Professional Documents
Culture Documents
Studies on Mechatronics
Supervised by:
Janosh Nikolic
Michael Bloesch
Author:
Felix Renaut
Declaration of Originality
Author(s)
Felix Renaut
Supervising lecturer
Janosch Nikolic
With the signature I declare that I have been informed regarding normal academic
citation rules and that I have read and understood the information on Citation
etiquette (http://www.ethz.ch/students/exams/plagiarism_s_en.pdf). The
citation conventions usual to the discipline in question here have been respected.
The above written work may be tested electronically for plagiarism.
Signature
1 Co-authored work: The signatures of all authors are required. Each signature attests to the
originality of the entire piece of written work in its final form.
Contents
Symbols
1 Introduction
2 Working principles
2.1 MEMS accelerometers . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1 Principle of a displacement-based MEMS accelerometer . . .
2.1.2 Capacitive-based displacement accelerometers . . . . . . . . .
2.1.3 Piezoelectric and piezoresistive-based displacement accelerometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.4 Resonance-based MEMS accelerometers . . . . . . . . . . . .
2.2 MEMS gyroscopes . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Principle of a gyroscope . . . . . . . . . . . . . . . . . . . . .
2.2.2 Tuning forks or resonant beam MEMS gyroscopes . . . . . .
2.2.3 Vibrating plate MEMS gyroscopes . . . . . . . . . . . . . . .
2.3 Inertial measurement units . . . . . . . . . . . . . . . . . . . . . . .
2.3.1 Definition and purpose of an IMU . . . . . . . . . . . . . . .
2.3.2 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
3
3
4
5
6
7
7
7
8
9
9
9
.
.
.
.
.
.
.
.
.
.
.
.
.
11
11
11
12
12
14
15
15
16
18
18
20
20
20
4 Conclusion
4.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
23
23
Bibliography
25
iii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Symbols
Symbols
roll, pitch and yaw angle
constant sensor bias
3-axis gyroscope measurement
, ,
m
Indices
x
y
z
x axis
y axis
z axis
Eidgen
ossische Technische Hochschule
Micro Electro-Mechanical System
Global Positioning System
Inertial Measurement Unit
Power Spectral Density
Angle Random Walk
Linear Quadratic Estimation
Defense Advanced Research Projects Agency
Chapter 1
Introduction
Inertial navigation is a technique in which the position and orientation of an object is
determined through the data given by accelerometers and gyroscopes. The results
obtained are therefore calculated only from the initial position, orientation and
movement of the object itself, and not compared to an external frame of reference,
as it is the case for other navigation systems such as GPS or radars.
The first inertial navigation systems were developed for the first long-range rockets
in World War II, since no external reference could be used to determine the position
of those objects in flight. However, the development of inertial navigation mostly
took place during the cold war through the american and the soviet space programs,
and the first space missions were only possible due to groundbreaking progress in
inertial sensor technology. Inertial sensors are today used in a wide range of objects,
from airplanes or the automotive industry to smartphones or running shoes.
Most of todays inertial sensors are micro electromechanical systems (MEMS). This
technology was first used for commercial purposes in the 1990s, and enabled new
applications through high miniaturization and cost reduction. Inertial sensors began
to be used in completely new domains, such as toys. However, this miniaturization
and cost reduction influences the performance of the accelerometers and gyroscopes,
which explains why some inertial sensors based on previous technologies are still
used for high-performance purposes.
In the following report, the physical working principles of recent MEMS inertial
sensors will be explained, and their performance studied. The aim of this studies on
mechatronics report is to have a better understanding of these widely used sensors
and of their applications.
Chapter 1. Introduction
Chapter 2
Working principles
In the following, different types of MEMS inertial sensors and the physical properties
they rely on are described.
2.1
2.1.1
MEMS accelerometers
Principle of a displacement-based MEMS accelerometer
(2.1)
where xr is the displacement of the proof mass with a known mass m relatively
to the object, i.e. if the object is taken as frame of reference, a the unknown
acceleration of the object in a desired frame of reference (e.g. the earth), k the
spring constant and d the damping constant of the mechanical system.
a(t) =
k
xr (t)
m
(2.2)
In many applications, the acceleration is the only desired parameter. However, for
navigation purposes, the position and velocity can also be determined through integration when knowing the initial velocity and position 1 . In almost all sensors, the
proof mass is only able to move in one direction, which means three different linear accelerometers have to be combined to measure a three-dimensional movement.
There are several ways of measuring the displacement of the proof mass. However, the most current ones are capacitors, piezoelectric crystals and piezoresistive
materials.
2.1.2
Capacitive-based accelerometers are using the fact that the capacitance of a given
plate capacitor is inversely proportional to the distance between the plates [2,
p. 182ff.] :
Cplate (d) =
0 r A
d
(2.3)
If this distance changes due to a movement of the proof mass, the capacitance
changes and the change in the electrical potential between the electrodes can be
measured. In order to augment the sensors sensitivity and to reduce the noise,
many capacitors can be combined linearly to create the output signal. Both in-plane
accelerometers, in which the proof mass moves in the plane of the device and outof-plane accelerometers, in which the proof mass can rotate perpendicularly to the
plane of the device on which the electrodes are put on, can use this physical property.
An example of in-plane and out-of-plane capacitive-based MEMS accelerometers is
shown in Fig. 2.1.2 .
1 given a constant orientation of the sensor, i.e. for a stable-platform IMU. In the case of
strap-down IMUs, the orientation of the object is also needed, since the direction of the measured
acceleration can vary in time. For more information, please refer to section 2.3.2.
-4#&'"4')&4)&'#-03',*"1*'%&)$2.)'"4'-'>E?6!'F>A
Navigation
Sensors
?>G
H'0&-I'1*-4#&'"4'1-0-1".-41
>AA'B?#3'".'")'4&1&))-%/'.+'%&)+2:&'7+."+4'+6'D'9'>A?>E'73'+%'-8+$.'EE;G'&2&
7-))'0&%'1-%%"&%'1/12&;'
?90@A"0
6->
3->
3-
6-
!"0
&15
+,-&(
./01#
+*=B#"
3-
6-
6->
6-;0<&9&*=0
8#9)&:0
3->
Figure 2.2: Typical structure of in-plane (left) and out-of-plane (right) capacitivebased MEMS accelerometers. On the left, the measured acceleration is parallel to
the plane of the device, whereas on the right, it is perpendicular to it (from [3]).
