Professional Documents
Culture Documents
Series Editor
Dr. Nikolas Xiros
School of Naval Architecture and Marine Engineering
University of New Orleans
2000 Lakeshore Dr. Ste 914
New Orleans, LA 70124
USA
E-mail: nxiros@uno.edu
A High-Rate
Virtual Instrument
of Marine Vehicle
Motions for Underwater
Navigation and Ocean
Remote Sensing
ABC
Author
Chrystel Gelin
San Diego, CA
USA
This work was conducted on the basis of the author’s thesis within Florida
Atlantic University’s Ocean Engineering Graduate Program requirements with
the advice of Dr. N. Xiros and advisory committee members Drs. M. Dhanak,
F. Driscoll, P. Beaujean and J. VanZwieten.
This book is dedicated to my family, friends and colleagues who have supported
me through the years and kept on believing in the work I did and its happy ending.
I’m dedicating it particularly to my dear husband, Gregory, who has put up
with these many years of me working long hours, my teeth grinding preventing
him from sleeping and the overall stress on our household during that journey.
I also dedicate this work to my tiger team, my grandmother, Lea, my mother,
Chantal, my father, Jacques, my little brother, Cyril and my best friend Anne.
Finally I look forward for my son, James, to be of age to share that work and
experience with him and why not inspire a scientific path…
Thank you. Stubbornness does pay sometimes …
Contents
1 Introduction ...................................................................................................... 1
1.1 Autonomous Surface Vessels for Hydrographic Data Acquisition ............. 1
1.2 Gateway USVs ............................................................................................ 2
1.3 Proposed System ......................................................................................... 4
1.4 Problem Statement ...................................................................................... 4
1.5 Contributions ............................................................................................... 5
1.5.1 Book Outline ..................................................................................... 6
6 Conclusion ....................................................................................................... 85
References ............................................................................................................ 93
Fig. 39 For the sets 1 (a), 3 (b) and 5 (c), velocity obtained by data fusion,
in blue, plotted against the expected velocity VZ, in red. The difference
between the two signals is in black. .......................................................... 45
Fig. 40 For the sets 1 (a), 3 (b) and 5 (c), position obtained using a high pass
filter on the integrated velocity, in blue, plotted against the expected
position Z, in red. The difference between the two signals is in black. ..... 46
Fig. 41 For the sets 1 (a), 3 (b) and 5 (c), position obtained by data fusion,
in blue, plotted against the expected position Z, in red. The difference
between the two signals is in black. .......................................................... 47
Fig. 42 IMU, tilt sensor, and TCM2 compass attached to a rigid plate attached
to the cart. .................................................................................................. 47
Fig. 43 Methodology used to find the data fusion frequency between IMU
and GPS measurement to recover full frequency estimate of the system’s
position and velocity. ................................................................................ 49
Fig. 44 Square path, as perceived by the DGPS. ................................................... 49
Fig. 45 Square path proceeding in a zigzag pattern between corners,
as perceived by the DGPS. ........................................................................ 50
Fig. 46 Circle path as perceived by the DGPS. ..................................................... 51
Fig. 47 Roll and Pitch of the cart measured by the tilt sensor during the first
trajectory ((a) and (b)), the second trajectory ((c) and (d)) and the third
trajectory ((e) and (f)). ............................................................................... 52
Fig. 48 PSD of the north component, in blue, and the east component, in red,
of the IMU acceleration during square trajectory (a), square path
by processing in zigzag course (b) and the circle trajectory (c). ............... 53
Fig. 49 Influence of frequencies above 2Hz on the IMU acceleration
measurements for the three trajectories of the on shore test of the data
acquisition system. .................................................................................... 54
Fig. 50 PSD of the DGPS position (a), DGPS velocity (b) and IMU acceleration
(c) for the first trajectory of the on shore test, following a square path.
The blue signal corresponds to the north component of the measurement
and the red signal to the east component. .................................................. 55
Fig. 51 Data fusion diagram between the IMU acceleration data and
the DGPS velocity measurements in order to obtain the enhanced
velocity estimate. ....................................................................................... 56
Fig. 52 PSD at particular steps of the data fusion process between
the DGPS north component velocity and the IMU north component
acceleration................................................................................................ 57
Fig. 53 Comparison in the time domain between the merged velocity (red)
and the velocity obtained by direct integration of the raw IMU
acceleration signal (black). The blue signal is the DGPS velocity
measurement. The upper panel shows the north component of the
signal (a) and the lower, the east component (b) ....................................... 58
XIV List of Figures
Fig. 54 Data fusion diagram between the DGPS position measurement and the
merged velocity estimate obtained by fusing the IMU acceleration data
and the DGPS velocity. ............................................................................. 59
Fig. 55 Crosscorrelation (a) (respectively (b)) between the north, (respectively
east) component of the DGPS velocity and the north (respectively east)
component of the merged velocity estimates. Similarly, (c) (respectively
(d)) corresponds to the crosscorrelation between the north (respectively
east) component of the DGPS position and the north (respectively east)
component of the merged position estimates............................................. 60
Fig. 56 The Florida Current ................................................................................... 61
Fig. 57 Trajectory perceived by the DGPS during the first (a) and second (b)
maneuver at sea. ........................................................................................ 62
Fig. 58 Close ups around the data fusion frequency, 0.05Hz, of the PSD of the
velocity measurement from the DGPS (blue), the acceleration estimate
from the IMU (black) and the enhanced estimate of the velocity obtained
by data fusion (red).................................................................................... 63
Fig. 59 Time series of the vessel’s enhance velocity measurement obtained
by data fusion with its north (east, down) component in blue (red, black)
for the first maneuver (a, b, c) and second maneuver (d, e, f) ................... 63
Fig. 60 Diagram of the necessary reference frame transformations to transform
the vessel’s enhanced velocity measured by the data acquisition system
into the ADCP Beam coordinate frame. .................................................... 65
Fig. 61 Ship velocity, in blue, along beam 2 (a), and 3 (d) compare to the
contaminated measurement of the water current, in black, along beam
2 (b) and 3 (e), and to the true water current, in red, along beam 2 (c)
and 3 (f) during the first maneuver while the beams 2 and 3 are looking
forward. ..................................................................................................... 66
Fig. 62 Ship velocity, in blue, along beam 1 (a), and 4 (d) compare to the
contaminated measure of the water current, in black, along beam 1 (b)
and 4 (e), and to the true water current, in red, along beam 1 (c) and 4 (f)
during the first maneuver while the beams 1 and 4 are looking aft. .......... 67
Fig. 63 Uncorrected ADCP velocity profile along beam 2, looking forward,
during the first maneuver going south then east ........................................ 68
Fig. 64 Corrected ADCP velocity profile along beam 2, looking forward,
during the first maneuver going south then east. ....................................... 69
Fig. 65 Uncorrected ADCP velocity profile along beam 3, looking forward,
during the first maneuver going south then east ........................................ 69
Fig. 66 Corrected ADCP velocity profile along beam 3, looking forward,
during the first maneuver going south then east. ....................................... 70
Fig. 67 Uncorrected ADCP velocity profile along beam 1, looking aft,
during the first maneuver going south then east ........................................ 70
