FPGAs. Even though FPGAs
circular and elliptical orbits, periodically repeat the same
typically operate at lower frequencies than processors,
shaped path around the Earth. Closed orbits steadily decay
FPGAs can attain performance gains that are magnitudes
and eventually end when the satellite loses velocity and
faster.
plummets to Earth or when the satellite is moved into the
Since previous work has shown that image-processing Junk orbit (an orbit considered not operationally useful due
applications are amenable to FPGA speedup [14][15], to the orbit’s high altitude). Open orbits, including
FPGAs are suitable for image-processing space missions, hyperbolic and parabolic orbits, eventually escape the
which typically capture large and/or many images and Earth’s gravitational pull by achieving enough energy to
slow/limited downlink bandwidth from satellites to ground reach escape velocity after a single orbit. Since satellites in
stations makes transmitting such large data sets difficult, if open orbits only orbit the Earth once, our work focuses on
not impossible. FPGAs have been employed in image- closed orbits.
processing missions to compress/coalesce image data before
A sun-synchronous orbit is typically a circular, retro-grade
transmission to reduce downlink requirements. Specialized
orbit at a 98o inclination and an altitude between 600 and
applications with data-filtering algorithms can benefit from
800 km. Sun-synchronous orbits are popular orbits for
significant speedups that, with regards to overall power
CubeSat’s for two reasons: 1) sun-synchronous orbits
consumption, can reduce the total energy consumption by
typically experience near-constant sun illumination and thus
using FPGAs as compared to processors. For example,
provide high power; and 2) the CubeSat passes over the
Kovac et al. [16] used a fully-pipelined VLSI (very large
same Earth geographical position at the same time every
scale integration) architecture to perform JPEG compression
day. Figure 2 depicts a sun-synchronous orbit that never
at a rate of 100 million pixels per second, which is fast
enters the Earth’s eclipse by keeping the orbital plane close
enough to compress thirty 1024x1024 pixel frames in one
to orthogonal to the sun’s direction.
second. Jacobs et al. [15] showed that an FPGA
implementation of SAR afforded a 6x speedup and a 50%
reduction in energy consumption as compared to a software
implementation.

3. ORBITAL PATTERNS
Since a CubeSat’s orbital pattern dictates the amount of
sunlight a CubeSat experiences, it is critical to consider how
the orbit influences the amount of available power during
mission design. CubeSats are typically placed in low Earth
orbit (LEO), which ranges from 160 Km to 2000 Km in
altitude [23] and offers a variety of mission options for Figure 1: Ground-tracking of a circular orbit at a 75o
inclination

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 4

For all circular orbits other than sun-synchronous, the
following equation determines the amount of time the
satellite spends in the Earth’s eclipse [23]:

r -ll( ~\I'l t
t
s l
;<COS
\. ---
COS
-
f3 ) j lr
(1)
where,
where,

R ; radius ratio ; l~q !~at

l~q ; equatorial radius of the Earth

rra! ; geocentric radius of the satellite; l~q + hsa!
h rar ; altitude of the satellite
f3 ; Sun-orbit-plane angle

r ; orbital period of the satellite; 2/Z' ~1~~r 0' Ii

Ii ; gravitational constant of the Earth

Figure 2: Sun-synchronous orbit showing constant sun

The Sun-orbit-plane angle
The angle fJ-or
β—or Beta angle-is
angle—is thethe angle
angle illumination
between the the geocentric unit
unit position vector to the
the sun and
the satellite's
the satellite’s orbit
orbit plane. The worst-case fJβ is 0°,
plane. The 0o, which Xilinx Spartan-3 devices and a Virtex-4 device. The device
that the
indicates that the satellite is spending the
the maximum amount usage allows power consumption estimators provided by
time in the
of time the Earth's
Earth’s eclipse
eclipse as possible for a given
given altitude
altitude Altera [3] and Xilinx [30] to extrapolate our FPGA power
the best-case fJβ is ±± 90°,
and the 90o, which
which indicates that
that the
the orbit
orbit consumption to inspect devices that were not physically
never enters
never enters eclipse:
eclipse: simulated. We also present two case studies based on
existing 1U and 3U CubeSat designs that will be used to
evaluate of our energy reserve budgets in Section 5.
(2)
4-1. Experimental Setup
where rrsun is the
where the geocentric unit
unit position vector of the
the Sun
We evaluated the power consumption for five low-power
and hsat is the
the unit
unit angular momentum vector of the
the satellite:
satellite:
Xilinx Spartan-3 devices (XC3S400A, XC3S400AN,
XC3S700A, XC3S700AN, and XC3S400), which represent
rcos~un c~sarunl sin Osin r "I the smallest Spartan-3 family with enough resources for the
r.wn ; 1cos ": '"
slnb~"J
«: h,at = -cosn~ini
COSt
Canny filter (for power consumption reasons, we assume a
designer would select the smallest device based on the
., .. f (3)
(3)
(4) mission resource requirements) and these devices allow us
where αsun is the geocentric, equatorial right ascension of the to evaluate device usage trends within the Spartan-3 family.
Sun, δsun is the geocentric, equatorial right declination of the We also evaluated one Virtex-4 device (XC4VFX12) for
comparison with a non-low-power FPGA. We created a
Sun, Ω is the right ascension of the ascending node
post-place and route simulation model in ISE [31] and a vcd
(satallite) (RAAN), and i is the satellite’s inclination [23].
file in ISIM [31]. The vcd file recorded the image filtering
αsun and δsun are characteristic of the Sun’s relative position process at the bit-flip level and served as input into Xpower
to the Earth and are a function of the time of year and the (Xilinx ISE Webpack [31]) to calculate the dynamic and
inclination and RAAN are characteristics of the satellite static power for each of the devices, as well as the power
[23]. The satellite’s orbital period can be determined using a consumed by the FPGA fabric resources (logic, BRAM,
gravitational constant µ = 398,600.4418 km3/s2 and the etc.). From these power consumption results, we obtained
satellite’s altitude. the device usage details and power consumption of the
FPGA while executing a Canny filter, which comprises
4. POWER CONSUMPTION EVALUATION AND several image filter algorithms in series to process an image
CASE STUDIES before edge detection takes place.