30"4#
,#150*'
6-7(&'8#
(! )$**+,-&)-! &.! (#$! /&%%$0/12*+3024$! 1-+5*2-$! 2//$*$0&%$($0'! 20$! 50&626*7! (#$! 8-2*&3!
?!2-4!89;<@>?A!"#$!*2(($0!%$2'B0$'!*2($02*!2//$*$02(1&-'!1-!()&!2C$'!)1(#!2!-&1'$!.*&&0!&.!=
Figure 2.3: Functioning principle of in-plane (left) and out-of-plane (right)
2/7!20&B-4!=?+>?!%3A!
capacitive-based MEMS accelerometers, from [4].
%$-(! 2//$*$0&%$($0'! /2-! 6$! &5$02($4! &5$-! &0! /*&'$4! *&&5A! G&*1607'! HI$.A! JKL! #2'! 0$5&0(
2.1.3 Piezoelectric and piezoresistive-based displacement ac.&0! (#$! IMK???! '$01$'!
2//$*$0&%$($0'! &5$02(1-3! &5$-! *&&5! 1-! (#$! &B(+&.+5*2-$! 2-4
celerometers
' 2! /2-(
(1&-'A! N&(#! 4$:1/$'!
20$! 2**+'1*1/&-! /&-'(0B/(1&-O! )1(#! (#$! &B(+&.+5*2-$! 4$:1/$! #2:1-3!
A piezoelectric material is a crystal which generates a voltage when mechanical
''! '2-4)1/#$4! 6$()$$-!
B55$0!
*&)$0!
$*$/(0&4$'A!
P$0.&0%2-/$!
QB&($4! 1'! R=@?! S3! 1-+0
stress is applied
to it.2-4!
On the
opposite,
applying an electric
field on the crystal
Figure
8:
MEMS
Pendulous
Accelerometer.
generates
a deformation.
This effect
is due to the physical disposition of the atoms
=+%3! (B0-+&-! 0$5$2(261*1(7O!
T??+55%! '/2*$+.2/(&0! 2//B02/7O! 2-4! 0$/(1.1/2(1&-! &.! U>! S
in piezoelectric crystals, in which a deformation of the crystal separates the electric
charges 855*1/2(1&-+M5$/1.1/!
from one another. Commonly used
piezoelectricG10/B1(!
materials W8MVGX!
are quartz and
B0$! '$-'&0! 2-4! 4131(2*!
V-($302($4!
1'! 1-($302($4! )1(#! (#$
(# zirconate titanate, a ceramic perovskite crystal). For a perfectly uniform
PZT (lead
(&! .B0-1'#! &-+6&204!
T ! +&04$0! ($%5$02(B0$! /&%5$-'2(1&-A! G&*1607'! #2'! 2*'&! '#&)-! HI$.A! T?
piezoelectric crystal and a given direction of the force applied to the crystal, the
(#
*")'./0&'+6'-11&2&%+7&.&%'")'K+%.*%+0'L%$77-4M)'N"J13'+6',*"1*'+:&%'
charge difference
hence(&!
the 2!
output
voltage is *1-$201(7!
linear to this force.
*(2! > +&04$0! 0$3B*2(1&-!
*&&5!and
*$24'!
402%2(1/!
1%50&:$%$-(! 2-4! /&-'$QB$-(*7
:&%)"+4)' *-:&' 8&&4' 5&:&2+0&5' F.-1."1-2'
#%-5&' -45' "4&%."-2' #%-5&H' -45' *-:&
X
Y1602(1&-!I$/(1.1/2(1&-!Z00&0!WYIZXA!
Qi =
dij Fj
(2.4)
?W2X!'#&)'!2!'/#$%2(1/!&.!2!(B--$*1-3!2//$*$0&%$($0A!"#$!/&-(0&*!$*$/(0&4$!$*$/(0&'(2(1/2**7!
$:$0! 1-(&! (#$! (B--$*1-3! 5&'1(1&-! W]=! S%! 2-4! R@?! YXA! 8! '$0:&! %$/#2-1'%! #&*4'! /&-'(2-(!
(#$!(B--$*1-3!(15!W\13B0$!=?W6XX!2-4!(#$!/2-(1*$:$0O!2-4!#$-/$!#&*4'!/&-'(2-(!(#$!(B--$*1-3!/B0
&'()*+,'(#*8('1+5/0&'(6*7*)#+(66*'*8!7*#*8'"4',*"1*'0%++6'7-))'5")02-1&7
&B(5B(!'13-2*!1'!(#$!/#2-3$!1-!:&*(23$!2(!(#$!$*$/(0&4$!B-4$0!2//$*$02(1&-A!"#$'$!4$:1/$'!20$!4
-0-1".-41&'-1%+))'.*&'1+78'6"4#&%);'O*")'-11&2&%+7&.&%'")'7$1*'7+%&')&4)"
$! 2//$*$02(1&-'! 1-! (#$! -2-&+3! 02-3$! 2-4! 0$QB10$! *&)+0$'&-2-(! .0$QB$-/7! 50&&.! %2''$'! 2
"#*.'
F%-.*&%' .*-4' .+0?.+?8+..+7H' 5"%&1."+4;' O*&' 1+78"4-."+4' +6' U?-9")' -45'
!0$'&*B(1&-!0$24&B('A!I$/$-(!%1/0&.2601/2($4!(B--$*1-3!2//$*$0&%$($0'!#2:$!0$'&*:$4!@?!-3D
0."7"U&5'
)/).&7' :+2$7&3' )"41&' .*%&&' -9&)' +6' -11&2&%-."+4' 7&-)$%&7&4.' 1
A>!,EF!HI$.A!T=L!)1(#!2!/*&'$4+*&&5!47-2%1/!02-3$!&.!&:$0!K?!4NA!E&)$:$0O!%2C1%B%!2//$
%$-(!/25261*1(7!1'!:$07!*&)!WR=!%3X!)1(#&B(!.B0(#$0!*&&5!%&41.1/2(1&-A!8!(B--$*1-3!2//$*$0&%
*"0);'
$.A!T@L!#2'!$C#161($4!3&&4!*&)!.0$QB$-/7!0$'&*B(1&-!&.!=>!-3DEF!&:$0!=!(&!=??!EFA!