Fig. 68 Corrected ADCP velocity profile along beam 1, looking aft,
during the first maneuver going south then east. ....................................... 71
List of Figures XV
Table 1 Specifications of the 300 KHz ADCP RDI Workhorse Sentinel ............... 9
Table 2 Specifications of the BEI Inertial Measurement Unit MotionPakII ......... 11
Table 3 Specifications of the TCM2 biaxial inclinometer and a triaxial
magnetometer compass module ............................................................... 12
Table 4 Specifications of the Fredericks Company ± 60 Degree Angle Range
tilt sensor .................................................................................................. 12
Table 5 Specifications of the GARMIN Global Positioning System
76 receiver ................................................................................................ 13
Table 6 Mean and standard deviation of the tilt sensors’ roll and pitch
as well as the influence it could have on the IMU acceleration if not
considered for the three trajectories of the on shore test of the data
acquisition system. ................................................................................... 52
Table 7 Results from the peaks of frequency detection corresponding
to the cart’s motion for the three trajectories............................................ 56
Table 8 Estimates of the standard deviation of the merged velocity signal
for the three trajectories of the on shore data acquisition test. ................. 60
Table 9 Ship’s enhanced velocity measurement, uncorrected and corrected
ADCP water current measurement in beam coordinates, at the first bin,
during the first maneuver ......................................................................... 67
Table 10 Estimates of uncorrected and corrected ADCP water current
measurement looking at the velocity profiles in beam coordinates
during the first maneuver. ...................................................................... 73
Table 11 Ship’s velocity, uncorrected and corrected ADCP water current
measurement in beam coordinates, for the first bin, during the second
maneuver ................................................................................................ 75
Table 12 Estimation of uncorrected and corrected ADCP water current
measurement looking at the velocity profiles in beam coordinates
during the second maneuver. .................................................................. 76
Table 13 Ship’s velocity, uncorrected and corrected ADCP water current
measurement in North-East-Up coordinates, for the first bin,
during the first and second maneuver..................................................... 79
Table 14 Estimation of uncorrected and corrected ADCP water current
measurement looking at the velocity profiles in NEU coordinates
during the first and second maneuver..................................................... 84
Table 15 Estimates of the standard deviation of the merged velocity signal
for the three trajectories of the on shore data acquisition test. ............... 89
XVIII List of Tables
C. Gelin: A Low-Cost, High Rate Motion Measurement System, NAMESS 1, pp. 1–6.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
2 1 Introduction
floor (LBL) or surface (inverted LBL) are used to triangulate vehicle position. If a
LBL is used, the UUV is restricted to operate within the beacon grid to obtain
geodetic position data. Offshore deep water deployment of LBL arrays is difficult
and if moored on the surface, buoyed beacons are not clandestine and therefore
vulnerable if deployed in a hostile theater.
In a Short-Base-Line (SBL) system, arrays of transducers are hull-mounted on
large vessels and typically separated by several tens of meters. Such large vessels
are easily detected, leading to a non clandestine solution. Alternatively, in an
Ultra-Short-Base-Line (USBL) positioning system, arrays of transducers are
separated by up to several centimeters, and potted into a single small hydrophone
array. Low system complexity and small size makes USBL an ideal tool to help
navigate UUVs because they are easy to deploy and small enough to be
clandestine. In addition, there is no need to deploy arrays of transponders because
there is only a single transceiver (Vickery 1998). Thus, the USBL is an ideal
UUV acoustic positioning system for GATEWAY type USVs. USVs are ideal
mobile GATEWAY platforms that can provide communications and positioning
to UUVs through the air-sea interface when mounted with a USBL and acoustic
modem. Unfortunately, little work exists on operating UUVs and USVs in a
cooperative manner. One such system, the Advanced System Integration for
Managing the Coordinated Operation of Robotic Ocean Vehicles (ASIMOV)
project, was developed with the objective of achieving coordinated operation of an
Autonomous Surface Craft (ASC) and an UUV for marine data acquisition while
ensuring a fast communication link between the two vehicles (ISR-IST 2000).
In this project, two
robotic ocean vehicles are
used: the DELFIM ASC
and the INFANTE AUV.
The DELFIM ASC is a
small catamaran that is
3.5m long and 2m wide,
with a mass of 320 Kg
(Figure 2). The DELFIM
performs automatic
marine data acquisition
and serves as an acoustic Fig. 2 Autonomous Surface Craft DELFIM, part of the
relay between submerged ASIMOV project, designed, and built by the Institute
craft and a support vessel. for System and Robotics, beginning in 1998
Besides operating as a
communications link, the DELFIM has a stand-alone sensor suite capable of
maneuvering autonomously and performing precise path following while carrying
out automatic marine and bathymetry data acquisition. This sensor suite includes
on-board systems for navigation, guidance and control, and mission control; an
Ultra Short Baseline unit (USBL) to position the AUV; an RF above water
communication link; and a high data rate underwater acoustic communication
system. Navigation is done by integrating motion sensor data obtained from an
attitude reference unit, a Doppler logger, and a DGPS. Transmissions between the
4 1 Introductioon
AUV, this ASV, the fixeed GPS station, and the control center installed on-shorre
are achieved with a radio link that has a range of 80 Km. In order to achievve
higher bandwidth acoustiic co-mmunications between the USV and the AUV, thhe
vertical channel (high data rate underwater acoustic system) is used.
The USBL hydrophone is mounted to the USV on a long rigid strut a distance
from the GPS antenna. As a result, as the USV moves and responds to ocean
waves, the USBL will also move. If the USBL is to provide accurate positioning,
the position and orientation of the hydrophone must be measured at a high rate to
correct the measured bearing and position offsets. However, standard GPS
receivers are unable to provide the rate or precision required when used on a small
vessel. To overcome this, a high rate and high precision position and orientation
measurement system is developed. The work integrates a set of low cost inertial
sensors and a GPS receiver to calculate the USV’s inertial motion. This will be
used to correct/transform USBL based position and bearing measurements even
when the surface vessel is required to operate in rough seas. Fundamental to any
navigation and control system is the measurement of the vehicles geodetic
position, orientation and velocity in 3 or 6 degrees of freedom. The system
developed in this book provides this information. Included in the navigational
instrumentation suite is an ADCP that measures the water velocity, but it can also
measure the speed of the USV over the ocean floor.
ADCPs measure the relative velocity between its sensor heads and the water
using the Doppler shift and time dilation of an acoustic pulse. By transmitting
acoustic pulses at a fixed frequency and listening to the Doppler shift of echoes
returning from sound scatterers in the water, water velocity estimates can be made.
While ADCPs non-intrusively measure water flow, they suffer from the inability
to discriminate between motions in the water column and self-motion. When
mounted on a moving platform, the measured velocity is the sum of the platform
velocity and the water velocity. Thus, the vessel motion contamination needs to be
removed to analyze the data and avoid long average times. The system developed
in this book provides the motion measurements and processing to accomplish
this task.
1.5 Contributions
The work presented in this book integrates a set of instruments and develops a
software package that measures and calculates the motion of the USV (Unmanned
Surface Vehicle) to aid in the navigation and control and enhance the performance
of the USBL positioning system. As well, the motion measurement system
actively controls the onboard ADCP and corrects the water velocity measurements
for ship motion contamination - ship surge, sway, heave, roll, pitch and heading.
The simplicity of the data acquisition system allows it to be easily deployable and
adaptable to new applications after setting the correct initial parameters.
