SSC14-III-6

Design and Testing of a Low-Cost MEMS IMU Cluster for SmallSat Applications
Daniel R. Greenheck, Robert H. Bishop, Eric M. Jonardi
Marquette University
Milwaukee, WI 53233; 715-571-5690
[email protected]

John A. Christian
West Virginia University
Morgantown, WV 26506

ABSTRACT
Small satellite missions are characterized by tight constraints on cost, mass, power, and volume that generally make
them unable to fly inertial measurement units (IMUs) required for orbital missions demanding precise orientation
and positioning. Instead, small satellite missions typically fly low-cost micro-electro-mechanical system (MEMS)
IMUs. The performance characteristics of MEMS IMUs make them ineffectual in many spaceflight applications
when employed in a single IMU system configuration. The challenge for small satellite designs aiming to tackle
more aggressive missions is to creatively employ advanced software algorithms coupled with embedded system
architectures to create an effective precision IMU from clusters of low-cost MEMS IMUs. The objective of this
work is to develop and demonstrate a MEMS IMU cluster whose composite output provides high performance while
remaining within the mass, power, and volume constraints of a 1U CubeSat. Successfully achieving this objective
will represent a new class of inertial navigation performance for the small satellite platform. We investigate the
practical issues associated with implementing an IMU cluster in a form factor suitable for use on a 1U CubeSat. The
results show that in general, simple averaging of the sensor outputs approaches the predicted square root of N
improvement in performance for the RMS noise and bias stability of the sensors. However, some sensors exhibited
lower performance improvements than other sensors, indicating a higher correlation between individual sensors.

INTRODUCTION change in orientation is measured relative to a set of

three mutually orthogonal axes [2]. These two sensors
Definitions and Applications allow the complete state of a body (position, velocity,
An inertial measurement unit (IMU) is a device capable orientation) to be determined.
of sensing non-gravitational accelerations and angular
Vehicle navigation is one of the primary applications of
rates that can be employed to propagate vehicle
an IMU. While this work focuses on satellite
position, velocity, and attitude between external updates
applications, IMUs are also extensively used in
(e.g. Global Positioning System or a star tracker) [1].
This can be accomplished by measuring accelerations robotics, self-driving cars, aircraft, nautical vessels and
acting on the body as well as tracking any changes in a host of other vehicles. Outputs from the IMU are
utilized by signal processing and control algorithms
the orientation of the body to allow the reference frame
from which changes in orientation and velocity may be
of the body to be transformed back to an inertial frame.
sensed. For many space applications, the gyros play a
Accelerations are most often measured using linear more important role than the accelerometers, except
accelerometers. A linear accelerometer outputs an during propulsive maneuvers. Otherwise, in low-Earth
analog or digital electrical signal proportional to the orbit, the linear accelerometers only sense aerodynamic
linear acceleration experienced by the sensor along a drag and these accelerations may be small compared to
predefined axis. An accelerometer is typically selected the noise on the acceleration measurements. Thus, it is
that measures accelerations along three mutually common for an acceleration threshold to be set below
orthogonal axes to provide a full description of the which the output of the accelerometer is not considered
accelerations acting upon the body. A gyroscope (or and spacecraft translational motion is described by
gyro) is an instrument able to track changes in orbital mechanics.
orientation over time. Similar to the accelerometer, the