>&,C KE,).4/3)'-H42<<2(1L
F :*-+8'-2,)'104'/4.2/()'A2.4A04,..'1'*-,8412)&4'-4MN
Chapter 2.
'
&
EI
N
UF
wx
Working principles
wx '
wt %
% L ! wx wx % 6
7'2)
!"#$$#!%$!
U = R(F ) I
&'()*+,+-./+01.(23*4*5,000000$6$7%8!$7%%
(2.5)
Piezoelectric and piezoresistive-based accelerometers usually have a higher robustness and range than capacitive-based accelerometers but are less precise and are
therefore often used in high-shock applications.
2.1.4
9:)5.7;/<4'-=>.0?@@.(-+
P,)H24Q&38'1B.242<<2(1/
Figure 2.4: Amplitude as a function of frequency for a general vibration movement.
1 is the resonance frequency, the excitation frequency (from [4]).
This physical property can be of great use for sensors : by exciting a mechanical
&'()*+,+-./+01.(23*4*5,000000$6$7%8!$7%%
structure at its resonant frequency !"#$$#!%$!
through an actuator,
a small change in the
frequency, for example due to an additional force on the structure, results in a
great change in amplitude of the vibrating structure. If this amplitude change can
be measured accurately, a sensor with high sensitivity can be obtained.
Likewise to displacement-based accelerometers, the change of frequency can be measured capacitively or with the use of a piezoelectric crystal. The actuator, which
enables the proof-mass to vibrate, usually uses the same physical property than
the sensor used to measure this change of frequency, e.g. a piezoelectric crystal or
a capacitor powered by alternative current with a certain frequency. Resonancebased accelerometers usually are more precise, but more cost-effective than simple
displacement-based accelerometers due to the additional presence of an actuator.
2.2
2.2.1
MEMS gyroscopes
Principle of a gyroscope
7'2)
trajectory. The apparent force on the object is called Coriolis force, and is proportional to the rotation speed of the frame of reference. Its direction is perpendicular
to the rotation direction of the reference frame and to the velocity of the moving
object.
F Coriolis = 2mobject (
v )
(2.6)
By measuring the effect of this force on a proof mass which is moving at a known
velocity, the rotational speed of the frame of reference, i.e. the one of the object
the sensor is placed on, can be determined.
2.2.2
Since the proof mass has to be in constant motion for the Coriolis effect to appear,
MEMS gyroscopes are mostly based on resonant structures. Those structures have
to be put in motion using an actuator, which can be for instance a piezoelectric
crystal put under a varying electric tension or two electrodes which are alternatively
switched on and off. Usually, the physical principle used for the actuator is the same
as the one used to measure a change in frequency of the proof-mass.
A quite simple example of a resonant structure used for MEMS gyroscopes is what
is called a tuning fork, which is lately nothing more than the combination of two
beams of same length and material on a common shaft, which is excited at their
resonance frequency, resulting in a type of balanced oscillator where the two beams
oscillate 180 out-of-phase. Compared to a single beam, this structure is more energy efficient and more accurate. When a rotation of the system occurs in a direction
perpendicular to the vibration direction of the beams, the resulting Coriolis force
changes the frequency of the vibration, which can be detected using a piezoelectric
or piezoresistive material on the common shaft. An example of a MEMS tuning
fork gyroscope is Systron Donners quartz gyroscope shown in Fig. 2.6 which was
very successful in the 1990s as a low-cost yaw-rate sensor [3] .
Figure 2.5: Working principle of a tuning fork MEMS gyroscope, without rotation
(left) and when a rotation of the object occurs (right).
!"#$$#!%$!
! structure comFigure 2.6: Systron Donner QRS and its characteristic H-formed
posed of two tuning fork-shaped piezoelectric quartz elements. It measures a change
Figure 13: Systron Donner Quartz Rate Sensor (QRS)
of orientation normal
to the plane [3] .
( BEI Systron Donner Inertial Division, printed with permission).
"#$%&'(!)*+%&,*#-!.#,$-%&,$/!01%2!3).04!5&(!&!(6$7*8*7!*#(9:&,*#-!(1(,$;!82%!,5$!<*+%&,*#-!$:$;$#,(!3(*;*:&%!
,2!,5$*%!=.)>!&77$:$%2;$,$%4!,2!%$/97$!$#$%-1!:2(($(!29,!28!,5$!(,%97,9%$?!"9,69,!#2*($!*(!@?@A!3/$-B(4BCD!*#!
2.2.3 Vibrating plate MEMS gyroscopes
12'(90:*4,+'4'(*30;,)*+(*<.0=>*+(2?
&!A@@!CD!EF?!G&-$;'(!H9&6&(2#!-1%2!5&(!829%!I9&%,D!,*#$(!$J,$#/*#-!96K&%/!8%2;!&!72;;2#!+&($L!,5$%$+1!
%$/97*#-!9#K&#,$/!7%2((M7296:*#-!8%2;!/%*<$!,2!($#($!NO$8?!P@Q?!!