The motion measurement system of the USV consists of an Inertial
Measurement Unit (IMU) with accelerometers and rate gyros, a GPS receiver, a
flux-gate compass, a roll and tilt sensor, and an ADCP. Interfacing all the sensors
is challenging because of their different characteristics. Some of the instruments
have digital output (Compass/ADCP/GPS) while others have an analog output
(IMU/tilt sensor). Among the sensors using RS232 serial port communication two
different output formats are used. The TCM2 compass and the GPS use the
NMEA 0183 (National Marine Electronics Association) standard while the RDI
6 1 Introduction
ADCPs use the Doppler effect to acoustically measure water velocity. They
transmit sounds, in the form of acoustic pulses (pings), perpendicular to the
transducer faces (the source and receivers) at a fixed frequency and record the
echoes at discrete intervals of time (depth bins). The ping is reflected by scatterers
moving with the water and the reflected signal is Doppler Shifted if the water has
a relative velocity component parallel to the acoustic beam. Using four transducers
pointed in different directions, each tilted at equal angles from the vertical axis
of the ADCP and oriented in pairs that point in perpendicular planes, the
ADCP computes the 3-Dimensional water velocity vector for each bin. All beams
C. Gelin: A Low-Cost, High Rate Motion Measurement System, NAMESS 1, pp. 7–18.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
8 2 Instrumentation and Data Acquisition System
The 300 kHz RDI Broadband Workhorse ADCP (Figure 4) is mounted 45o
counter-clockwise from its normal beam 3 forward orientation (Figure 6), so that
beams 2 and 3 are looking ahead, and beams 1 and 4 are looking aft. In this
configuration, all four beams detect similar magnitudes of Doppler shift in relation
to surge and sway, which will aid in removing errors during post-processing
(Figure 5).
2.1 The Sensors 9
Fig. 6 ADCP Beam orientation with beam 3 at 45 degrees with respect to the heading,
looking from underneath the boat.
RANGE ±15°
ACCURACY ±0.5°
PRECISION ±0.5°
RESOLUTION 0.01°
COMPASS
ACCURACY ±2°
PRECISION ±0.5°
RESOLUTION 0.01°
MAXIMUM TILT ±15°
10 2 Instrumentation and Data Acquisition System
0.14
Standard Deviation [m]
0.12
0.1
0.08
0.06
0.04
0.02
2 4 6 8 10 12 14 16
Bin Siz e [m]
Fig. 7 ADCP velocity standard deviation in function of the size of the bins and number of
pings per ensemble chosen on the mission command set.
For this work, the single ping data is motion corrected and the variance of a
single ping signal provides a basis to set the maximum level for the accuracy and
precision of the vessel motion measurements. A value of less than 1 cm/s is
desired for velocity measurements, which is about one half the value of the lowest
single ping standard deviation (with an 8m bin size).
The low cost BEI Systron Donner Inertial Division MotionPak II contains three
orthogonally mounted micromachined quartz angular rate gyroscopes and three
2.1 The Sensors 11
Table 4 Specifications of the Fredericks Company ± 60 Degree Angle Range tilt sensor
PARAMETER VALUE
UPDATE RATE 0.5Hz, continuous
GPS ACCURACY < 15 M (49 Ft) RMS 95% typical
DGPS (USCG) ACCURACY 3-5 M (10-16 Ft), 95% typical
3 M (10 Ft), 95% typical with DGPS
DGPS WAAS ACCURACY
corrections
Fig. 13 Overview of the data acquisition system, including the sensors, computers and links
Systems. As the IMU and tilt sensor both output an analog signal, a Diamond-
MM-32-AT AD/DA converter board is included in the PC104 stack and is
configured with 32 single-ended channels. A Simpletec flash IDE 1 GB drive
mounted on its controller board is used to store data. A VGA card can be installed
when needed, to facilitate software debugging.
Fig. 15 Block diagram of the acquisition hardware, including the sensors, computers and
links
While the data flow from the IMU, GPS, tilt sensor, and the compass is one
way, from the instrument to the data acquisition computer, data flow is
bidirectional between the data acquisition system and the ADCP over a digital
serial communication line. Unlike the other instruments which are independent
from the data acquisition computer and output data continuously (analog) or at
fixed rates (digital), the ADCP is programmed prior to sampling and is
interrogated to trigger each sample. Thus, the ADCP requires a dedicated serial
port and communicates at 115200 baud with a sample rate set to 1Hz.
Fig. 17 Block diagram of the links between the host PC, the target PC104 stack, the
sensors, and Operating Systems of the entities
18 2 Instrumentation and Data Acquisition System
One guiding goal in the software development is that the program must be
quickly and easily understandable and be developed in a framework that readily
allows future modifications and additions without understanding of the program in
its entirety. Thus, a modular approach is adopted that sees the software broken
down into five key sections: 1) system initialization, 2) the data acquisition,
translation and conversion; 3) rotational motion estimation; 4) translational motion
estimation; and 5) ADCP signal decoding and correction. Each one of these main
modules are themselves composed of sub-modules.
Chapter 3
Data Processing
The methods developed in this book are used to calculate high rate position and
orientation states of a moving body relative to the earth using a comprehensive
sensor suite. Unfortunately, the different sensors move relative to each other and
measure quantities along different, uncoupled axes. For example, the
MotionPack’s accelerometers measure acceleration and its rate gyros measure
angular velocity along three perpendicular directions that move with the ship with
the acceleration signals contaminated by gravitational acceleration, which is
always orientated towards the center of the earth. Similarly, the TCM2 compass
measures the heading of the moving body with respect to the magnetic North, and
the tilt sensor measures the roll and pitch angles with respect to the ship.
Conversely, the GPS computes geodetic latitude, longitude and height above sea
level relative to the rotating earth. Therefore, the data measured from each one of
the sensors need to be rotated into a common frame of reference. The first section
of this chapter, section 3.1.1, defines the different reference frames used. The
second section, section 3.2, derives the transformations necessary to convert
vectors between the reference frames and into a common reference frame. The
third section, section 3.3, describes the data fusion algorithm and its
implementation. Finally, the last section, section 3.4, defines the method used to
process the ADCP data.
From the GPS geodetic data, the GPS position is calculated in the Earth-Centered
Earth-Fixed frame, ECEF, described firstly in the following section. The GPS
velocity is obtained using the ground speed and the course over ground provided
by the GPS and expressed in the North-East-Down (NED) frame described
secondly in the following section. In the NED frame is also expressed the TCM2
compass heading. The third section explains the body-fixed frame, where the tilt
sensor, the IMU and the ADCP outputs.
C. Gelin: A Low-Cost, High Rate Motion Measurement System, NAMESS 1, pp. 19–35.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
20 3 Data Processing
Fig. 18 Representation of the axis of the Earth Centered Earth Fixed and Earth Centered
Inertial Frames
The Earth-Centered Inertial frame (ECI, i-frame) has its origin at the center of the
earth, with axes [Xi, Yi, Zi]T that are non-rotating with respect to the fixed stars.The
Earth-Centered Earth-Fixed (ECEF, e-frame) has its origin at the center of the
earth, with axes [Xe, Ye, Ze]T fixed to the Earth that rotates at a rate of 15.041067
°/hr (7.2921.10-5 rad/s) with respect to the ECI. The Zi and Ze axes point from the
center of the Earth upwards towards the North Pole. The Xi and Xe axes point
horizontally in the plane of the equator from the center of the Earth towards the
equator at zero latitude. The Yi and Ye axes are chosen to complete the right hand
coordinate system (Figure 18). The ECI and ECEF frame are taken into account in
the process of obtaining the GPS position data as described in section 3.2.1.