Greenheck 1 28th Annual AIAA/USU

Conference on Small Satellites
An IMU can be used in conjunction with external countercurrent light paths. As the drum rotates about its
navigation updates, such as GPS, or act independently, axis, the transmit time for each path is altered, resulting
resulting in a completely self-contained navigation in a phase difference which can be detected at the point
solution. An example of when self-contained navigation of entry via an interference measurement. This phase
is often employed is during spacecraft re-entry. During difference can be measured and related to the angular
hypersonic re-entry of a spacecraft, a plasma sheath rate [5].
surrounds the vehicle as a result of the extreme heating
of the air generated by strong shock waves at the One of the technologies developed in recent years is the
vehicle leading edges. The plasma, consisting of ions MEMS gyroscope. These sensors are inexpensive,
and free electrons, can reflect and attenuate the small and have few moving parts [5]. Although there
propagation of electromagnetic waves, including those are several sub-classifications within the grouping of
at frequencies used for radio communications. This MEMS gyroscopes, only Coriolis vibratory gyroscopes
results in a “communication blackout” for some period will be considered here. In a single-axis configuration,
of time during re-entry. The spacecraft must rely on two proof masses are attached on either side of a fixed
inertial navigation during this time since external platform. The masses are driven at a specific frequency
navigational updates cannot be communicated to the along the horizontal plane. If a rotation is applied to the
spacecraft [3,4]. axis normal to the plane, the Coriolis force causes the
two masses to deflect in opposite directions. A
Although the concept of a completely self-contained differential capacitive measurement results in a value
navigation system is enticing, one disadvantage of corresponding to the angular rate [5].
IMUs is that navigation errors tend to quickly build up
over time. Determining the position requires integrating Several accelerometer technologies include the force-
the measured acceleration twice with respect to time. If feedback accelerometer, the vibratory accelerometer
a constant bias is present in the linear acceleration and the pendulous mass MEMS accelerometer. The
measurement, it would result in a position error that mechanical force-feedback accelerometer consists of a
grows quadratically with time. Similarly, the output of pendulum with an attached proof mass. Initially, the
the gyro must be integrated once with respect to time in pendulum is at a standstill in the equilibrium position.
order to determine the orientation. A constant bias error As an acceleration is applied, the pendulum moves
in the rate of change of the orientation measurement from its equilibrium position. The deviation is directly
would result in a ramp error in the orientation proportional to the applied acceleration.
measurement. Because of this, much effort has been put
in to developing new technologies to minimize the The vibratory accelerometer has two quartz crystal
various errors found in inertial navigation systems. beams positioned back to back, each supporting a proof
mass. The beams vibrate at their own resonant
Accelerometer and Gyroscope Technologies frequency. When an acceleration is applied along the
sensitive axis, one beam undergoes compression while
Different types of inertial sensor technologies for both
the other undergoes expansion, altering the resonant
accelerometers and gyros have been developed to
frequency. The difference in resonant frequencies is
mitigate errors, each having different tradeoffs between
directly proportional to the applied acceleration.
performance, mass, volume and power consumption.
The three most common types of gyroscopes are the There are two main types of MEMS accelerometers:
ring laser gyroscope (RLG), the fiber optic gyroscope pendulous mass and vibrating beam. Only pendulous
(FOG), and the micro-machined electromechanical mass accelerometers will be considered here. In this
system (MEMS) gyroscope. configuration, a polysilicon structure is suspended over
a silicon substrate by a set of polysilicon springs,
Mirrors are aligned in the RLG to produce a closed
creating a differential capacitor. As the structure
light path. Two counter-propagating laser beams are
deflects due to accelerative forces, the capacitance of
directed along this path. As the apparatus is rotated
the differential capacitor changes. The magnitude and
about the axis normal to the plane of the light path, the
direction of the acceleration can be determined via
path length of each beam is altered resulting in a phase
phase-sensitive demodulation techniques [5,6].
difference which can be measured by allowing the two
beams to undergo interference. This phase difference is Aims of Study
proportional to the angular rate [5].
In this study, we seek to demonstrate that placing
FOGs operate on a similar principle. A coil of fiber MEMS sensors in a cluster configuration results in the
optic cable is wrapped around a cylindrical drum. Light predicted improvements [7]. The cluster configuration
is directed through a beam splitter creating two consists of placing multiple MEMS sensors on a single