Another structure used for MEMS gyroscopes is a vibrating plate, which can be
in round or rectangular shape. The plate is suspended by spring structures such
as folded
beams.
actuator makes this structure vibrate at a specific frequency
2.4.2.2 Vibrating
Plate
MEMSAn
Gyros
and phase. This vibration can be in an in-plane direction (x-axis), in which case
=%&6$%!R&+2%&,2%1'(!S9#*#-!T2%U!01%2MV!3ST0MV4!(52K#!*#!T*-9%$!AW!72#(*(,(!28!,K2!(*:*72#!6%228!;&((!6:&,$(!
a rotation normal to the plane (in the z-axis)
causes the device to vibrate in the
=,>+34126,<6?+#6@+./A6BC+(4+96D,E4+36F,1*G6H(E7
(9(6$#/$/!2<$%!&!-:&((!(9+(,%&,$!+1!82:/$/!+$&;(!&#/!<*+%&,*#-!*#M6:&#$!AX@!/$-!29,!28!65&($?!S5*(!/$(*-#!*(!
other in-plane direction (y-axis), inducing for example a change of capacitance as
&:(2!%$8$%%$/!,2!&(!&!/29+:$M$#/$/!ST0?!=*;$#(*2#(!&%$!2#!,5$!2%/$%!28!Y@@!J!W@@!Z;?!S5$($!&%$!&:(2!*#M6:&#$!
described in section 2.1.2. Fig. 2.7 and 2.8 show examples of in plane vibrating
-1%2(?! S5$! 29,M28M6:&#$!
gyroscopes ($#($!
[1, 4] .;2,*2#! *#/97$/! +1! ,5$! [2%*2:*(! 82%7$! *(! /$,$7,$/! +1! 75&#-$(! *#! 7&6&7*,$!
A+
Fs = 2m v !:
There are several other structures possible, such as single vibrating beams or viO2#$%!2)4$-!./!0+1%&2+(3!45&2$!<I<@!3)%.-!2#&2!,.!(.2!%$5)!.(!2#$!6.'1!,%+0$!#&0$!1$$(!,$0$5.4$,;!"#$-$!
disks
which
have different
properties
due,%+0$;!
to their
mass distribution and the
,$0+6$-! ,.! (.2! %$P=+%$! -=6#!brating
(&%%.?! 3&4-!
&(,!
2+3#2! /&1%+6&2+.(!
2.5$%&(6$-!
&-! 2#$! 6.'1!
Q$2! R%.4=5-+.(!
S&1T-! UQRSV! <I<@! 3)%.! WN$/;!
?#+6#! &!of
XALOZ!
%$-.(&2+(3!
65.0$%! -#&4$9!
-=-4$(,$,!
/.=%! principle is the
wayDXY9!
the+(!change
frequency
is [A5$&/!
measured.
However,
the 1)!
working
2
-4%+(3-!&(,!6.(2&+(+(3!&!0$%2+6&5!4.-2!4%.0+,+(3!2#$!'&+(!+($%2+&5!'&--9!+-!,%+0$(!+(!&!%.6*+(3!'.2+.(!&1.=2!&(!
same,
which
is
why
it
wont
be
described
in
detail
here
.
2 -45&($!
16 ./! 2#$! 65.0$%5$&/;! "#+-! +-! &(! .=2A./A45&($! 3)%.! U+;$;9! +2! %$-4.(,-! 2.! %&2$! &1.=2! 2#$! >A&B+-!
RTO-EN-SET-116(2011)
&B+-! +(! 2#$!
4$%4$(,+6=5&%! 2.! 2#$! 45&($! ./! 2#$! 65.0$%! 5$&/V;! K! ,.=15$! 3+'1&5! -2%=62=%$! ,%+0$(! &(,! -$(-$,!
$5$62%.'&3($2+6&55)! #&-! 1$$(! ,$0$5.4$,! +(! Q&4&(! WN$/;! DJY;! @$(-.(.%! "$6#(.5.3+$-! U\.%?&)V! +-! ,$0$5.4+(3!
2 for more details on e.g. the principle of a resonant ring gyro, please refer to [1, p. 202ff]
2#$! @KNDEE! (.0$59! #+3#A4%$6+-+.(!
<I<@! :=22$%/5)! 3)%.9! ,$-+3($,! 2.! &6#+$0$! &(! KN]! ./! E;EEX! ,$3^#9!!
+(A%=(!1+&-!-2&1+5+2)!./!E;E[!,$3^#9!&(,!1+&-!%$4$&2&1+5+2)!./!E;F!,$3^#!WN$/;!D[Y;!"#$!%&2$!-$(-+2+0$!&B+-!./!2#$!
@KNDEE! 5+$-! +(A45&($! ?+2#! =44$%! &(,! 5.?$%! -$(-$! 45&2$-! 2#&29! &-+,$! /%.'! +(6%$&-+(3! -$(-+2+0+2)9! &55.?! /.%!
2=(+(3!./!2#$!-$(-$!&(,!,%+0$!%$-.(&(6$-!&(,!&62+0$!6.'4$(-&2+.(!./!2#$!P=&,%&2=%$!.//-$2;!
>*+(20;,)*+(*<.7B.(*3C01,<.
2.4.2.3
N$-.(&(2!%+(3!<I<@!3)%.-6.4$-!#&0$!&(!&,0&(2&3$!+(!2#&2!2#$!%+(3!-2%=62=%$!'&+(2&+(-!2#$!,%+0$!&(,!-$(-$!
0+1%&2+.(&5!$($%3)!&55!+(!.($!45&($;!_.?$0$%9!2#$%$!+-!&5-.!&!,+-&,0&(2&3$!+(!2#&2!2#$!%+(3!#&-!&!5.?!0+1%&2+(3!
'&--! &(,! #$(6$! 5.?$%! -6&5$! /&62.%;! Z+3=%$! FDU&V! -#.?-! &! -+(35$A6%)-2&5! -+5+6.(! 0+1%&2+(3! %+(3! 3)%.! /%.'! 7;!
2.3
2.3.1
An inertial measurement unit (IMU) is a combination of different inertial sensors,usually three linear accelerometers and three gyroscopes, which provides a 3
dimensional measure of the systems orientation and motion. Such units are used
for inertial navigation purposes, e.g. in rockets or satellites, to determine the position of an object. They are sometimes combined with positioning systems (GPS)
to have more precise information.