In this section, we describe our methodology for evaluating Our experiments evaluate an in-house VHDL Canny filter,
device usage (toggle rate, block RAM (BRAM) writes and which is a standard image-processing filter for locating and
reads, etc.) and component power consumption using a highlighting edges in an image. Image-processing filters
sample multistage image filter (Canny filter) for several traverse images pixel-by-pixel and perform several

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 5

 (a) (b)
Figure 3: a) The original Wall Street bull image used for Canny filter processing. b) The output image after Canny
filter processing.

operations on the pixels to create a modified version of the camera. The Canny filter processed the serialized image
image. Filtering applications are highly parallelizable, data and stored the output into a FIFO (first-in-first-out)
which makes these applications highly amenable to FPGA buffer, which was then compressed by the FPGA and
processing speedups as compared to microprocessors. transformed to a viewable JPEG image.
Figure 3 (b) depicts an example of a Canny filter’s output
after processing an image of the Wall Street bull in Figure 3 4-2. Power Results
(a). We refer the reader to [12] for additional Canny filter Table 2 depicts the dynamic, static, and total power for our
details. evaluated devices while performing the Canny filter
operation and Table 3 depicts the associated resource
A Canny filter’s front-end processing is a Gaussian filter
requirements (look-up tables (LUTs) shift registers (shift
and a Sobel filter in sequence, which obtains the first
regs), flip-flops (FFs), clock fanout, input/output pins (I/O),
derivatives of the image to remove high-frequency
BRAMs, distributed clock manager (DCM), digital signal
components and determines the image’s intensity gradient.
processors (DSPs), and multiplication units (MULT)). Since
Based on results obtained from Xpower averaged over all
the results revealed that the resource requirements were
Spartan-3 devices, the Canny filter’s static power
device-independent for the Spartan-3 devices, Table 3 lists
consumption ranges between 40 and 55 mW, the dynamic
only a single column for the Spartan-3 devices. The results
power is approximately twice that of the static power, and
revealed only a 4.6% increase in the logic resource
the average resource utilization is approximately 2,500 logic
requirements (LUTs, shift registers, FFs) when comparing
slices and 18 BRAM slices.
the Spartan-3 devices to the Virtex-4 device.
The 1024x768 JPEG of the Wall Street bull was serialized
4-3. Case Studies
using a python script before being input to the FPGA. We
note that this serialization could have been done on the Due to the 1U CubeSat’s highly constrained size and power
FPGA but is not necessary for our case studies since we are budget, the energy reserve budget is most beneficial and
effectively simulating an incoming data stream from a
Table 3: Per-family FPGA resource requirements
Table 2: FPGA power consumption for the Canny filter
Resources Spartan-3 Virtex-4
Power Consumption (mW) Logic (LUTs) 1528 1634
Device Dynamic Static Total Logic (Shift regs) 46 44
Spartan-3 Logic (FFs) 1055 1071
XC3S400A 97.52 40.11 137.63 Clock fanout 1119 1156
XC3S700A 103.34 46.40 149.74 I/O (input pins) 11 11
XC3S400AN 98.66 48.62 147.28 I/O (output pins) 12 12
XC3S700AN 110.35 55.68 166.03 BRAM 18 18
XC3S400 104.19 95.29 199.48 DCM Single Single
Virtex-4 DSP 0 0
XC4VFX12 161.00 199.10 360.10 MULT 0 0

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 6

 Table 4: Power budget for the M-Cubed CubeSat

Subsystem Description Minimum Power (mW) Maximum Power (mW)

ADCS Passive Magnetic ACS 0 0
C&DH Atmel AVR32 7002 -216 -292.5
Uplink 144 MHz, .33-m dipole -616.5 -616.5
Downlink 437 MHz, .5-m monopole 0 -1000
Primary Payload CMOS camera 0 -250
Secondary Payload COVE board 0 -5088.5
Board I/O power and loss -227.5 -1254
EPS 4 Polymer Li-ion Batteries -960 +4091
EPS 6 Solar Panels +2010 +2010