Greenheck 2 28th Annual AIAA/USU

Conference on Small Satellites
circuit board and synthesizing their outputs to create a where N is the setting. At present, the FPGA collects
single output which has improved performance the data in parallel from the sensor channels and
characteristics over that of a single MEMS sensor. We transmits it serially over USB; all processing is done
examine the benefits of simple averaging of the sensor via an external processing unit. However, the authors
outputs. would like to mention that the FPGA has sufficient
internal logic resources to implement signal filtering
A custom board was designed and fabricated to test the and navigation algorithms within the FPGA itself in the
objective of fusing many simultaneous MEMS IMU future. With the addition of an on-board wireless
measurements. What follows is a description of the system, this would eliminate the need for an external
physical board layout, the process used to calibrate the processing unit. Our current goal of determining how
sensors, and a comparison between the RMS noise and the performance of a MEMS IMU cluster compares
bias stability of a single MEMS sensor and a custom with single MEMS gyro and accelerometer does not
MEMS IMU cluster. Additionally, the MEMS IMU require on board processing.
cluster described here is compared to a number of
current IMUs on the market of varying degrees of A UART/FIFO IC (FTDI FT2232HL) serves as a
performance. A comparison that scales the performance communication bridge between the FPGA and an
by the power and mass of the devices is also included, external Mini-USB connection. The connection is used
as these constraints are often just as important as the to both collect data via a host computer as well as
performance of the IMU in small satellite applications. controlling the internal settings of the IMU. A second
Mini-USB port provides auxiliary power to the board as
BOARD DESIGN AND LAYOUT the power consumption of the board exceeds the 500
The IMU is 10 cm x 10 cm circuit board designed to fit mA maximum limit for USB.
within a 1U CubeSat form factor. The sensor cluster is
Figure 1 illustrates the rendered exploded view of IMU
made up of 16 individual sensor groups. Each sensor
assembly and custom enclosure and Figure 2 shows the
group contains two dual-axis analog MEMS gyros (ST
rapid prototype of IMU enclosure. The enclosure was
Microelectronics LPY410AL) as well as a single tri-
designed to contain the entire IMU cluster in
axial analog MEMS accelerometer (Analog Devices
preparation for sounding rocket flight testing. The
ADXL335), for a total of 32 gyros and 16
schematic of the integrated system is shown in Figure 3.
accelerometers on a single board, or 112 individual
The main elements of the unit are the battery, voltage
inertial measurements. Dual-axis gyros were chosen
regulator and charging circuit, the MEMS IMU cluster,
instead of tri-axial because the selection of single-
and the Raspberry Pi processor. We collect data
package tri-axial analog output MEMS gyros is limited.
through a wireless dongle to a remote laptop. We
Each pair of gyros is perpendicularly aligned to provide
a measurement of all three axes. The seven analog currently do not perform significant calculations on the
sensor outputs of each sensor channel are sampled onboard processor, but that is a target for our coning
and sculling algorithms as we further develop the IMU
sequentially by a high speed 16-bit analog-to-digital
cluster.
converter (ADC). The high sampling speed of the ADC
(Texas Instruments ADS8332) compared to the
External Processing
maximum output bandwidth of the analog sensors
External processing is carried out on a Raspberry Pi
(500kSPS vs 1.6kHz) allows near-simultaneous
Model B (512 MB of RAM, Broadcom BCM2835
sampling of the seven sensor outputs within each sensor
700MHz ARM1176JZFS processor). The Raspberry Pi
channel.
was chosen for the prototype due to its simplicity,
Each sensor channel has its own ADC, allowing wireless capabilities and small form factor. Currently,
the Raspberry Pi receives the data from the IMU board
simultaneous sampling of a single IMU axis (e.g. the x-
over a serial connection and transmits it wirelessly to a
axis on the MEMS accelerometer) across all sensor
host computer via TCP/IP connection. The wireless
channels. An op-amp buffer in a unity gain
connection is established over WiFi using an 802.11n
configuration is placed between each sensor output and
wireless adapter (Edimax EW-7811Un).
the ADC as a precaution to ensure sufficient drive
strength into the ADC inputs.

A FPGA (Lattice XP2-8) located at the center of the

board serves as a massively parallel I/O processor to
simultaneously acquire data from the ADCs. The FPGA
has two onboard decimation units to control the rate of
data output. Each will discard N of N+1 data sets,

Greenheck 3 28th Annual AIAA/USU

Conference on Small Satellites
 Figure 3: Flow diagram of IMU connections

Figure 1: Rendered exploded view of IMU

assembly and custom enclosure

Figure 4: IMU test fixture mounted on rate table

as the cables are bulky and add unnecessary weight to
the assembly. Linear power supplies were used on the
IMU board instead of switching supplies to minimize
electrical noise. A number of additional noise reduction
techniques were also employed to minimize electrical
noise as much as possible.