IMUs based on MEMS sensors are strap-down systems, which means the sensors
orientation depends of the orientation of the object it is on. Theoretically, all types
of previously shown MEMS inertial sensors can be combined in an IMU. However,
to win space, most MEMS IMUs designs combine two in-plane accelerometers and
gyroscopes (x- and y-axis) with an out-of-plane accelerometer and gyroscope (zaxis), which permits to place all sensors on a single chip.
2.3.2
Algorithms
Figure 3:
3: Stable
Stable platform
platform inertial
inertial navigation
navigation algorithm.
algorithm.
Figure
Figure 2.9: Model of stable platform inertial navigation algorithm, from [5].
Figure 4:
4: Strapdown
Strapdown inertial
inertial navigation
navigation algorithm.
algorithm.
Figure
Figure 2.10: Model of strap-down inertial navigation algorithm, from [5].
66
10
The implementation of the gyroscopes output signal into the algorithm results in
adding the gyroscope error to the one of the accelerometer. This specific problem of
strap-down inertial systems may decrease the accuracy of the inertial measurement
unit significantly compared to a similar stable platform system, since the largest
errors in MEMS inertial technology are usually originated by the gyroscope, as
shown in section 3.2.3.
Chapter 3
The total bias of a sensor is usually defined as the average output signal that has
no correlation with the input signal, i. e. the acceleration or rotation, for specific
operation conditions and a specific time [6]. It is generally expressed in degree
per hour (deg/h) for a gyroscope and in meter per second square (m/s2 ) or g
(1g 9.80665m/s2 ) for an accelerometer. Additionally to the bias comes noise,
which can change the output value and add an additional error to the sensor.
3.1.1
The bias offset of a sensor is defined as the value of the output signal when the
input signal is zero. For an accelerometer, it would therefore be the acceleration
given by the sensor when it is not actually moving; for a gyroscope, the angular
rotation given when the sensor is not undergoing any rotation.
Basically, a change in any physical property such as pressure, temperature or height
can induce such a bias. Nevertheless, for MEMS inertial sensors, temperature variation, wether it is due to the environment or to the heating of the sensor itself, is
the main cause of bias. Because of the many physical properties depending on the
temperature, the bias caused by temperature fluctuation is almost always nonlinear
to the temperature change itself, which means it is difficult to correct. However,
many IMUs 1 contain temperature sensors which are used to correct those bias
directly [5].
The problem of inertial sensors is that the input signals are time integrated, as
can be seen in Figure 2.10 : a constant bias of therefore causes an error which
grows linearly with time for a gyroscope, i.e. f (t) = t and quadratically for an accelerometer, i.e. xf (t) = 21 t2 [5] . This means for a 100g bias in an accelerometer,
the error in position results in 0.05m after 10 seconds and 500 m after 1000 seconds
2
. The bias of a gyroscope creates an even more important position error, since
it creates second order errors in velocity and third order errors in position when
combined with the accelerometer. This means that for e.g. a 0.2deg/h gyroscope
bias, the final error in position is 0.0016m after 10 seconds, and grows to 1600m
after 1000 seconds [6] . The error due to the accumulation of small bias over time
is called bias drift.
1 e.g.
the Xsens Mti or the ADIS 16488 from Analog Devices (see table 3.4).
for 1000s : xf (1000) = 1/2 100 10 106 10002 = 0.05m where g = 10m/s2 is
2 calculation
assumed
11
3.1.2
12
Noise
Noise is an additional signal that interferes with the output signal of the sensor.
It may come from other sensors or from the sensor itself, but it is present in any
sensor and difficult to characterize. Since it is not systematic, noise can not be
removed from the data directly and a stochastic modeling of the noise is necessary
beforehand. Integrating a white noise with a constant standard deviation 0 one
time, i.e. as it is the case for the output of the gyroscope or for the velocity, gives
a zero-mean random walk with a standard deviation that grows proportionally to
the square root of time, and integrating it twice, as it is the case for the position
error resulting of the accelerometer noise, gives a second-order zero-mean random
walk with a standard deviation linear to t3/2 [5] .
Fig. 3.1 shows a measurement made on 1000 identical gyroscopes by the company
crossbow for the same noisy signal. Those sensors have an angle random walk
(ARW) 3 of 0.99 deg/ sec. The resulting distribution, after 1000 seconds, should
therefore have a standard deviation of 31.5 degrees, which
appears
to be Walk
the case
Angle
Random
here.
300
250
200
150
100
50
0
-150
-100
-50
50
100
150
Endpoint (deg)
Figure 3.1: Random walk of 1000 identical gyroscopes as a function of time. On the
right, the final distribution, i.e. after 1000s, of the output values of the gyroscopes
We can clearly see the same characteristics of the random walk that we saw on the football field. Figure 3
is given
(reprinted from [7]).
on the left shows how the distribution widens with time, and on the right how the distribution falls into a
bell curve.
3.1.3
modeling
We can Stochastic
make a quantitative statement
about the distribution of the end points for this sensor.
1/2
The sensor
has an ARW of 0.99 deg/sec . So after 1 second, the standard deviation of the distribution will be about 1
degree; after
seconds,
about 9.9 of
degrees;
and after
1000therefore
seconds, about
31.5 degrees.
will be the its
To study
the 100
noise
influence
a sensor
and
being
able toThis
categorize
same whether the sensor is moving, or sitting still. Without some other angle reference, this will be a
performances,
stochastic modeling techniques have to be used. In this section, the
fundamental uncertainty in the result of the angle calculation.
two most important stochastic modeling techniques and their purpose are briefly
Notes: For
Converting
Angle Random
Walkstochastic
and PSD/FFT
Noise Values
depicted.
more information
about
modeling
techniques applied to
inertial sensors, please refer to [8].
Different manufacturers will quote noise specifications in different ways. Some will quote an angle
random walk (ARW); some will quote a PSD or FFT noise density; some will quote a total noise, one or
three sigma variation in the output of the sensor. Below are some methods to convert between the various
PSD
specifications. See IEEE Std. 952-1997 C.1.1 for more complete discussion of this.
ThePSD
power
spectral density (PSD) is the most common tool used to analyze data,
? ARW
especially periodic signals. It gives
the periodicity of an output signal for different
&, ) 2
#
1
frequencies.