3.1 Reference Frames 221
The body-fixed frame, ℑB, is a moving coordinate frame rigidly attached to eitheer
a ship or sensor packagee to which the sensors’ axes of sensitivity are aligned.
Traditionally, The x-axis points
p forward, the y-axis points starboard, and the z-axxis
completes a right-hand orthogonal system by pointing downward. The IMU, thhe
tilt sensor and the ADCP uses this coordinate system as described in sections 3.2.2
and 3.2.3.
22 3 Data Processing
Fig. 20 Ship-fixed coordinate reference frame (red) and 6 degrees of Freedom motion
variables for a marine vessel (sway, surge, heave, pitch, roll and yaw) (Fossen 1994)
Fig. 21 Diagram of the sensors output variables and the coordinate transformations
The following section describes the transforms used to rotate data from the
different sensor frames to the NED reference frame, the common frame of
reference (Figure 21).
. (88)
3.2.2 Transformatio
ons from Component Reference Systems to
Body Fixed Reeference System
The IMU and the ADCP output data along axes that are defined relative to theeir
individual orientation. Often
O because the axes of the instruments may not bbe
aligned with the axes of th
he ship, this data needs to be rotated and translated to thhe
ship fixed frame prior to transformation to the NED frame. For examplle,
accelerations measured using
u the IMU are transformed to the body fixed fram me
using:
, ,
, , , (99)
, ,
3.3 Data Fusion 25
cos cos cos sin sin sin cos cos sin cos sin sin
cos sin cos cos sin sin sin sin sin cos cos sin . (10)
sin cos sin cos cos
The inverse transformation from the NED frame to the body frame is:
cos cos cos sin sin
cos sin sin sin cos cos cos sin sin sin cos sin . (11)
cos sin cos sin sin sin sin cos cos sin cos cos
Unfortunately rotational velocities in the body fixed frame do not directly measure
the Euler rates; instead they measure angular velocities about a fixed set of axes.
These angular rates do not constitute a vector space and therefore can’t be
integrated to provide a measure of orientation. Thus, the angular rates are
converted to Euler rates using:
1 sin tan cos tan
Γ 0 cos sin , (12)
0
Γ Γ , (13)
·
where the dot above the variable (·) represents a time derivative, .
but the update is slow and the resolution coarse. The derivative measurement
is often high frequency, but is subject to a bias that results in drift of the
integrated signal. Thus, each one of these signals cannot be used independently for
applications that require high quality navigation data. However, they have
complementary characteristics that can be leveraged to obtain a combined useful
signal. The data fusion techniques developed in this work combine the
complementary outputs of sensors measuring a related state to eliminate the drift
of integrating the measurement , while increasing the rate and resolution of
. The pre-emphasized signal, , is obtained by summing the signal
with the derivative signal , such that (Mudge and Lueck 1994):
, (14)
Ω
where the scaling factor Ω , denoting the cutoff frequency, is a real positive
constant. The choice of Ω is determined by the characteristics of the
complementary region of the two sensors. For frequencies that are small compared
to the cutoff frequency (Ω Ω , the signal portion of the spectrum comes
predominantly from , while for Ω Ω , the signal is predominately from
. In the frequency domain, the pre-emphasized signal is:
Ω
Ω 1 Ω , (15)
Ω
Ω Ω , (16)
Ω
yielding:
Ω
Ω Ω Ω 1 Ω Ω , (17)
Ω
Ω
matrix Γ(12) , converting angular rates to Euler rates, and thus, they need to be
known a priori to calculate from w (13). The Euler angles, , are obtained by
merging the low-frequency Euler angles, , and , calculated from the tilt
measurements and , and the compass heading, , with the high-frequency
IMU angular rates, , , T . The estimated Euler angles are then used to
convert the IMU acceleration from body-fixed frame, , , T , to the NED
T
frame, , , , where gravity is removed and the signal is merged with the
GPS velocity, , , . The result of the last data fusion leads to a high
quality merged measure of the ship’s velocity , , .
, (18)
and
sin
asin (19)
cos
The Euler angles calculated from the tilt sensor data were calibrated using the
accelerometers. This was done using an accelerometer to measure acceleration
along its axis of sensitivity and at frequencies where the translational acceleration
of the IMU is negligible so that the accelerometers only measure gravity. Thus, at
low frequencies, with little translational motion, the accelerometers can provide an
independent measure of the Euler angles, (Figure 23), (Lueck, Nahon 2000):
, sin , (20)
and
To compare the angular measurements of the tilt sensor with these of the
accelerometer, the IMU and tilt sensor were mounted on the same rigid plate and
very slowly rotated through the expected tilt range. Data for each sensor was
simultaneously acquired and recorded (Figure 24). The vessel is expected to rotate
less than 20°, and within this range, the angles measured by both the sensors agree
within ± 0.54366° for and within ± 0.48737° for . At angles greater than 30°,
the accuracy of the tilt sensor decreases with range. At the maximum tilt of 42°,
the tilt sensor overestimated the tilt by up to 10° and this difference results
because of the non-linear response of the tilt sensor, which can be corrected for
using a calibration table. When the tilt sensor is steady, local vibrations can cause
the fluid to slosh; this coupled with processes internal to the sensor causes a
± 1.1541° fluctuation in the tilt measurements. The tilt sensor can only provide
reliable data at frequencies less than 1Hz and higher frequency signal is removed
by applying 1st order Butterworth low-pass filter at a cutoff frequency of 1Hz to
the tilt sensor roll, , and pitch, .
3.3 Data Fusion 29
50 (a)
Phi [Deg]
0
-50
0 100 200 300 400
40 (b)
Theta [Deg]
20
0
-20
-40
0 100 200 300 400
T ime [s]
Fig. 24 Comparison between the low frequency estimates of Euler angle (a) ( (b))
obtained from the IMU (blue) and from the Tilt sensor (red).
The TCM2 compass provides a direct measure of the yaw Euler angle, , but
similar to the tilt sensors, it also has a finite response time and sudden changes in
heading cannot be directly measured with this instrument. Thus, only the low-
frequency yaw signal is useful and the low-frequency estimate of , , is
obtained by applying a 1st order Butterworth low-pass filter to the compass
heading, with a cut-off frequency of 1Hz.
At higher frequencies (> 1/30Hz) the rotation rates, , , T measured
T
with the IMU provide the needed angular measurements, , , ,
that are used as angular derivative signal in the data fusion. The compass, tilt
sensor and IMU signals are combined to create the pre-emphasized signal
T
, , , according to (14):
, (22)
Ω
, (23)
Ω
and
. (24)
Ω
30 3 Data Processing
Fig. 25 Diagram of the data fusion IMU / TCM2/ Tilt sensor to obtain Euler angles, .
40 (a)
Phi [Deg]
20
0
-20
-40
0 50 100 150 200 250
40 (b)
Theta [Deg]
20
0
-20
-40
0 50 100 150 200 250
400 (c)
Psi [Deg]
200
0
300 320 340 360 380
T ime [s]
Fig. 26 Comparison between the Euler angles (a), (b) from accelerometers, blue and
(a), (b) from data fusion, red and between the compass heading, blue and Euler angle
from data fusion in red (c). The black is the difference of blue and red signals.