CALIBRATION
Figure 2: Rapid prototype of IMU enclosure
Measurement Setup
Power Calibration of the IMU board was performed on a
The entire assembly is powered by an off-the-shelf single-axis rate table (Trio-Tech Model 1102). The rate
consumer Lithium-polymer battery pack with a capacity table itself was calibrated by a trained technician before
of 6600 mAh and regulated output voltage of 5V. The testing. Before any measurements were taken, the table
battery pack has two USB-A outputs which provide 1A was first leveled using a built-in bubble level on top of
and 2A of current. There is also a Micro USB port on the rate table. The test setup is shown in Figure 4.
the side for charging the battery pack. This battery pack
was chosen because of its sufficiently small form factor Data Collection
and weight, as well easily interfacing with the USB Sensor data was collected at 100 °/s clockwise, 100 °/s
connections on the MEMS IMU board and the counter-clockwise, and at zero rate. Data was recorded
Raspberry Pi. The battery is able to power the IMU and in six different orientations: each sensing axis of the
Raspberry Pi for approximately 4-5 hours at full charge. IMU was aligned to be parallel with the gravitational
Future work includes eliminating the USB connections vector, as well as anti-parallel. Accelerometer data was

Greenheck 4 28th Annual AIAA/USU

Conference on Small Satellites
collected at zero rate with the gravitational force being where 𝒂!"#$ is a vector of the digital outputs of the
the input acceleration. accelerometer, 𝒂!"#$ is the acceleration truth vector, 𝑩!
is a vector of the accelerometer bias values for each
Accelerometer Calibration axis, 𝑺! is the accelerometer sensitivity matrix and 𝚪! is
The bias level, scale factors, and misalignment for each the accelerometer misalignment matrix. The sensitivity
individual accelerometer were calculated using the test matrix and misalignment matrix are defined below
procedures outlined in [8]. The bias level was
calculated by mounting each sensing axis of the 𝑆!,! 0 0
accelerometers both parallel with the gravitational 𝐒! =   0 𝑆!,! 0 (6)
vector parallel as well as anti-parallel, for a total of six 0 0 𝑆!,!
orientations. For a single axis, the bias level is given by
0 𝛿!,!" 𝛿!,!"
𝑎! + 𝑎!
𝐵! = (1) 𝚪! =   𝛿!,!" 0 𝛿!,!" (7)
2
𝛿!,!" 𝛿!,!" 0
where 𝑎!  is the digital output of the accelerometer with
an input acceleration of +g and 𝑎!  is the digital output Gyroscope Calibration
of the accelerometer with an input acceleration of –g.
The units of 𝐵! are given in µg. The scale factor for a The process of calibrating the gyros is analogous to the
single axis is calculated similarly by process of calibrating the accelerometers [9]. The
equations for the x-axis bias level, scale factor and
𝑎! − 𝑎! misalignment angles are given by
𝑆! = (2)
2𝐵!
𝜔! + 𝜔!
𝐵! = (8)
!"# 2
with the units of 𝑆! given in .
!"
𝜔! − 𝜔!
𝑆! = (9)
Calculation of the inter-axial error, or misalignment 2𝐵!
angles, requires four separate measurements. Using the
x-axis as an example, the x-axis is first aligned 𝜔!,!! − 𝜔!,!!
horizontally with the y-axis pointing downwards. The 𝛿!,!" = (10)
2𝑆!
output of the x-axis is recorded (𝑎!,!! ). The same
procedure is repeated, except with the y-axis pointing
𝜔!,!! − 𝜔!,!!
upwards. The output of the x-axis is recorded (𝑎!,!! ). 𝛿!,!" = (11)
The misalignment angle with respect to the y-axis is 2𝑆!
then given by (3). Equation (4) follows similarly.
where 𝜔! is a positive rotation about the axis being
𝑎!,!! − 𝑎!,!! sensed, 𝜔! is a rotation rate equal in magnitude to 𝜔!
𝛿!,!" = (3) except with opposite sign, 𝜔!,!! , 𝜔!,!!    are positive
2𝑆!
and negative rotations of equal magnitude about the y-
𝑎!,!! − 𝑎!,!! axis, respectively, and 𝜔!,!! , 𝜔!,!!    are positive and
𝛿!,!" = (4) negative rotations of equal magnitude about the z-axis,
2𝑆!
respectively. The equations for the y and z axes follow
similarly. The parameters can be combined into a
Knowing the bias level, scale factor and misalignment
model of the gyroscope, given by
angles for each individual sensor allow for a basic
calibration of the IMU. However, this calibration does
not take into account environmental thermal 𝝎!"#$ = (𝐈 + 𝑺! )(𝐈 + 𝚪! )(𝝎!"#! + 𝑩! ) (12)
fluctuations. To minimize this effect, the data was
acquired at night when the temperature of the room was where 𝝎!"#$ is a vector of the digital outputs of the
the most stable. The parameters can be combined into a gyro, 𝑩! is a vector of the gyro bias values for each
model of the accelerometer, given by axis, 𝑺! is the gyro sensitivity matrix and 𝚪! is the gyro
misalignment matrix and 𝝎!"#$ is the angular rate truth
𝒂!"#$ = (𝐈 + 𝑺! )(𝐈 + 𝚪! )(𝒂!"#$ + 𝑩! ) (5) vector.