challenging for these CubeSats. Furthermore, since 1U energy reserve budget, which consists of these power
CubeSats cost less than larger CubeSats, previous work modes, can be used to calculate the orbital-pattern-
shows that 1U CubeSats are the most popular to construct. dependent amount of time a CubeSat can spend in each
Therefore, our first case study focuses on the 1U CubeSat, power mode.
which is based on the M-Cubed CubeSat [4].
Since energy reserve budgeting is useful for any platform
Table 4 shows the minimum and maximum power size, we also evaluate an energy reserve budget case study
consumptions in mW for the M-Cubed’s subsystems [4], for a larger 3U CubeSat design based on the QuakeSat,
where positive (+) numbers indicate that a subsystem is which was the first 3U launched into space in 2003[17].
producing power and negative (-) numbers indicate that a Even though the QuakeSat was short-lived due to a quickly
subsystem is consuming power. The M-Cubed’s secondary degenerating orbit, the QuakeSat’s design could have easily
payload contained an on-board verification experiment sustained a much longer mission and is therefore a good
(COVE) and required a high-power Virtex-5QV Single- evaluation choice for a sample 3U CubeSat.
event Immune Reconfigurable FPGA [24]. Since this
payload was the highest power subsystem, the M-Cubed Table 5 shows the minimum and maximum power
designers carefully considered the overall system’s power consumptions in mW for the QuakeSat’s subsystems, where
consumption during mission design. In order to provide positive (+) numbers indicate that a subsystem is producing
enough power for the secondary payload, the system power and negative (-) numbers indicate that a subsystem is
included four Polymer Li-ion batteries, which each stored consuming power. Since the QuakeSat’s mission dictated
3.89 Wžh. These batteries were chosen by the M-Cubed that the payload would be active for a significant portion of
designers to provide a larger discharge rate as compared to the orbital time to collect data, the QuakeSat was built using
other higher capacity batteries such that there was enough a very conservative power budget [6]. Due to the high
power to run the secondary payload [4]. power production provided by the twelve solar panels (14
Watts), the QuakeSat easily supported the constant power
The 1U case study contains a single payload and has four draw of 3.6 Watts for the CPU (C&DH), uplink, and board
power modes: 1) a power-storing mode; 2) a I/O for continual operation [17]. Prior to launch, the
communication-overpower mode for data transmission to a designers estimated that the maximum power consumption
ground station; 3) a communication-uplink-overpower mode of 12.6 Watts (when all components were operating) would
for receiving ground station transmissions; and 4) a not exceed the average power production of the solar panels
processing-overpower mode for high-performance FPGA when in sunlight.
payload processing. In Section 5, we demonstrate how the

Table 5: Power budget for the QuakeSat CubeSat

Subsystem Description Minimum Power (mW) Maximum Power (mW)

ADCS Passive Magnetic ACS 0 0
C&DH Prometheus CPU -2500 -2500
Uplink Tek-net 9600 Baud-Rate Solid-State -750 -750
Downlink transmitter -1000 -1400
Primary Payload Magnetometer -600 -2200
Board I/O power and loss -850 -850
EPS 2 Li-ion Batteries -8800 +11700
EPS 12 Solar Panels +14000 +14000

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 7

The 3U case study focuses on CubeSat resource though inexact in the early design stages, the energy reserve
optimization, the addition of fault-tolerant networking budget is beneficial at the beginning of the design process
methodologies, and efficient usage of the available power. by enabling designers to better quantify the maximum
Whereas the 1U case study focuses on working within tight payload power available as a function of both orbital time
power constraints the 3U case study has more power and the usable energy produced during a single orbit. By
resources and can thus leverage multiple FPGAs in a COTS analyzing the orbital pattern and the energy reserve budget,
distributed C&DH network. Due to the larger number of the designer can calculate the payload(s) operational time(s)
FPGAs and thus increased cumulative payload processing and compare these times to the mission requirements. If the
time as compared to the 1U case study, we incorporate fault- mission requirements are not met, the designer may redesign
tolerant methodologies to ensure high system availability the subsystems to achieve lower power consumption or
since the COTS FPGAs are not radiation hardened. Using create another iteration of the energy reserve budget to
multiple FPGAs allows implementation of a dependable improve the payload(s) operational time(s). One potential
multiprocessor (DM) fault-tolerance methodology, which side effect of using an energy reserve budget is that early
ensures that FPGA failure does not cause entire system design stages may require additional time due to the
failure. This fault-tolerant operation leverages a central additional analysis time required to create the energy
processor, which monitors the payload FPGAs to detect reserve budget. However, this additional time is recovered
FPGA failures [26]. Detailed DM operation is beyond the by quickening component selection and CubeSat
scope of this paper and we refer the interested reader to [26] construction.
for additional information.
We first establish the process of determining an energy
The 3U case study contains three payloads and the energy reserve budget for a system with two power modes
reserve budget requires two iterations to exemplify how the (processing-overpower and power-storing), and then
energy reserve budget can indicate an overly conservative generalize the process for systems with an arbitrary number
CubeSat design with respect to power consumption. The of power modes. The first step in developing the energy
first iteration pinpoints subsystems where increasing the reserve budget is to quantify the payload’s power
operational time would be advantageous to meeting the consumption for the overpower mode Ppayload (the payload is
mission’s goals. The first iteration leverages five power fully operational) and the power-storing mode Pstore (the
modes: 1) a power-storing mode; 2) a communication- payload is completely powered off). These modes’ power
overpower mode for data transmission to a ground station; consumptions are combined with an assumed orbit to
3) a processing-overpower-P1 mode for using a single determine the CubeSat’s eclipse time ts. During the eclipse
FPGA for payload processing; 4) a processing-overpower- time, the solar panels do not generate any power, which
P2 mode for using two FPGAs for payload processing; and decreases the overall energy stored in a single orbit. The
5) a processing-overpower-Pall mode for using all on-board following equation calculates the total energy Jproduced
FPGAs for payload processing. The second iteration refines available during a single orbit:
the first iteration’s output to further optimize the power
usage. (5)
(5)
(5)
where Ppanels is the average power produced by the solar
5. ENERGY RESERVE BUDGET panels (this data can be obtained from the solar panels’
In this section, we use the power results from our Canny datasheets or via physical testing) while in sunlight and τ is
filter (Section 4-2) and the 1U and 3U CubeSat designs the orbital period. Using Jproduced, the following equation
(Section 4-3) to create three energy reserve budgets: two calculates the operational time tp available for the
budgets for the 1U CubeSat (single-payload in Section 5-2) processing-overpower mode during a single orbit.
and one budget for a 3U CubeSat (triple-payload in Section
tt Iproducfld + l' * PStOTfI
Iproducfld + l' * PStOTfI
5-3). The energy reserve budgets’ power modes are given as == ......;....--------
......;....--------
p PStOTfI - Ppayload
total power consumed. Using the eclipse time equations p PStOTfI - Ppayload
(6)
(Section 3), we calculate the total energy produced during a (6)
(6)
single orbit, which, when combined with the energy reserve where Pstore and Ppayload are the power consumptions during
budget, enables designers to calculate the percentage of time the power-storing and processing-overpower modes,
the system can spend in each power mode. respectively.