To quantify the accuracy of the merged Euler angle signals at lower frequencies
(< 5 Hz), they are compared with an independent measure of the Euler angles
calculated using the accelerometer (Figure 26) for motion set 1. Figure 26 shows
the comparison between the Euler angle from the accelerometers in blue and
from Data fusion, red (a) from the first part of the test. On the second panel (b),
from the first part of the test, the Euler angle from accelerometers in blue is
compared to from the data fusion in red. Finally, on the last panel (c), from the
second part of the test, the unwrapped heading from compass is in blue, and from
Data fusion is in red. The black is the difference of the merged and IMU Euler
angle. During the post processing, a low-pass filter (cutoff at 1 Hz) is applied to
both the Euler angles derived from the IMU and the Euler angles obtained by the
data fusion and the standard deviation of their difference is ± 1.9° for , ± 1.5° for
(most likely due to inaccuracy of the tilt sensor pass 30°) and ± 8° for .
The raw signals are filtered both prior to and during data fusion and this creates
a frequency dependant phase distortion. Thus, to quantify any potential signal
delay, the cross-correlation between the Euler angles measured strictly with the
accelerometer are compared to the merged Euler angle estimates and normalized
so that the autocorrelations at zero lag are identically to 1.0. Comparing the Euler
angles from the accelerometer and from the data fusion, and are 99.23%
correlated with a delay of 7 samples (0.0547s), and are correlated at 99.66%
and delayed by 11 samples. The compass heading and correlate at 99.38% with
no delay. The compass is a digital serial device that is first processed and coded
into RS-232 format at the compass, broadcast, then received and processed by the
data acquisition computer. Any delay is due to the lag associated with this
overhead and should be accounted for within the motion correction of the ADCP
although in our case an agreement of more than 99% is considered acceptable.
32 3 Data Processing
To validate the high frequency component of (Figure 25), the Euler rates
are integrated and compared to the Euler angles, , yield through the data fusion
(Figure 27). The figure shows the third part of the test during the high frequency
set of motion. In the first panel (a), the high frequency component of the integral
of Euler rate , blue, is compared to the high frequency component of the merged
Euler angle , red. In the 2nd panel (b), the high frequency component of the
integral of Euler rate , blue, is compared to the high frequency component of the
merged Euler angle , red and in the last panel (c), the high frequency component
of the integral of Euler rate , blue, is compared to high frequency component of
the merged Euler angle , red. The black is the difference of the two signals in
each plot.
As expected, the difference between directly integrating yields to a low
frequency drift. The standard deviation is ±0.44° for , ± 0.38° for and ± 0.58°
for . To quantize the phase agreement between the Euler rates and the Euler
angles, the corresponding cross-correlations are computed, and are 99.99%
correlated with no delay, and are correlated at 99.99% with no delay. Looking
at the first 10s of Figure 27 third panel (c) the high frequency component of the
integral of Euler rate is not following the same behavior of the high frequency
component of the merged Euler angle . This is due to the high-pass filtering
process itself and leads to and being delayed by 4 samples and 84.30%
correlated.
(a)
Phi [deg]
20
0
-20
400 410 420 430 440 450 460 470 480
(b)
Theta [deg]
20
0
-20
400 410 420 430 440 450 460 470 480
50
(c)
Psi [deg]
-50
400 410 420 430 440 450 460 470 480
T ime [s]
Fig. 27 During the third part of the test, high frequency set of motion, comparison between
the high frequency component of the integral of Euler rate (a), (b), (c) in blue, and
the high frequency component of the merged Euler angle (a), (b) and (c) in red. The
black is the difference of the two signals in each plot.
3.3 Data Fusion 33
(a)
PSD [dB/Hz]
0
10
-2 -1 0
10 10 10
(b)
PSD [dB/Hz]
0
10
-2 -1 0
10 10 10
(c)
PSD [dB/Hz]
0
10
-2 -1 0
10 10 10
Frequency [Hz]
Fig. 28 In red, PSD of Merged Euler angle (a), (b) and (c); in blue, PSD of Euler
angle from tilt sensor (a), (b), and (c); in black, PSD of integrated Euler rate (a),
(b), and (c).
Fig. 29 Diagrammatic representation of the data fusion of the IMU data and the GPS data
used to obtain the ships velocity .
34 3 Data Processing
The enhanced velocity of the ship, , is obtained by directly fusing the high
frequency (128Hz) acceleration measurement from the IMU, , and the low
frequency (0.5Hz) velocity obtained from the speed and course overground output
T
from the GPS, , , (Figure 29). For this calculation, the
accelerations are rotated to the NED frame from its instrumentation frame and the
gravitational contamination is removed,
T
where 00 . (25)
T
The and , , are then combined to create the pre-
emphasized velocity using (14):
, (26)
Ω
1
Ω (27)
and
1
(28)
Ω
After this, is to be convolved with a 1st order Butterworth filter with a cutoff
frequency Ω to obtain the full frequency measure of the ships velocity, ,
according to (17).
The position is then obtained by merging the enhanced velocity,
T
, , , with the latitude and longitude measured with the GPS. The latitude
and longitude, measured by the GPS, are converted to NED position
measurements using (6) and (8),
. (29)
, ,
1.4619 , , , (31)
°
and
, , , ,
4 cos 20° (32)
0.2666 , , , , .
where the index i is the number of the bin, each bin representing a different depth
cell of the profiled water column, from 1 to 128 in our case. The Earth fixed water
velocity of bin i is obtained using the body to inertial transformation matrix
and subtracting the measured velocity of the ship,
. (33)
Chapter 4
Motion Observation and Experimental Results
This chapter details the experiments used to choose the best data fusion point (Ω )
to obtain the full frequency measure of the ships velocity, , and its position, .
The first experiment investigates properties of the vertical NED acceleration and
different methods available to obtain the merged vertical velocity, and merged
position, . The second experiment investigates properties of the data acquisition
system on shore, without the ADCP in order to validate the data fusion point of
the velocity and position data.
C. Gelin: A Low-Cost, High Rate Motion Measurement System, NAMESS 1, pp. 37–60.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
38 4 Motion Observation and Experimental Results
0
Az[m/s2]
-2
-4
-6
16
550 (b)
(a)
500 14
450
12
350 10
300
8
250
6
200
150 4
100
2
50
14 3
12
2.5
10
2
8
1.5
6
1
4
2 0.5
2.2 2.5
(e) (f)
2
1.8
2
PSD:Az5 filtered, Az5
1.6
PSD:Az5 filtered, Az5
1.4
1.5
1.2
1
0.8 1
0.6
0.4 0.5
0.2
Fig. 33 spectrum from top to bottom for the set 1 (a), 3 (c) and 5 (e) of periods about 5,
15 and 25 s (left side) and filtering effect on the signal (right side) for the set 1 (b), 3 (d)
and 5 (f).
A close up, in time domain, of three sets of period 5 (a), 15 (b), and 25s (c), is
shown in Figure 34. The black signal represents the recorded data and the red
signal the filtered data with a 2nd order bandpass Butterworth with cutoff
frequencies 0.01 and 0.4Hz. Figure 34 highlights the filtering necessity as higher
the period of the set, higher the noise and lower the signal to noise ratio.
40 4 Motion Observation and Experimental Results
(a) (b)
2 1
1.5 0.8
1 0.6
0.4
Az Set1 [m/s2]
Az Set3 [m/s2]
0.5
0.2
0
0
-0.5 -0.2
-1 -0.4
-1.5 -0.6
-0.8
-2
-1
200 250 300 350 200 250 300 350
T ime[s] T ime[s]
0.8 (c)
0.6
0.4
Az Set5 [m/s2]
0.2
-0.2
-0.4
-0.6
-0.8
200 250 300 350
T ime[s]
Fig. 34 Measured and filtered acceleration for periods about 5 (a), 15 (b) and 25s (c).