Greenheck 5 28th Annual AIAA/USU

Conference on Small Satellites
EXPERIMENT DESIGN sensor clusters. The Allan variance method is a way of
separating out different noise sources which contribute
Introduction
to the overall signal noise. In basic terms, an Allan
When comparing the performance of IMUs, there are variance plot shows the variance of the noise as a
often several standard metrics used to facilitate a function of the averaging time.
standard comparison between different technologies.
The Allan variance is calculated according to [3] using
Bias stability is the deviation from the mean value over
the following formula
a period of time given a constant input. With units of
deg/hr for gyros or µg for accelerometers, a higher
!!!
value for bias stability indicates that the bias has large 1 !
fluctuations about the mean. Such fluctuations have 𝜎!! 𝛥𝑡 = 𝑥! − 𝑥!!!   (14)
2(𝑁 − 1)
implications on how IMU biases should be treated in !!!
Kalman filters that may process these measurements.
Rate noise density is an indication of noise at the output where 𝑥! , 𝑖 = 1, … , 𝑁 is the time series data and Δ𝑡 is
of the sensor. The faster a sensor is sample, the more the sample time. Then a new sequence is defined
the noise will affect the final measurement. For gyros,
this measure is often expressed as angle random walk 𝑥!! + 𝑥!!!! 𝑁
(ARW) in units of degrees per root hour. ARW is the 𝑦! = , 𝑗 = 0,1, … ,   (15)
standard deviation of the error introduced by integrating 2 2
noisy gyro measurements. Scale factor stability is a
measure of the deviation of the scale factor for constant and the Allan variance 𝜎!! 2Δ𝑡 is computed for that
conditions over time and between different runs. sequence. This process continues for longer and longer
Finally, non-orthogonality refers to the errors in the averaging times until a new 𝑦! can no longer be formed.
alignment of the sensor axes. This includes angular The square root of the Allan variance is then plotted on
deviations of the axes from their true orientation as well
a log-log plot for analysis. Although it will not be
as inter-axial angular errors.
shown here, the bias stability can be determined by
looking at the minimum point on the Allan deviation
For this study, the authors choose to experimentally
curve.
determine bias stability and noise characteristics for
both a single MEMS gyro/accelerometer, as well as the
Experimental Setup
MEMS cluster.
The IMU cluster assembly was mounted to the rate
RMS Noise table to provide a flat surface for testing. The negative
z-axis of the IMU was aligned parallel with the
For a deterministic signal with additive zero-mean gravitational vector. The onboard decimation units were
Gaussian noise which is uncorrelated with the signal, set to produce a data output rate of 18.9 Hz. Data was
the improvement in the signal to noise ratio (SNR) by continuously collected for eight hours, resulting in N =
averaging N separate sample records is given by 545410 data points for each of the 112 sensor axes.