5-1. Energy Reserve Budget Development In order to generalize these equations to consider a system
with an arbitrary number of power modes, assumptions
Developing the energy reserve budget is an iterative
must be made about the required operational times (required
process, which becomes more refined and concrete
time spent in each power mode based on the mission’s
throughout the mission design process. Since early design
requirements) of all of the power modes except for one
stages have little information about the exact launch and
power mode. For example, if a communication-overpower
deployment details, designers must estimate these details in mode is added to the energy reserve budget for the previous
order to create an initial energy reserve budget and construct two-power-mode example, the designer must estimate the
power modes based on anticipated orbital patterns. Even

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 8

new communication-overpower mode’s required operational lower power camera and used the FPGA power estimation
time and power consumption in order to calculate the new from our power results) and the subsystems’ estimated
processing-overpower mode’s operational time tp. Similarly, power usage from Table 4. In our energy reserve budgets,
for a system with n power modes, the designer must the single payload’s subsystems are a camera and an FPGA.
estimate each power mode’s operational time and power The camera is the same OmniVision 2655 CMOS image
consumption in order to calculate the new processing- sensor used on the M-Cubed, which consumes 250 mW
overpower mode’s operational time tp. The following during operation. The FPGA is the Spartan-3 XC3A400T,
equation, which is a generalization of Equation (6), the lowest power consumption device from our power
calculates a power mode’s operational time tp, for a system results. Using these subsystems, we assess the energy
with n power modes: reserve budget for the Canny filter (Section 4-1).

We evaluate the energy reserve budgets for ten orbital

patterns. Table 6 shows the orbital patterns’ altitude, RAAN
(7) and inclination, which determine the Beta range, τ (orbital
where Pn and tn are the power consumption and operational period), and ts (eclipse times) for the given Beta ranges. The
time during a single orbit for power mode n, respectively. In ts (low), ts (high), and ts (avg) columns specify the lowest,
this example, the communications-overpower mode’s highest, and average eclipse times, respectively, for a
required downlink operational time is based on the satellite over a single year. The values in the table are based
mission’s requirement and must be estimated by the on an in-house orbital evaluation tool, which uses the
designer before tp can be calculated (7). equations in Section 3.

If the mission’s requirements dictate that a particular power The first energy reserve budget iteration requires several
mode requires the maximum operational time available—a assumptions about the mission’s required uplink and
maximized mode—Equation (7) can be further generalized downlink operational times. The communication
to calculate the maximum operational time τA by holding all subsystem’s transmission and reception frequencies
other power modes’ operational times constant, except one determine the maximum distance at which the
power mode—the donor mode. For example, if a mission communication signal can be reliably received, with lower
requires an operational time of at least three minutes per frequencies traveling further. Since our communication
orbital period for the processor-overpower mode and the subsystem is based on the M-Cubed, we assume that the
communication-overpower mode is the maximized mode, communication subsystem requires the same
the processor-overpower mode’s operational time is set to a communication time per orbit, which is five minutes of
constant of three minutes, the power-storing mode is the downlink time and ten minutes of uplink time. Even though
donor-mode, and τA is calculated as: not every orbit will require communication with the ground
station, many orbital patterns allow for multiple
communication opportunities per day. Therefore, it is
necessary to evaluate our energy reserve budget using the
(8) maximum available uplink and downlink operational times
where PA and PB are the power consumptions of the to calculate the available processing-overpower operational
maximized and donor modes, respectively. time.

5-2. Single-Payload 1U Case Study The power modes’ functionalities are defined based on the
mission requirements, and for this case study, we assume
For the single-payload 1U case study, we use the M-Cubed the following mission requirements. In the power-storing
subsystems (except for the payload, which we replace with a
Table 6: Orbital patterns

Orbit Altitude Inclination RAAN β range τ ts (low) ts (high) ts (avg)

Equatorial 300 km 0 0 -23o:23o 90.5 35.8 36.6 36.2
Inclination 22.5 400 km 22.5 30 -43o:45o 92.6 31.6 36.1 35.2
Inclination 45 500 km 45 60 -61o:65o 94.6 14.5 35.7 32.7
Inclination 67.5 600 km 67.5 90 -88o:85o 96.7 0 35.5 26.1
Inclination 90 (polar) 300 km 90 60 36o:60o 90.5 27.0 34.4 32.3
Inclination 112.5 400 km 112.5 30 -88o:88o 92.6 0 36.1 29.7
Inclination 135 500 km 135 0 -67o:63o 94.6 8.1 35.8 32.6
Inclination 157.5 600 km 157.5 0 -46o:46o 96.7 29.2 35.5 34.3
Inclination 180 800 km 180 0 -23o:23o 100.9 33.6 35.1 34.4
Sun-synchronous polar 300 km 90 0 71o:79o 100.9 0 0 0
Sun-synchronous nonpolar 400 km 98 0 -75o:82o 92.6 0 0 0
Epoch time of all Orbits is considered to be January 1st, 2011

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 9

 Table 7: Energy reserve budget for the single-payload 1U case study

Power Mode’s Power Consumptions (mW)