Acceleration measurements are in black while filtered accelerations are in red.
5 120
set 4 set 1
pic at 0.049 Hz (a) (b)
4.5 => period of 20.07 s
pic at 0.1838 Hz
=> period of 5.44 s
100 set 3
4 pic at 0.0705 Hz
=> period of 14.17 s
3.5 set 5
pic at 0.0417 Hz 80
3 => period of 23.95 s
PSD
PSD
2.5 60
2 set 2
pic at 0.1035 H z
set 6 40 => period of 9.66 s
1.5 pic at 0.0288 Hz
=> period of 34.71 s
1
20
0.5
0 0
0.01 0.02 0.03 0.04 0.05 0.06 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22
Hz Hz
Fig. 35 Az PSD for the set 1, 2 and 3 (b) of periods about 5, 10 and 15 s, and for the set 4, 5
and 6 (a) of periods about 20, 25, and 35 s.
4.1 Vertical Motion 41
.
. (34)
0.4 0.05
0.2
0
0
-0.2
-0.05
-0.4
-0.6 -0.1
-0.8
200 250 300 350 200 250 300 350
T ime[s] T ime[s]
0.04 (c)
0.03
Az5 BP,Theo,Theo-BP [m/s2]
0.02
0.01
-0.01
-0.02
-0.03
-0.04
-0.05
200 250 300 350
T ime[s]
Fig. 36 Close up of the acceleration for the set 1 (a), 3 (b) and 5 (c) of periods about 5, 15
and 25s with the expected motion in red, the system acceleration in blue and the difference
between the signals in black.
The black signal’s behavior emphasizes that the slower the motion, the smaller
its resulting standard deviation, the higher the agreement between expected
and obtained signal. The black signal’s standard deviation is respectively
9.6 ± 8.5891 cm/s2 (taking into account the specification of the instrument),
1.5239 ± 1.2652 cm/s2 and 0.66908 ± 0.44294 cm/s2. According to a
crosscorrelation calculation conducted, they agree respectively at 98%, 97.7% and
96% with a delay of 0.015s each.
42 4 Motion Observation and Experimental Results
A close up of the effect of integrating the acceleration (using the Matlab function
cumsum) and using a low frequency contamination removal function (detrend) is
shown for the sets 1 (a), 3 (b), and 5 (c) in Figure 37. Detrending a signal refers to
applying the matlab function detrend to the signal.
The red signal represents the expected velocity obtained by integrating the
expected acceleration, the blue signal is obtained integrating then detrending the
obtained acceleration and the black signal is the difference between expected and
obtained velocities. The black signal’s standard deviation is respectively
6.27 cm/s, 2.3 cm/s and 1.9 cm/s. The lower the standard deviation of the black
signal the better the method since the goal of the method is matching obtained
signal with expected motion. This method fails at removing all low frequency
perturbations leading to the need for another method based on a high-pass
filter use.
4.1 Vertical Motion 43
(a)
0.6 (b)
0.25
0.4 0.2
Vz1 BP,Theo,Theo-BP [m/s]
0.1
Vz5 BP,Theo,Theo-BP [m/s]
0.05
-0.05
-0.1
-0.15
Fig. 37 Difference, in black, between the expected velocity VZ, red, and the obtained
velocity using the detrend function on the integrated acceleration in blue for the set 1 (a), 3
(b) and 5 (c).
A close up of the effect of high-pass filtering on the integrated signal is shown for
sets 1 (a), 3 (b), and 5 (c) in Figure 38. The expected velocity is in red, the blue
signal is the velocity resulting from integrating then high-pass filtering the
obtained acceleration and the black signal is the difference between expected and
obtained velocities. A 4th order Butterworth high-pass filter with cutoff frequency
of 0.021Hz is found to be the most suited filter to minimize phase delay and
eliminate low frequency perturbation due to the integration process. The
Butterworth filter is selected because typical and an Elliptic filter could also have
been used. The black signal’s standard deviation is respectively 6 cm/s, 1,8 cm/s
and 1.3 cm/s which is overall lower than the standard deviations obtained through
the first method. This method fails at removing all perturbations at low frequency
leading to the blue signal following a trend different than the red signal. But the
method provides better results considering the standard deviation of the black
signal than the first method involving use of the detrend Matlab function.
44 4 Motion Observation and Experimental Results
(a) (b)
0.6 0.25
Vz1 CSHP,Theo,Theo-CSHP [m/s]
-0.4 -0.15
-0.2
-0.6 -0.25
(c)
0.15
Vz5 CSHP,Theo,Theo-CSHP [m/s]
0.1
0.05
-0.05
-0.1
-0.15
Fig. 38 For the sets 1 (a), 3 (b) and 5 (c), velocity obtained using a high-pass filter on the
integrated acceleration, in blue, plotted against the expected velocity VZ, in red. The
difference between the two signals is in black.
Results for the data fusion method is shown for the sets 1 (a), 3 (b), and 5 (c) in
Figure 39. The red signal is the expected velocity, the blue signal the obtained
velocity resulting from the data fusion technique, and the black signal the
difference between expected and obtained velocities. The IMU, which is assumed
accurate only at high frequencies, is merged with a null signal at low frequency by
replacing VLFZ by 0 in (28). The black signal is the difference between expected
and obtained velocities. A data fusion point at 1/100Hz is found to be the best
compromise to merge IMU with null signal. With this method, the standard
deviation of the black signal is respectively 6.9 cm/s, 1.9 cm/s and 1.5 cm/s.
Looking at the standard deviation of the black signal for each one of the three
methods and since the best method will have the lowest standard deviations, this
method is not as satisfying as the second method and better than the first method
for the set 3 and 5. This method is selected over the second method because it
introduces less delay than using a 4th order Butterworth filter.
4.1 Vertical Motion 45
0.1
0.2
0.05
0 0
-0.05
-0.2
-0.1
-0.4 -0.15
-0.2
-0.6
-0.25
200 250 300 350 200 250 300 350
T ime[s] T ime[s]
0.15 (c)
0.1
Vz5 F,Theo,Theo-F [m/s]
0.05
-0.05
-0.1
-0.15
Fig. 39 For the sets 1 (a), 3 (b) and 5 (c), velocity obtained by data fusion, in blue, plotted
against the expected velocity VZ, in red. The difference between the two signals is in black.
Two methods are applied to obtain the vertical position from obtained velocity
data. The first method is to integrate the velocity estimates at fixed time steps
using the cumulative summation and to high-pass filter the result. The second
method is to merge the velocity obtained in 4.1.2 with a null signal at low
frequency by replacing Z,LF by 0 in (28). This sub-section describes each of the
two aforementioned techniques. In each of the sub-sections the red signal
represents the expected motion, the blue the system motion, and the black the
difference between the two. The standard deviation of the black signal is used to
quantify the accuracy of the position signal.
46 4 Motion Observation and Experimental Results
4.1.3.1 Vertical Position Calculated Using the High Pass Filtered Integrated
Velocity
0.5
0.4
Z1 CSHP,Theo,Theo-CSHP [m]
0.3
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
200 250 300 350
T ime[s]
0.5 0.5
0.4 0.4
Z3 CSHP,Theo,Theo-CSHP [m]
Fig. 40 For the sets 1 (a), 3 (b) and 5 (c), position obtained using a high pass filter on the
integrated velocity, in blue, plotted against the expected position Z, in red. The difference
between the two signals is in black.