𝑆 RESULTS AND DISCUSSION

𝑆𝑁𝑅 = 𝑁   (13)
𝜎 RMS Noise

Since the IMU cluster has N=16 sensor groups that will Figures 5 and 6 illustrate the difference in RMS noise
be averaged together, our expectation is that the between a single MEMS sensor and the sensor cluster
improvement in bias stability and noise will be on the for each axis of the gyros and accelerometers,
order of 𝑁 or a factor of 4. Because we are using respectively. The results are summarized in Table 1.
multiples of the same sensor, not all of the noise The first column of Table 1 shows the RMS noise for a
sources will be completely uncorrelated. Therefore, the single sensor, the second column shows the RMS noise
actual improvement is expected to be less than what for the sensor cluster and the final column shows the
theory predicts. ratio between the single sensor value and the sensor
cluster value.
Method for Determining Bias Stability
The authors would like to point out that for the
The bias stability was calculated by creating an Allan ADXL335 accelerometer, the noise density specified
variance plot for each individual sensor as well as the for the z-axis of the accelerometer (referred to as the y-
axis in this paper) is twice the value for the other two

Greenheck 6 28th Annual AIAA/USU

Conference on Small Satellites
axes. This explains why the RMS noise for the y-axis is Single Sensor
IMU Cluster
twice that of the x and z axes.
4 X-axis
x 10
2

As predicted, the improvement in noise is close to a 1

micro-g
factor of 4. The gyros showed the highest average 0

-1
performance realization. The gyro x-axis was the best
-2
performing, with a ratio of 3.9 between the single 500 1000 1500 2000 2500
sensor performance and the cluster performance. The Time (s)

accelerometers also saw solid improvements in

4 Y-axis
x 10

performance. The accelerometer z-axis did not perform 2

as well as the other two axes (performance ratio = 2.7), 1

micro-g
0
suggesting that the correlation of the noise between
-1
sensors increases as they are subjected to an
-2
acceleration. 500 1000 1500 2000 2500
Time (s)

Table 1: RMS noise comparison between single 4

x 10 Z-axis

sensor and sensor cluster 2

1

micro-g
Accelerometer Single (µg) Cluster (µg) Ratio 0

-1
X-axis 3670 1060 3.5
-2
Y-axis 6250 1800 3.5 500 1000 1500 2000 2500
Time (s)
Z-axis 3590 1330 2.7
Average 4500 1400 3.2
Figure 6: RMS noise of single MEMS accelerometer
Gyroscope Single (deg/s) Cluster (deg/s) Ratio vs. MEMS cluster
X-axis 0.418 0.108 3.9
Y-axis 0.393 0.109 3.6 Bias Stability
Z-axis 0.422 0.111 3.8 Figures 7 and 8 show the Allan variances for both the
Average 0.411 0.109 3.8 gyros and accelerometers in the single sensor and
cluster configurations. For the case of a single sensor,
Single Sensor the Allan variances for each individual sensor were
2
X-axis IMU Cluster calculated. For the cluster, the Allan variances were
averaged across each group of sensors. The bias
1
stability for each axis is indicated by a horizontal dotted
deg/s

0
line corresponding to the minimum of the graph. The
-1 results are summarized in Table 2. The average
-2 performance ratio for the accelerometers (ratio = 2.8)
500 1000 1500 2000 2500
Time (s) and the gyros (ratio = 3.0) shows a significant
2
Y-axis
improvement via simple averaging. However, in
1
comparison with the RMS noise, the performance
improvements are not quite as pronounced. This may
deg/s

0
have been caused by changes in the bias due to
-1
temperature fluctuations since the experiment was not
-2
500 1000 1500 2000 2500
carried out in a controlled thermal environment.
Time (s) However, data was collected during the night to
Z-axis
2 minimize any temperature fluctuations. The gyro y-axis
1
stands out as the improvement ratio was only 2.4,
compared to 3.3 for both the x-axis and z-axis. Data
deg/s