The
two-sided
PSD
S()
ARW / hr =
PSD$* ' Hz ! corresponds to the Fourrier transform of the
60
hr
!
covariance K( ) [8] :
%$+ (
Z" +
S() =
K( )ej d
(3.1)
FFT ? ARW
1
&, )
#
(
)
FFT $* ' Hz !
the angle random60walk is+ahrcurrently
given value for gyroscopes It is simply defined as the
(
ARW / hr =
"
standard deviation of the gyroscope output for a white noise input, i.e. a first order random walk
afters,1BW
second
and corresponds to the proportionality factor between this standard deviation and
? ARW
the square root of time
ARW / hr =
1 , )
1
!* ',
60 + hr ( BW ( Hz )
where s is the standard deviation of the signal, and BW is the effective bandwidth of the sensor in Hz. This
creates an estimate of the noise density that is then used to estimate the angle random walk.
Page 3
expressions of noise are shown together with Allan variance expressions in Chapter Four.
The typical characteristic slopes are shown in Figure (3.9), where the actual units and
13
3.1. MEMS inertial sensors error characteristics
frequency range are hypothetical. With real data, gradual transitions would exist between
different
PSD slopes
(IEEE
the input,
sharp transitions
in Figure
Athe
useful
property
of the
PSDStd1293-1998),
is that for a rather
whitethan
noise
the output
power
spectral density can give directly the transfer function of the system. The PSD of
(3.9); and the slopes might be different than 2, -1, 0, and +2 values in Figure (3.9). A
a random process usually gives linear log-log slopes corresponding to the powers of
frequencies.
This means the different random processes such as noise or bias drift
certain amount of noise or hash would exist in the plot curve due to the uncertainty of the
are represented through straight lines, wich have different slopes [6] [8].
measured PSD.
Figure
3.9 Slopes
Hypothetical
Gyro in Single-sided
PSD
(after
IEEE processes
Std952-1997)
Figure
3.2:
and appearance
frequencies
of Form
various
random
for a
hypothetical gyro in single-sided PSD form, from [8].
3.4.2.3 TEST RESULTS
These characteristics can be used to analyze random processes, for instance the bias
The same data sets used in section 3.3.2 are used here for power spectral density analysis.
instability or the angle random walk of a gyroscope. The resulting PSD obtained as
aApplying
function the
of the
frequency
is usuallypreviously,
averaged using
frequency
techniques,
PSD
method described
the PSD
result onaveraging
log-log plot
is shown
which enable a clear visible slope. An example of such an analysisis shown in
Fig.
3.3, which
shows
e.g. anXangle
random
of about
0.0015deg
in Figure
(3.10)
for CIMU
-axis gyro
data.walk
Because
of the
bunching/ ofh [6].
the high
frequency data points in the log-log plot, it is difficult to identify noise terms and obtain
parameters in such conditions. Hence, the frequency averaging technique (IEEE Std
66
Figure
CIMU
X-Gyro
Results gyro
with Frequency
Averaging
Technique
Figure
3.3:3.11
PSD
output
for a PSD
real MEMS
after frequency
averaging
techniques,
from [6].
14
Allan Variance
Allan Variance is a procedure first defined to analyze the rendering of high precision
oscillators [9]. It has later on been applied more generally for the study of oscillator
stability, and can be used quite efficiently to determine the stability and random
2
drift characterization of MEMS inertial sensors [10]. The allan variance
( )
describes the frequency stability of a random process, i.e. the variance of the change
of frequency value between two observations done over a certain sample period . It
can be calculated from the PSD estimation with the formula given in equation 3.2,
but there is no inverse formula [8].
( ) = 4
S (f )
0
sin4 (f )
df
(f )2
(3.2)
"###!$-./0.10!$2#3"4"3.-"5/!4516.-!78"0#!./0!-#$-!2153#081#
GB+
"" !$!"%&#$
'
()*
++
,
GBG+
/#
() (
(
.$
1& 2
GBGG+
(
()(
0#
- # ()((.$ #
(.$
GBGGG+
GBG+
GB+
+G
+GG
+GGG
!!"#$
345678!-)*9:;<=8>4?8!78@78?8AB<B4CA!CD!#E@CB#8B4=<;!57EC!4A!F;;<A!G<74<A=8!DC7H
Figure 3.4: Slopes of various random processes for a hypothetical gyro in an Allan
variance plot from [8].
+GBG
2
K%& D' (&$ &#' & IJ)
20 2
*
3.1.4
+BG
Data filtering
2
2*2
'2
L
-2
GB+filter4 is a mathematical algorithm
The Kalman
providing an optimized output for
*
a noisy input signal by using a predictive model. The predictive control system is
2
a copy of the
studied system, which estimates how the answer
should be without
2/
GBG+
the noise or error created by the actual sensor. Using aiding sensors which give
additional information to the system, it provides a more observable system than
the original.
GBGG+ By averaging the output value of the real control system and the
predictive GBGGGGGG+
control system,
an optimalGBGGGG+
signal can beGBGGG+
obtained [6,GBGG+
p. 43ff.].
GBGGGGG+
GBG+
D!"IJ$
4 also
345678!-)M9:48=8>4?8!78@78?8AB<B4CA!CD!#E@CB#8B4=<;!5E7C!4A!?4A5;8N?4O8O!:PQ!DC7H
!! !
velocity vector is then time integrated, yielding the position vector. These steps are
continuously iterated throughout the navigation process (Verplaetse 1995). Figure 1.1
shows this concept in a schematic form. This procedure is, usually, considered as IMU
mechanization. The mechanization
results will
be classic
fed into inertial
the Kalman
filterperformances
to correct
3.2. Comparison
with
sensor
15
1.1 Inertial
Navigation
Schematic
Plot (after
El-Sheimy
2003)
Figure 3.5: Figure
Schematic
plot of
a, inertial
navigation
algorithm
for an
IMU using a
Kalman filter, from [6].