Results obtained using the first method are shown for the sets 1 (a), 3 (b), and
5 (c) in Figure 40. The red signal is the expected position obtained by integrating
the expected velocity, the blue signal is the position estimate obtained by
integrating then high-pass filtering the velocity estimates in 4.1.2, and the black
signal is the difference between expected and obtained position. A 4th order high
pass Butterworth filter with cutoff frequency of 0.021Hz is found to be the most
suited filter limiting phase delay and minimizing low frequency amplification due
to the integration process. The Butterworth filter is selected because typical and an
Elliptic filter could also have been used. The standard deviation of the black signal
with the first technique is 4.8 cm, 4.3 cm, and 3.9 cm respectively.
Results obtained using the second method are shown for the sets 1, 3, and 5 in
Figure 41. The red signal is the expected position, the blue signal the position
obtained using the data fusion technique with the velocity obtained in 4.1.2
merged with a null signal at low frequency by replacing Z,LF by 0 in (28), and the
black signal is the difference between expected and obtained position. A cutoff
frequency of 1/50Hz for the data fusion is found to be the best compromise to
4.2 Data Acquisition System Lab Testing 47
merge the obtained velocity with the null signal. The black signal’s standard
deviation is respectively 6.7 cm, 5.6 cm and 7.6 cm. Although this method
provides a high standard deviation measurement for the black signal, it is selected
for the processing of the vertical position moving forward since it is compatible
with real time applications.
0.5
0.5
0.4
0.4
0.3
Z3 F,Theo,Theo-F [m]
0.3
0.2
Z1 F,Theo,Theo-F [m]
0.2
0.1
0.1
0
0
-0.1
-0.1
-0.2
-0.2
-0.3 -0.3
-0.4 -0.4
-0.5 -0.5
0.5
0.4
0.3
Z5 F,Theo,Theo-F [m]
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
Fig. 41 For the sets 1 (a), 3 (b) and 5 (c), position obtained by data fusion, in blue, plotted
against the expected position Z, in red. The difference between the two signals is in black.
consists of two data fusions: the first data fusion process, involving the IMU
acceleration and the GPS velocity measurement leads to a full frequency
assessment of the velocity measurement. This frequency for the data fusion is
selected so that at frequencies lower than the cutoff frequency (to be determined)
the GPS provides an accurate measure of the system’s velocity, and at frequencies
above the cutoff frequency the IMU provides an accurate estimation of the
velocity. The second data fusion is performed between the merged velocity,
obtained from the first data fusion, and the GPS position measurement. Similar
reasoning to that used for the first data fusion is applied, i.e. for frequencies lower
than the cutoff frequency the position estimate is provided by the GPS, while for
frequencies above the cutoff frequency the measure of position is derived from the
merged velocity signal.
The experiments take place in an open parking lot to ensure the GPS system
has a clear and unimpeded signal. The experimental setup consists in mounting
the IMU, the tilt sensor and the compass on a rigid plate that is fixed to a cart
(Figure 42) where the rest of the data acquisition system (without the ADCP) is
mounted. Once the sensors’ signals are acquired, decoded, and synchronized, they
are sent to the PC104 logger stack to be saved to the flash drive. These signals are
later post processed to find the best frequencies for the data fusions for the
evaluated sensors.
During each test the cart is initially stationary for at least two minutes. Because
of the lack of automatic motion control, the cart is then moved manually between
four spots on the ground that mark the corners of a square with 7.88m leg and the
corners pointing towards the four cardinal points. The use of industrial foam and
the choice of a cart with large wheels are among the precautions taken to minimize
the vibrations caused by the uneven ground. Three trajectories are selected: a
square path, a zigzag course and a circle. These trajectories are repeated at least
three times each at different speeds. The path of the trajectories, speed and
periodicity are selected to test the system’s ability to accurately measure the cart
motion.
Conventions used in this chapter are as follows:
The GPS position vector is denoted which is composed of , its north-
south component, , its east-west component, and , its vertical component.
Similarly, and represent the north-south and east-west velocity
components, respectively, and is the vertical component of the GPS velocity
vector, :
T
, , , (35)
T
, , . (36)
The determination of the data fusion frequency for the data fusion process is based
on 4 steps applied to each one of the three trajectories. The steps are as follows:
4.2 Data Acquisition System Lab Testing 49
1) processing of the raw data and selection of the data fusion frequency,
2) Observing the signal spectrum to validate the choice of the data fusion
frequency, 3) Observing the crosscorrelation of the low-passed merged and low-
passed GPS measurements to quantify their agreement with each other and Step4;
calculation of the standard deviation of the merged signal using a high-pass filter
to remove any motion contamination (Figure 43).
Fig. 43 Methodology used to find the data fusion frequency between IMU and GPS
measurement to recover full frequency estimate of the system’s position and velocity.
1. South 4. North
W e st W e st
0
-2
X gps[m]=>North
-4
-6
-8
-10
-12
2. South 3. North
East East
-8 -6 -4 -2 0 2 4
Y gps[m]=>East
The path is repeated four times in approximately three minutes. The first two
squares are each completed in approximately 57.8s and the last two in 34.35s. The
DGPS reflects the motion of the cart within its 3 meter accuracy. According to the
DGPS velocity data, the cart is moving at 0.55m/s during the first two square
paths and about twice as fast for the last two, at 0.93m/s. The second path,
following the same general square trajectory but proceeding in a zigzag pattern
between corners, is repeated four times 90.22s each with a total of a little over six
minutes, (Figure 45). The DGPS responded sharply to the sudden change of
direction of the cart. According to the DGPS velocity measurement, the cart
moves at approximately 0.39m/s.
-2
-4
X gps[m]=>North
-6
-8
-10
-12
-4 -2 0 2 4 6
Y gps[m]=>East
Fig. 45 Square path proceeding in a zigzag pattern between corners, as perceived by the
DGPS.
The last path follows a 39m perimeter circle which is repeated five times in
approximately six minutes. The fourth circle travelled has an ellipsoidal shape due
to the DGPS position’s error. The first three circles represent a 39.3m distance
travelled in 83s. The ellipsoidal path is 32.83m and is conducted in 64s. Finally,
the last circle takes 60.9s to travel 36.29m (Figure 46). The cart swerved less than
3 meters when attempting to manually recreate the same trajectory four times in a
row. The DGPS measurements reflect both the cart’s swerve and its 3 meters
accuracy range.
4.2 Data Acquisition System Lab Testing 51
-2
X gps[m]=>North
-4
-6
-8
-10
-12
-6 -4 -2 0 2 4 6
Y gps[m]=>East
The uneven track induced the data acquisition systems to tilt and those angles
are measured by the tilt sensor (Figure 47). Although pitch and roll applied to the
data acquisition system impacts the IMU accelerometers, these are taken into
account in the processing of the IMU acceleration measurement. As an example of
that impact, the error induced on the IMU acceleration by tilting is calculated for
the first trajectory test. The tilt sensor’s roll and pitch measurement of the square
path have a mean of -0.4° and -1.275° respectively, with a standard deviation of
± 1.44° and ± 0.93° respectively. At the maximum roll angle, 1.84°, it influences
the east component of the acceleration by -0.315m/s2 and at the maximum pitch
angle, 2.2°, the north component of the acceleration by 0.376m/s2. The tilt
sensor’s roll and pitch for the three trajectories are shown in Figure 47 and Table 6
presents their mean and standard deviation as well as the influence of the
maximum deviation of the data on the IMU east and north component of the
acceleration.