0
was recorded at different orientations to determine if
-1
this artifact was the result of g-dependent errors, but the
-2
500 1000 1500 2000 2500 results did not indicate any specific correlation with the
Time (s)
orientation. The cause of the reduction in performance
for the y-axis has not yet been determined.
Figure 5: RMS noise of single MEMS gyro vs.
MEMS cluster The performance ratio for the other axes are consistent
with the expected results given the fact that there are

Greenheck 7 28th Annual AIAA/USU

Conference on Small Satellites
unmodeled, correlated errors between the individual 4
Allan Variance for Single Gyroscope
10
sensors. X-axis
Y-axis
Z-axis
Table 2: Bias stability comparison between a single 10
3

sensor and sensor cluster

deg/hr
Accelerometer Single (µg) Cluster (µg) Ratio 2
10
X-axis 62 23 2.7
Y-axis 152 56 2.7
1
10
Z-axis 61 21 2.8 10
-2 -1
10 10
0
10
1
10
2 3
10
4
10
τ (sec)
Average 92 33 2.8
Gyroscope Single (deg/hr) Cluster (deg/hr) Ratio 3
Allan Variance Plot for Gyroscope Cluster
10
X-axis 29.5 9.0 3.3 X-axis
Y-axis
Y-axis 25.3 10.6 2.4 Z-axis
2
10
Z-axis 26.6 8.0 3.3

deg/hr
Average 27.1 9.2 3.0
1
10

Allan Variance Plot for Single Accelerometer (Average variance)

4 0
10 10
-2 -1 0 1 2 3 4
X-axis 10 10 10 10 10 10 10
Y-axis τ (sec)
Z-axis
3
10
Figure 8: Allan variance of single MEMS gyro vs.
micro-g

MEMS cluster
2
10
Comparison with Other IMUs

1
A secondary aim of this work is to compare the
10
10
-2
10
-1
10
0 1
10 10
2 3
10
4
10 performance of the MEMS IMU cluster to existing
τ (sec) IMU sensors on the market. A direct comparison
Allan Variance Plot for Accelerometer Cluster
between a MEMS sensor and a RLG or FOG based
sensor shows that much progress must be made before
4
10
X-axis
Y-axis MEMS can be used in precision applications. However,
3
Z-axis
for applications where quantities such as mass, power
10
and volume are highly constrained, it is worth
micro-g

examining the tradeoff between mass, power, and

10
2 performance. Figures 9 and 10 show the comparison
between the MEMS IMU cluster and other types of
IMUs on the market [10,11,12,13,14]. It is notable that
10
1
-2 -1 0 1 2 3 4
the IMU cluster shown here is consistent with the mass,
10 10 10 10
τ (sec)
10 10 10
power, and performance trends seen with other IMUs.
Figure 7: Allan variance of single MEMS
By incorporating more precise MEMS IMUs, it is
accelerometer vs. MEMS cluster
anticipated that the IMU cluster performance can be
significantly improved with little change in the mass or
power consumption. Replacing the analog sensors with
digital output sensors would eliminate the need for the
op-amps and ADCs, which would drastically reduce
power consumption and shrink the board size due to
fewer physical components being on the board.

Greenheck 8 28th Annual AIAA/USU

Conference on Small Satellites
 1
10
Future work involves implementing signal processing
techniques more advanced than simple averaging, such
ADIS%16405% as using Kalman filtering. The board design could also
be improved by using digital sensors instead of analog,
0
10

ADIS%16485% which would reduce power consumption and reduce

−1
IMU%Cluster% board size by eliminating the need for the buffers and
ARW, deg/rt−hr

10
LN0200S% the ADCs. Implementing the algorithms on the FPGA
would also eliminate the need for an external processor,
−2
10 drastically reducing the mass, weight and volume of the
Desired'direc)on' MIMU% current design.
of'IMU'system'
performance''
−3
10
Acknowledgments

−4 SIRU% This work was made possible by NASA cooperative

10 −2 −1 0 1
10 10
mass, kg
10 10
agreement NNX13AQ79A with Marquette University
under subcontract to West Virginia University.