Subsystem Power-Storing Communication-overpower Uplink-overpower Processing-overpower
ADCS 0 0 0 0
C&DH 216 292.5 216 292.5
Uplink 0 616.5 616.5 0
Downlink 0 1000 0 0
Payload 0 0 0 387.5
Board 960 1254 960 1532
Total 1176 3163 1792.5 2212
mode, only the C&DH subsystem is active to enable entire five minutes of the required downlink operational
switching between power modes. This power mode time and five minutes of the required uplink operational
switching process is designer-defined and can be triggered time, leaving five minutes of operational time for the
by either external events (e.g., transitioning to eclipse) or an uplink-overpower mode to account for the remaining
internal schedule. In the communications-overpower mode, required uplink operational time. Table 8 shows the
the uplink and downlink subsystems are active to allow for processing-overpower mode operational times tp given the
the required uplink and downlink operational times and in worst-, best-, and average-case eclipse times (low, high, and
the uplink-overpower mode, only the uplink subsystem is avg, respectively) based on the required communications-
active. In the processing-overpower mode, the overpower and uplink-overpower modes’ operational times.
communication subsystems are turned off and all power is These results reveal that during eight of the ten orbital
used by the CD&H. patterns, the processing-overpower mode operation time tp
is negative for some orbits, which means that these orbital
Table 7 depicts the energy reserve budget with one power- patterns are not suitable for this CubeSat’s mission
storing mode and three overpower modes (two for requirements. Furthermore, the negative operational times
communications (communication- and uplink-overpower) indicate an excess power consumption due to the other
and one for payload operation (processing-overpower)), the power modes, which reveals that either the uplink-
power modes’ power consumptions per subsystem, and the overpower or communication-overpower mode is using too
power modes’ total power consumptions. Based on the EPS much energy to be sustained over that orbital pattern.
subsystem’s solar panel power production given in Table 5,
the power-storing mode will store energy over a single orbit, Given this energy reserve budget analysis, designers may
which is required to evaluate the operational times of all the either accept this energy reserve budget, which limits their
other higher power consumption modes in the energy mission to two orbital patterns or the designer may modify
reserve budget. Since the uplink subsystem consumes the mission’s requirements or subsystems to allow this
relatively little power, power is also stored in the batteries CubeSat to be amenable to more orbital patterns, and
during the communication-uplink-overpower mode. therefore more launch opportunities. Modifications to the
mission’s requirements include decreasing the power
We evaluate our energy reserve budget using the uplink and modes’ required operational times or power consumptions
downlink operational time requirements of ten and five or by completely replacing/redesigning subsystems.
minutes, respectively. Five minutes are spent in the Modifications to the Cubesat can be quickly evaluated using
communications-overpower mode, which accounts for the the existing energy reserve budget and the payload
operational time can be recalculated. Additionally, the
Table 8: 1U payload operational time in minutes during
designer may replace, redesign, or add new subsystems,
a single orbit
however, these modifications are more drastic since even
modification to a single component may affect the
Orbit tp (low) tp (high) tp (avg)
CubeSat’s capabilities and adversely affect the overall
Equatorial -31.87 -33.42 -32.65 system. For example, using a lower power downlink
Inclination 22.5 -22.03 -30.76 -29.02 subsystem may reduce the downlink communication
Inclination 45 12.75 -28.38 -22.56 resources, which in turn may require more efficient on-
Inclination 67.5 42.58 -26.30 -8.06 board data compression/preprocessing. After changes are
Inclination 90 (polar) -14.80 -29.15 -25.08 applied and new tp values can be calculated and evaluated
Inclination 112.5 39.28 -30.76 -18.35 and further changes to the CubeSat may be assessed.
Inclination 135 25.17 -28.57 -22.36
Inclination 157.5 -14.08 -26.30 -23.97 5-3. Triple-Payload 3U Case Study
Inclination 180 -19.23 -22.14 -20.78
For the triple-payload 3U case study, we use the QuakeSat
Sun-synchronous polar 45.96 45.96 45.96 subsystem design with the Magnetometer payload replaced
Sun-synchronous by the same camera used in the 1U case study (the
nonpolar 39.28 39.28 39.28

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 10

 Table 9: Energy reserve budget for the triple-payload 3U case study – first iteration

Power Mode’s Power Consumptions (mW)

Subsystem Power-Storing Communication- Processing- Processing- Processing-
overpower overpower-P1 overpower-P2 overpower-Pall
ADCS 0 0 0 0 0
C&DH 2500 2500 2500 2500 2500
Uplink 750 750 750 750 750
Downlink 0 1400 0 0 0
P (Cam) 0 0 250 250 250
P (FPGAs) 0 0 137.6 275.2 1651.2
Board 350 350 350 350 350
Total 3600 5000 3987.6 4125 5501.2