3
The aiding sensors used for the Kalman filter can be temperature and pressure
sensors, but also magnetometers or vision-based navigation algorithm, which use a
change of the environment to determine their position. Another often used solution
is the combination of IMUs with a geostationary positioning system (GPS) to obtain
more precise information on the position and orientation of the studied system [11].
3.2
3.2.1
section 3.1
16
Bandwidth
The bandwith of a sensor is a signal processing term which refers to the range of
the frequencies that can be measured (usually in Hz or rad/s).
Sensitivity
The sensitivity of a system is generally defined as the ratio between the output signal
and the input signal. The higher the sensitivity, the easier a change in the input
signal (which can result from a change in acceleration or orientation) is measured.
However, increasing the sensitivity of a sensor usually decreases its accuracy, since
the errors are increased as well.
Selectivity
The selectivity of a sensor is its capacity to differentiate the wanted signal from
other present signal. It can be defined as the ratio of the sensor sensitivity for two
different signals.
Nonlinearity
The nonlinearity of a sensor is the actual maximal variation of a constant sensitivity
over time for the total range or scale.
Bias stability/instability
Bias stability can be defined as the potential of the sensor error to stay within a
certain range for a certain time. It is an essential value for inertial sensors, since the
integration steps necessary in the signal processing of these sensors can lead to high
bias instability. Likewise to rate noise density, bias stability/instability is modeled
using stochastic modeling techniques which are described in section 3.1.3.
Repeatability
Repeatability is the potential of the sensor to respond in an identical way, i.e. with
the same output, to a same input signal given the same conditions.
Grades
Inertial sensors are sometimes graded depending on their performances. Those
grades are corresponding to the possible application domain corresponding to the
performance of the sensor. Table 3.1 shows the grades corresponding to different
in-run bias of inertial sensors.
Application Grade
Gyroscope
Accelerometer
Commercial
> 1deg/s
> 50mg
Tactical
1deg/h
1mg
Navigation
0.01deg/h
25g
Strategic
0.001deg/h
1g
3.2.2
Accelerometers
One of the main arguments of MEMS accelerometers is the different range there
are able to calculate an acceleration for. This enables their use in many different
application domains, as shown in Fig. 3.6.
17
Principle
Prize (USD)
Output signal type
Measured Directions
Size
Weight (grams)
Range (g)
Bandwith
Sensitivity
Sensitivity
change
due to Temperature
(%/ C)
Nonlinearity (% of
Full-scale-output)
Shock survival limit
Offset at 0g
Noise Density
before
filtering (g/ Hz)
R
Kistler miniature Piezobeam
8640A
Piezoelectric ceramic beam creates voltage when vibrating due
to acceleration
not given (low cost)
Analog
1
10.5mm3
3.5
3 versions available : 5, 10,
50
0.5 to 3000 Hz for 5 and 10g
range, 0.5 to 5000 Hz for 50g
1.000 V/g for 5g range, 0.500
V/g for 10 g range, 0.100 V/g for
50g range
0.12 for 5g version, 0,16 for
10g and 50g
0.3
7 000g
none (no operating current)
10 000g
1.5V on all 3 axis
150106 for x-, y-axis, 300106
for z-axis
not given
Table 3.2: Comparison of values given by the constructor for 2 different MEMS
Accelerometers (data taken directly from the datasheets of the constructors)
3.2.3
18
Gyroscopes
Even though the precision of MEMS gyroscopes has been improved in the last
years with the help of batch production techniques and new gyroscope designs,
they remain the biggest problem of MEMS inertial sensing technology and the main
reason why previous technologies such as fiber-optic gyroscopes (FOG) or ring laser
gyroscopes (RLG) are still vastly used in domains where high-precision sensors are
necessary. Todays best MEMS gyroscopes attain a bias stability in the range of
1 /h, which corresponds to a tactical grade.
Analog
Devices
ADXRS453
Technology
1 MEMS gyro
Prize (USD)
70
ST Microelectronics A3G4250D
3 MEMS gyros, 1
Temperature sensor
not given (< 250)
Size
350mm3
4 x 4 x 1.1mm
Range ( /sec)
Sensitivity
( /sec/LSB)
245
0.00875
0.05 %Full
and range
Nonlinearity
Temperature Sensitivity
Acceleration Sensitivity ( /sec/g)
Rate
Noise
Density( /sec/ Hz)
Bias Stability /h
scale
KVH DSP-3400
1 Fiber Optic Gyroscope (FOG)
not given (> 2000)
88.9 x 58.42 x 42.54
mm
375
not given
0.15%Full-scale and
range
<6 /h for a change
of max. 1 C/min
not given
0.03 /s/ C
0.01
not given
not given
0.015 at 25 C
0.03 at 25 C
0.0667
16
not given
Table 3.3: Comparison of values given by the constructor for 3 different gyroscopes
: 2 based on MEMS technology and 1 FOG (data taken directly from the datasheets
of the constructors)
3.2.4
MEMS IMUs
Due to the performances of the MEMS inertial sensors, and particularly to the
integration of the gyroscope errors, MEMS IMUs are globally still less performant
than previous IMUs. However, some recent inertial measurement units based on
MEMS technology can be used for navigation. Their prize is almost similar to some
IMUs based on previous technologies, but their reduced size and weight make new
application domains for navigation possible. The price range of MEMS IMUs, which
go from about 4$ for e.g. the LSM330DLC of ST Microelectronics, to more than 10
000$ for some high-precision IMUs, is also what enables their different applications.
In the following table, the main data of 3 MEMS IMUs of different constructors
and with different application domains are listed.