Figure 9: IMU Accuracy scaled by mass for the References

various types of IMUs
1. El-Sheimy, N., Haiying Hou, and Xiaoji Niu,
1
10 “Analysis and Modeling of Inertial Sensors
Using Allan Variance,” IEEE Transactions
ADIS%16405% on Instrumentation and Measurement, Vol. 57,
0
10
No. 1, 2008, pp. 140-149.
ADIS%16485%
IMU%Cluster% 2. Britting, K.R., Inertial Navigation Systems
−1
ARW, deg/rt−hr

10
LN0200S% Analysis, Wiley-Interscience, New York, 1971.

−2
3. Nawrat, A., Jędrasiak K., Daniec K., and Koteras
10
R., “Inertial Navigation Systems and its Practical
Desired'direc)on' MIMU%
of'IMU'system' Application,” found in Chapter 10 of New
performance''
−3
10 Approach of Indoor and Outdoor Localization
Systems, Fouzia Elbahhar and Atika Rivenq
−4
SIRU% (Ed.), InTech, Croatia, 2012.
10
4. Hartunian, R.A., Stewart, G.E., Curtiss, T.J.,
−1 0 1 2
10 10 10 10
power, W
Fergason, D., Seibold, R.W., and Shome, P.,
“Implications and Mitigation of Radio Frequency
Figure 10: IMU Accuracy scaled by power for the Blackout during Reentry of Reusable Launch
various types of IMUs Vehicles,” AIAA 2007-6633, AIAA Atmospheric
CONCLUSION Flight Mechanics Conference and Exhibit, Hilton
Head, SC, 2007
Our results show that performance gains resulting from
averaging of a cluster of MEMS sensors approaches the 5. Titterton, D.H., and Weston, J.L., Strapdown
predicted performance. The improvement in RMS noise Inertial Navigation Technology, 2nd Edition, IEE
for a cluster of 16 MEMS accelerometers was on Radar, Sonar, Navigation and Avionics Series,
average 3.2x, while the improvement in RMS noise for The Institution of Engineering and Technology,
a cluster of 16 MEMS gyros was 3.8x. Bias stability 2005.
saw less of a performance improvement, with an 6. http://www.analog.com/static/imported-
average improvement of 2.8x for the accelerometers files/data_sheets/ADXL335.pdf
and 3.0x for the gyros. This suggests that correlation
between bias stability errors across the individual 7. Crain, T.P., Bishop, R.H., and Brady, T,
sensors is higher than the correlation between errors “Shifting the Inertial Navigation Paradigm with
due to random noise. The y-axis of the gyros also MEMS Technology,” AAS 10-043, 33rd
showed less improvement in bias stability when Guidance and Control Conference,
compared to the other gyro axes. Further Breckenridge, CO, 2010.
experimentation will be needed to determine if this is a 8. “IEEE Standard Specification Format Guide and
result of the test setup, the cluster configuration, or Test Procedure for Coriolis Vibratory Gyros,"
because of the actual sensors themselves.

Greenheck 9 28th Annual AIAA/USU

Conference on Small Satellites
 IEEE Standard 1431-2004, IEEE Aerospace and
Electronic Systems Society, 2004, pp. 1-78.
9. “IEEE Standard Specification Format Guide and
Test Procedure for Linear, Single-Axis, Non-
Gyroscopic Accelerometers," IEEE Standard
1293-1998, 2011, pp. 1-249.
10. http://www51.honeywell.com/aero/common/docu
ments/myaerospacecatalog-
documents/MIMU.pdf
11. http://www.northropgrumman.com/Capabilities/S
IRU/Documents/ssiru.pdf
12. http://www.analog.com/static/imported-
files/data_sheets/ADIS16485.pdf
13. http://www.northropgrumman.com/Capabilities/
LN200FOG/Documents/ln200s.pdf
14. http://www.analog.com/static/imported-
files/data_sheets/ADIS16400_16405.pdf

Greenheck 10 28th Annual AIAA/USU

Conference on Small Satellites