OmniVision 2655 CMOS image sensor) and combine processing-capable power mode where all twelve FPGAs
twelve low-power Spartan-3 XC3A400Ts for a distributed are active (i.e., the processing-overpower-Pall mode), the
data processing network. This distributed data processing first energy budget iteration allocates no operational time to
network combines the processing power of the COTS the processing-overpower-P1 or the processing-overpower-
FPGAs with the dependability and fault-tolerance of a DM P2 modes. If the processing-overpower-Pall mode’s
network [26] using the Prometheus CPU as the central operational time given this energy reserve budget indicates
processor. Similarly to the 1U case study, all of the FPGAs that the mission’s requirements cannot be met, then the
execute the Canny filter (Section 4-1) and have the same designer can consider using a lower processing-capable and
power consumption (Section 4-2). This case study’s goal is thus lower power, power mode (i.e., processing-overpower-
to maximize the payload operational time for payload- P1 or processing-overpower-P2).
overpower-Pall, since this mode provides the maximum
processing capability using all of the FPGAs. Table 10 shows the processor-overpower-Pall mode’s
operational times for each orbit given the assumed
Table 9 depicts the first iteration of the energy reserve communication-overpower mode’s operational time and the
budget for the 3U case study. Unlike the 1U case study in energy reserve budget (Table 9). The operational times
which the power resources were more constrained, the reveal that this energy reserve budget enables the CubeSat
uplink is continually powered on for all power modes, but to operate in the highest processing-capable power mode
similarly to the 1U case study, all downlink communication (i.e., processor-overpower-Pall) for ~180 minutes in the
occurs in the communications-overpower mode. worst-case orbit (the equatorial orbit), which is 2x greater
than the total orbital period for this orbit. This excess
We set the communication-overpower mode operational operational time indicates that this energy reserve budget is
time to ten minutes for downlink, which enables a data too conservative and additional operational time can be
transfer rate similarly to that of the original QuakeSat allocated to other power modes and still enable the
(3MB/day). Since it is preferable to operate in the highest processor-overpower-Pall mode to operate for 100% of the
orbital period.
Table 10: Processor-overpower-Pall mode’s operational
time in minutes during a single orbit for the triple- Similarly to the 1U case study, the designer may either
payload 3U case study – first iteration accept the energy reserve budget as is, allowing the CubeSat
to operate successfully in any orbit, or the designer can
Orbit tp (low) tp (high) tp (avg) create another iteration of the energy reserve budget to
Equatorial 186.20 180.31 183.25 optimize the CubeSat’s capabilities such that the CubeSat
will leverage the available power resources more efficiently.
Inclination 22.5 228.61 195.48 202.10
Since the QuakeSat designers reported that the
Inclination 45 365.47 209.36 231.45
communications downlink was less efficient than desired
Inclination 67.5 483.74 222.32 291.54
[6], the second energy reserve budget iteration focuses on
Inclination 90 (polar) 251.00 196.51 211.97 increasing the downlink operational time while not
Inclination 112.5 461.31 195.48 242.60 compromising the amount of data produced by the payload.
Inclination 135 412.60 208.63 232.19
Inclination 157.5 268.71 222.32 231.16 In order to leverage the available power resources more
Inclination 180 259.29 248.24 253.40 efficiently and increase the downlink time, the second
Sun-synchronous polar 506.71 506.71 506.71 iteration of the energy reserve budget could either increase
Sun-synchronous the other subsystems’ operational times or modify the
nonpolar 461.31 461.31 461.31 communication-overpower mode operational time.

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 11

 Table 11: Energy reserve budget for a triple-payload 3U case study – second iteration

Power Mode’s Power Consumptions (mW)

Subsystem Power-Storing Processing- Processing- Processing-
overpower-P1 overpower-P2 overpower-Pall
ADCS 0 0 0 0
C&DH 2500 2500 2500 2500
Uplink 750 750 750 750
Downlink 0 1400 1400 1400
P (Cam) 0 250 250 250
P (FPGAs) 0 137.6 275.2 1651.2
Board 350 350 350 350
Total 3600 5387.6 5525 6901.2

However, since the maximum operational time available to during all modes to meet data transfer requirements, except
the payload-overpower-Pall mode per orbit is the orbital in the power-storing mode, where the downlink is powered
period minus the operational time used by all other modes, off to ensure minimal system usage and maximum energy
increasing the communication-overpower mode’s storage. These changes increase the available downlink
operational time is counterproductive to the goal of time, increase payload operational time, and enable
maximizing the payload-overpower-Pall mode’s operational complete control from a ground station during all overpower
time. Alternatively, reducing the communication-overpower modes since both the uplink and downlink are always active
mode’s operational time would increase the payload- (except in the power-storing mode). Since an analysis
overpower-Pall mode’s operational time, but would reduce indicates that there is no need to reduce the payload’s
the communication-overpower mode’s operational time and processing capabilities (i.e., use the payload-overpower-
thus would reduce the amount of transmitted data. Given P1/P2 modes) the energy reserve budget only requires two
these tradeoffs, the designer’s best option to improve the power modes: the payload-overpower-Pall mode and the
power resource utilization and to increase the downlink time power-storing mode.
is to modify the power modes’ subsystem power usage.
Table 12 depicts the payload-overpower-Pall mode’s
Table 11 depicts the energy reserve budget’s second operational time assuming the energy budget in Table 11.
iteration with the goal of maximizing the CubeSat’s Similarly to the first iteration, the results indicate that even
performance by leveraging the downlink subsystem during though the CubeSat has all subsystems operational in the
the payload-overpower-Pall mode and increasing available processor-overpower-Pall mode, this mode’s operational
operational time for the payload-overpower-Pall mode. This time for all orbits is still greater than the corresponding
second iteration includes two modifications: 1) the orbit’s orbital period. Depending on the mission
communications-overpower mode’s operational time is requirements, a designer may consider this design complete,
reduced to zero and is removed from the energy reserve in which case the satellite could be launched into any
budget, which increases the available operational time for circular LEO orbit and operate in payload-overpower-Pall
the payload-overpower modes by ten minutes; and 2) since mode 100% of the time. Alternatively, since the operational
there is sufficient available power, the downlink is active times still indicate excess operational time (i.e., the payload-
Table 12: Processor-overpower-Pall mode’s operational overpower-Pall mode’s operational time is still greater than
time in minutes during a single orbit for the triple- the orbital period), the designer could perform additional
payload 3U case study – second iteration energy reserve budget iterations to further improve the
subsystems or the energy reserve budget to increase the
Orbit tp (low) tp (high) tp (avg) CubeSat’s processing capabilities. For example, a more
Equatorial 133.28 129.89 131.59 accurate and increased controlling-capable ADCS could be
Inclination 22.5 157.71 138.63 142.45 added if more accurate directional control were desired,
more powerful FPGAs could be used to increase the payload
Inclination 45 236.53 146.63 159.35
processing capabilities, the Tek-net could be replaced with a
Inclination 67.5 304.64 154.09 193.95
more robust (higher power consumption) radio to improve
Inclination 90 (polar) 170.60 139.22 148.13
communication reliability, or additional payloads or FPGAs
Inclination 112.5 291.72 138.63 165.77 could be added to maximize the processing capability given
Inclination 135 263.67 146.20 159.77 the size and weight restrictions.
Inclination 157.5 180.81 154.09 159.18
Inclination 180 175.38 169.02 171.99
Sun-synchronous polar 317.87 317.87 317.87
Sun-synchronous
nonpolar 291.72 291.72 291.72