19
Prize (USD)
Size
Weight (grams)
Number of Sensors (A
: accelerometer, G: gyros, T : temperature,
M: magnetometer, P:
pressure)
Accelerometer
(g)
Range
Gyroscope
( /sec)
Range
Accelerometer
tivity (g/LSB)
Sensi-
Gyroscope Sensitivity
( /sec/LSB)
Accelerometer Noise
Density
before filtering
(g/ Hz)
Gyroscope Noise Density before
filtering
( /sec/ Hz)
Gyroscope Bias Stability
Signal processing
Applications
Invensense
MPU
6000
15
4 4 0.9mm
not given
3A, 3G, 1T
2, 4, 8, 16
selectable
250,
500,
1000,
2000
selectable
from 6.10 105
for 2g range to
4.88 104 for 16g
range
from 0.00763 for
250 /sec to 0.061
for 2000 /sec
4104 at 10Hz
2, 4,
lectable
8 se-
Analog
Devices
ADIS 16488
1700
47 44 14 mm
not given
3A, 3G, 1T, 3M, 1P
18
250,500,2000
selectable
1.221108
3.052 107
6.7105 at 25Hz
0.005 at 10Hz
0.03 at 10Hz
0.0066 at 25Hz
not given
11 /h average at
25 C
6.25 /h
Kalman filtering
Kalman filtering
Motion
Control,
Immersive
Simmulations,
Image
Stabilization,
Pedestrian
Navigation, Toys
Robotics,
Motion
capture,
Positioning
and
stabilization, Personnel navigation
and
tracking,
Unmanned
vehicle
navigation,
Healthcare
monitoring,
Motion
control, Immersive
simulations
Platform stabilization,
Navigation,
Personnel
tracking,
Instrument,
Robotics
Table 3.4: Comparison of values given by the constructor for 3 different MEMS
IMUs (data taken directly from the datasheets of the constructors)
3.3
3.3.1
20
Future developments
Multi-directional MEMS inertial sensors
3.3.2
The never-ending miniaturization of MEMS inertial sensors has opened the possibility for new applications. One often given example is smart ammunition : a
bullet equipped with a MEMS inertial IMU could calculate its position, compare it
to the one of its target, and redirect itself using e.g. small fins on the back of the
bullet. Such possibilities are currently under research in several companies for the
weapon industry6 , the most important problem being the shock resistance of the
6 e.g. at Honeywell, which specializes in producing high-precision inertial sensors for military
and aeronautical use
21
IMU necessary for its survival to the firing of the bullet. Another example of new
possible applications is microrobotics. Robots small enough to be implemented in
someones body have been build7 , and if adding an IMU to these robots to enable
precise navigation is not imaginable today, it might be in the upcoming years.
7 e.g.
22
Chapter 4
Conclusion
4.1
Conclusion
MEMS Inertial sensors have been a real breakthrough compared to previous inertial
sensing technologies in matters of size and price. Although they function with
relatively simple physical principles, they can enable measurements of acceleration
and angular rotation quite precisely. The miniaturization and low cost of inertial
sensors created by the MEMS technology has opened new possibilities and industrial
applications, such as their utilization in toys and smartphones, in small robots or
motion capture. New applications are envisaged in the future in different fields,
from micro-sized robots to smart ammunition.
However, there still are several drawbacks in MEMS inertial sensor performance,
especially for MEMS gyroscopes : Their low selectivity, for instance their high
sensitivity to temperature or pressure, combined with the integration necessary for
a strap-down technology, can lead to high bias drift and create important errors
in position after a certain time. Furthermore, the precision of micromachining
processes is limited, which can induce flaws in the basic structure of the sensor. All
these errors lead to the point that todays MEMS IMUs still are not comparable
in matters of performance to previous technologies, especially due to the MEMS
gyroscopes.
Nevertheless, one should not forget that MEMS inertial sensing technology is a
fairly recent technology, and that increasing the performance and design of MEMS
IMUs remains a very active development area. Another important point is that
the size and price of such an IMU enables their combination with other position
tracking devices, such as GPS, magnetometers or vision-based tracking, which significantly reduce the errors of a inertial measurement navigation system and make
high-precision navigation based on MEMS sensors a reality.
4.2
Acknowledgements
I would like to express my very great appreciation to the tutors of this project,
Janosch Nikolic and Michael Bloesch, who answered all my questions and showed
great support during all the semester. I also would like to thank the whole ASL
lab for enabling me to do this work in ideal conditions, and the ETH library for
providing a calm place to work.
23
Chapter 4. Conclusion
24
Bibliography
[1] D. Titterton, J. Weston: Strapdown Inertial Navigation Technology. The
American Institute of Aeronautics and Astronautics, second edition, 2004.
[2] S. Beeby, G. Ensell, M. Kraft, N. White MEMS Mechanical Sensors
Artech House - MEMS Series, pp. 173-208, 2004.
[3] N. M. Barbour: Inertial Navigation Sensors. Nato Report, C. S. Draper
Laboratory, Cambridge, USA,
[4] C. Hierold: Microsystems Technology Lecture, chapters : Transduction Techniques, Micro Resonators and Resonant Microsensors. At ETHZ, Zurich 2012.
[5] O. J. Woodman: An Introduction to inertial navigation. University of Cambridges Computer Laboratory Technical Report, Cambridge 2007.
[6] H. Hou: Modeling inertial sensors errors using Allan variance. M.S. thesis, MMSS Res. Group, Dept. Geomatics Eng., Univ. Calgary, Calgary, AB,
Canada, UCGE Rep. 20201, Sep. 2004.
[7] W. Stockwell:
Angle Random Walk. Crossbow Technology Inc.,
http://www.xbow.com.
[8] IEEE Standard Specification Format Guide and Test Procedure for Single-Axis
Interferometric Fiber Optic Gyros
[9] D. W. Allan: Statistics of Atomic Frequency Standards. Proceedings of the
IEEE, Vol. 54, No. 2, February 1966.
[10] N. El-Sheimy, H. Hou, X. Niu: Analysis and Modeling of Inertial Sensors
using Allan Variance. IEEE transactions on instrumentation and measurement,
vol. 57, NO. 1, 2008
[11] A. Kealy, G. Roberts, G. Retscher: Evaluating the performance of low
cost MEMS inertial sensors for seamless indoor/outdoor navigation. IEEE Position Location and Navigation Symposium, May 2010.
[12] R. Houlihan, M. Kraft: Modeling of an accelerometer based on a levitated
proof mass. Microelectronics Centre, University of Southampton, UK, June
2002.
25