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 12

 6. CONCLUSIONS validate the Virtex-5 FPGA for spaceborne image
processing," Aerospace Conference, 2010 IEEE, Big
Given the low-cost and fast design time afforded by Sky, MT, 6-13 March 2010
CubeSats, CubeSats are a popular platform for academia,
small companies, and countries without fully-funded space [5] P. Bernardi, M. Sonza Reorda, L. Sterpone, M.
programs. However, given the CubeSat’s tight design Violante,“On the evaluation of SEU sensitiveness in
constraints, leveraging high-performance payload SRAM-based FPGAs,” in IEEE IOLTS Proceedings,
processing components such as FPGAs is challenging. In 10., 2004, Los Alamitos : IEEE Computer Society,
this paper, we proposed an energy reserve budget and p.115-120., 2004
associated power modes to aid designers in most effectively
selecting CubeSat components and appropriate orbital [6] T. Bleier, P. Clarke, J Cutler, L. DeMartini, C. Dunson,
patterns based on a mission’s requirements. The energy et al, QuakeSat Lessons Learned: Notes from the
reserve budget enables quick calculation of the maximum Development of a Triple CubeSat, White Paper, June 4,
available operational time for a CubeSat’s payloads based 2003
on the solar panels’ power production and the batteries’
energy reserves. Using this information, designers can [7] H. Bock, U. Hugentobler, T.A. Springer, G. Beutler,
quickly select appropriate launch opportunities or redesign “Efficient precise orbit determination of LEO satellites
their system to meet a mission’s requirements based on a using GPS, Advances in Space Research,” Volume 30,
launch opportunity’s orbital pattern. Issue 2, July 2002, Pages 295-300, ISSN 0273-1177,
10.1016/S0273-1177(02)00298-3.
We evaluate our energy reserve budget using a one unit
(1U) and a three unit (3U) CubeSat for an image-processing [8] D. Harris, Electronic Design. "If only the original
mission using a Canny filter. Based on our selected FPGA Spartans could have thrived on so little power,"
device and ten sample orbital patterns, the energy reserve February 27, 2008
budget revealed that only two of these orbital patterns could
meet the mission requirements. Using these results, a [9] L. David (2006). CubeSat losses spur new development
designer could either refine the mission’s requirements or [Online] available at http://www.cnn.com/
change subsystem components to make their design more
amenable to more orbital patterns. Our future work includes [10] H. Heidt, C Turner, R. Twiggs, “CubeSat: A new
evaluating the Cyclone and Artix families, which show generation of picosatellite for education and industry
promising results based on power estimations. low-cost space experimentation.” in Proceedings of the
14th Annual AIAA/USU Conference on Small Satellites,
7. ACKNOWLEDGMENTS SSC00-V-5, Logan, UT, Aug. 2000

This work was supported in part by the National Science [11] H. Ashida, K. Fujihashi, S. Inagawa, Y. Miura, K.
Foundation (ECCS-0901706 and the I/UCRC Program Omagari, et al, “Design of Tokyo Tech nano-satellite
EEC-0642422). Any opinions, findings, and conclusions or Cute-1.7+APD II and its operation,” in Acta
recommendations expressed in this material are those of the Astronautica, Volume 66, Issues 9-10, May-June 2010,
author(s) and do not necessarily reflect the views of the Pages 1412-1424,ISSN 0094-5765,
National Science Foundation. The authors gratefully 10.1016/j.actaastro.2009.10.035.
acknowledge vendor equipment and tools provided by
Xilinx that helped make this work possible. [12] H.S. Neoh, A. Hazanchuk, “Adaptive Edge Detection
for Real-Time Video Processing using FPGAs,” GSPx
8. REFERENCES 2004 Conference, 2004.

[1] L. Alminde, M. Bisgaad, D. Vinter, T. Viscor, and K. [13] H. Ding, L. Arber, L. Sha, M. Caccamo, "The
Z. Østergard, "The AAU-CubeSat student satellite Dependency Management Framework: A Case Study of
project: architectural overview and lessons learned," in the ION CubeSat," in 18th Euromicro Conference on
Proceedings of the 16th IFAC Symposium on Automatic Real-Time Systems (ECRTS'06), pp. 52-64, 2006
Control in Aerospace, 2004
[14] A. Jacobs, G. Cieslewski, C. Reardon, and A. George,
[2] Altera inc., “Stratix III FPGAs vs. Xilinx virtex-5 "Multiparadigm Computing for Space-Based Synthetic
Devices Architecture and Perfomance Comparison”, Aperture Radar," in Proceedings of 2008 International
White Paper, October 2007 Conference on Engineering of Reconfigurable Systems
and Algorithms (ERSA), Las Vegas, NV, July 14-17,
[3] Altera inc, Altera Powerplay Early power Estimators, 2008.
copywrite 2005-2011
[15] A. Jacobs, C. Conger, A.D. George, "Multiparadigm
[4] D.L. Bekker, T.A. Werne, T.O. Wilson, P.J. Pingree, K. Space Processing for Hyperspectral Imaging," in
Dontchev, M. Heywood, et al, "A CubeSat design to

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 13

 Aerospace Conference, 2008 IEEE, Big Sky, MT, March [27] J. Schaffner, “The Electronic System Design, Analysis,
2008 Integration, and Construction of the Cal Poly State
University CP1 CubeSat” in 16th AIAA/USU on Small
[16] M. Kovac, N. Ranganathan, "JAGUAR: a fully Satellites Conference, Logan, UT, October 2002, pp. 1-2
pipelined VLSI architecture for JPEG image
compression standard," in Proceedings of the IEEE, [28] A. Toorian, K. Diaz, S. Lee, "The CubeSat Approach to
vol.83, no.2, pp.247-258, Feb 1995 Space Access," in Aerospace Conference, 2008 IEEE,
pp.1-14, 1-8 March 2008
[17] M. Long, A. Lorenz, G. Rodgers, E. Tapio, G. Tran, et
al, “A CubeSat Derived Design for a Unique Academic [29] Xilinx Inc. Spartan-3 vs. Cyclone I Performance
Research Mission in Earthquake Signature Detection,” Analysis, White Paper, May 24th 2005
in 16th Annual AIAA/USU Conference on Small
Satellites, Utah, USA, 2002. [30] Xilinx Inc. 7-series Power Estimator. copywrite 2011

[18] Mike’s list of CubeSat Satellite Missions [Online]. [31] Xilinx Inc. Xilinx ISE Design Suite 13.2. copywrite
Available at http://mtech.dk/thomsen/space/cubesat.php 2011

[19] K. Morris, “Veni! Vidi! Virtex! (and Kintex and Artix

Too),” In FPGA Journal June 21, 2010
9. BIOGRAPHY
[20] I. Nason, M. Creedon, J. Puig-Suari, “CubeSat Design
Specifications Document,” Revision V, pp. 1-6. Scott Sterling Arnold received his Bachelor of
Available at http://ssdl.stanford.edu/cubesat, Nov. 2001 Science in Electrical Engineering in 2011 from the
University of Florida. Currently he is a PhD student at
[21] I. Nason, J. Puig-Suari, R. Twiggs, "Development of a the University of Florida and is a student member of
the NSF Center for High Performance Reconfigurable
family of picosatellite deployers based on the CubeSat Computing (CHREC). During his undergraduate
standard," in Aerospace Conference Proceedings, 2002. studies, he has worked for GE Appliances. His
IEEE, vol.1, pp. 1-457- 1-464 vol.1, 2002 interests include CubeSat design, embedded systems
design, computer architecture, reconfigurable and
[22] OmniVision, “Datasheet: Preliminary Specifications adaptive computing, and space computing.
1/5” CMOS (2 megapixel) image sensor with Ryan Nuzzaci is a 2nd year Ph.D. student in
OmniPixel3-HS technology”, available at Computer Engineering at the University of Florida.
http://www.trulydisplays.com/ccm/specs/2.0M%20Sens He is the lead hardware development engineer for the
or%20OV2655%20Spec.pdf Small Sat Group and is currently designing the
payload board for INTA's OPTOS II CubeSat. He
received his B.S. and M.S. in Computer Engineering
[23] C. Eagle, Shadow Conditions of the Earth in Circular from the University of Florida in Dec. 2009 and Dec.
Orbits [Online]. Available at 2011, respectively. Following graduation, he plans on
http://www.cdeagle.com/html/omnum.html pursuing a career in embedded systems design engineering. He is an active
chair for UF IEEE.
[24] P. Pingree, T. Werne, D. Bekker, T. Wilson, J. Cutler, Ann Gordon-Ross (M’00) received her B.S and
M. Heywood, “The Prototype Development Phase of the Ph.D. degrees in Computer Science and
CubeSat On-board processing Validation Experiment Engineering from the University of California,
(COVE),” in IEEE Proc. 2011 Aerospace Conference, Riverside (USA) in 2000 and 2007, respectively.
She is currently an Assistant Professor of ECE at
Big Sky, MT, 2011 the University of Florida and is a member of the
NSF Center for High Performance
[25] J. Puig-Suari, C. Turner, W. Ahlgren, "Development of Reconfigurable Computing (CHREC). She is
the standard CubeSat deployer and a CubeSat class also faculty advisor for the Women in Electrical
PicoSatellite," in Aerospace Conference, 2001, IEEE and Computer Engineering (WECE) and the Phi
Sigma Rho National Society for Women in Engineering and Engineering
Proceedings. Vol.1, pp.1/347-1/353 vol.1, 2001 Technology. She received her CAREER award from the National Science
Foundation in 2010 and Best Paper awards at the Great Lakes Symposium
[26] J. Samson, J. Ramos, I. Troxel, R. Subramaniyan, A. on VLSI (GLSVLSI) in 2010 and the IARIA International Conference on
Jacobs, et al, “High-Performance, Dependable Mobile Ubiquitous Computing, Systems, Services and Technologies
(UBICOMM) in 2010. Her research interests include embedded systems,
Multiprocessor,” in Proc. of IEEE/AIAA Aerospace, Big computer architecture, low-power design, reconfigurable computing,
Sky, MT, Mar 4-11, 2006 dynamic optimizations, hardware design, real-time systems, and multi-core
platforms.

978-1-4577-0557-1/12/$26.00 ©2012 IEEE 14