52 4 Motion Observation and Experimental Results
10 a
Roll [deg] 10 c
Roll [deg]
0 0
-10 -10
250 300 350 400 450 200 300 400 500 600
5 b 4 d
Pitch deg]
Pitch deg]
2
0 0
-2
-5 -4
-6
250 300 350 400 450 200 300 400 500 600
Time[s] Time[s]
10 e
Roll [deg]
-10
4 f
Pitch deg]
2
0
-2
-4
-6
200 300 400 500 600
Time[s]
Fig. 47 Roll and Pitch of the cart measured by the tilt sensor during the first trajectory ((a)
and (b)), the second trajectory ((c) and (d)) and the third trajectory ((e) and (f)).
Table 6 Mean and standard deviation of the tilt sensors’ roll and pitch as well as the
influence it could have on the IMU acceleration if not considered for the three trajectories
of the on shore test of the data acquisition system.
Trajectory
Square in zigzag Circle
Square
course
Mean [°] -0.4 -0.631 -0.6321
ROLL STD [°] ± 1.44 ± 1.253 ± 1.096
Influence on Ay [m/s2] -0.315 -0.3225 1.7281
Mean [°] -1.275 -1.156 -1.165
PITCH STD [°] ± 0.93 ± 0.946 ± 0.8169
Influence on Ax[m/s2] 0.376 0.356 0.3392
The study of the tilts concluded that the cart has rolled 0.55° ± 1.26°, and has
pitched 1.19°± 0.89° on average. The second part of the processing of the raw
sensor’s data is to observe the measurements in the frequency domain. For the
study of the signals in the frequency domain, the power spectral density (PSD) of
each signal is used. The IMU PSD accelerations are obtained for the three
trajectories and shown in Figure 48. The notation PSD Ax/Ay means the graph is
representing both the PSD of Ax and the PSD of Ay in two different colors.
4.2 Data Acquisition System Lab Testing 53
a b
5 6
4 5
PSD Ax / Ay
PSD Ax / Ay
4
3
3
2
2
1 1
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Frequency [Hz] Frequency [Hz]
c
4
PSD Ax / Ay
0 10 20 30 40 50 60
Frequency [Hz]
Fig. 48 PSD of the north component, in blue, and the east component, in red, of the IMU
acceleration during square trajectory (a), square path by processing in zigzag course (b) and
the circle trajectory (c).
Figure 48 shows significant spectral content for frequencies above 2Hz present
in the measurements. The Notation PSD Ax/Ay indicates that the plot has both the
PSD of Ax and the PSD of Ay represented. The scale is chosen to show the noise
starting at 2Hz and most likely coming from the vibration of the cart and from the
batteries located nearby the system. These high frequencies are most likely a
combination of valuable high frequency component of the IMU measurement and
noise from the vibration of the IMU on the cart and from the battery nearby the
data acquisition system. Among the noise is valuable high frequency information
on the IMU acceleration measurements essential for the data fusion process with
the GPS. No low-pass filtering can be applied to the data prior to the data fusion
process and high frequency noise is going to be relatively attenuated by the low-
pass filter applied at the last stage of the data fusion process. To be aware of the
signal to noise ratio of the IMU acceleration measurement, a first order
Butterworth low-pass filter with a cutoff frequency at 2Hz is applied to the signals
to remove the aforementioned noise at 2Hz and the result can be seen in Figure 49.
The plots (a), (d) and (g) [respectively (b), (e), (h) and (c), (f), (i)] shows the north
(respectively east and down) component of the acceleration in blue and the signal
54 4 Motion Observation and Experimental Results
after low-pass filtering the noise at 2Hz in red. The signal to noise ratio is
approximately 1 to 2. The Butterworth filter is selected because typical and an
Elliptic filter could also have been used. The filtering is only done for a better
understanding of the frequency distribution of the signal and is not included in the
data fusion process.
Ax [m/s2]
2 a
0
-2
250 300 350 400 450
Ay [m/s2]
2 b
0
-2
250 300 350 400 450
Az [m/s2]
4 c
2
0
-2
-4
250 300 350 400 450
Time[s]
Ax [m/s2]
2 g
Ax [m/s2]
2 d
0 0
-2 -2
200 300 400 500 600 200 300 400 500 600
Ay [m/s2]
Ay [m/s2]
2 e 2 h
0 0
-2 -2
200 300 400 500 600 200 300 400 500 600
Az [m/s2]
Az [m/s2]
2 f 2 i
0 0
-2 -2
200 300 400 500 600 200 300 400 500 600
Time[s] Time[s]
Fig. 49 Influence of frequencies above 2Hz on the IMU acceleration measurements for the
three trajectories of the on shore test of the data acquisition system.
Finally, the DGPS position, velocity and the IMU acceleration are studied in
the frequency domain to determine the complementary regions of the sensor. The
detection of the frequency peak corresponding to the cart’s motion, which appears
in each one of the signals spectrum, is the first step in observing the
complementary regions of the sensors. Figure 50 presents the spectrum of the
DGPS position signal (Figure 50.a), DGPS velocity signal (Figure 50.b) and of the
IMU acceleration (Figure 50.c) during the first trajectory (square path). The
frequency corresponding to the first two square paths is close to 0.015Hz and for
the last two squares close to 0.02Hz. Once the frequency peaks corresponding to
the cart’s motion are detected, around 0.02Hz in this example, the visualization of
where the DGPS and IMU measurements have most of their significant spectral
content is used to choose the data fusion frequency.
4.2 Data Acquisition System Lab Testing 55
4000
a
3000
1000
PSD Ax / Ay
30
3
20
2
10
1
0 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0 0.02 0.04 0.06 0.08 0.1
Frequency [Hz] Frequency [Hz]
Fig. 50 PSD of the DGPS position (a), DGPS velocity (b) and IMU acceleration (c) for the
first trajectory of the on shore test, following a square path. The blue signal corresponds to
the north component of the measurement and the red signal to the east component.
The DGPS has most of its significant spectral content around the peak of
interest (Figure 50.a and Figure 50.b) corresponding to the cart’s motion and
almost no significant spectral content at higher frequencies. On the other hand,
Figure 50.c shows how the IMU sensed the low frequency motion of the cart,
smothered by other frequencies, while still responsive over a frequency range
greater than the DGPS. This observation shows the importance of the IMU data
being filtered out of the region where the DGPS delivers accurate measurements
in order to avoid perturbations on the low frequency estimate of the merged signal.
To do so, a 3rd order high pass Butterworth filter with a cutoff frequency at 0.1Hz
(Nyquist frequency) is applied to the IMU acceleration measurement prior to the
data fusion process. The Butterworth filter is selected because typical and an
Elliptic filter could also have been used. The detection for the frequency peak
corresponding to the cart’s motion is applied for all the trajectories and results are
compiled in Table 7.
56 4 Motion Observation and Experimental Results
Table 7 Results from the peaks of frequency detection corresponding to the cart’s motion
for the three trajectories.
Square
Frequency peak corresponding to Square in
First two Last two Circle
the cart’s motion [Hz] zigzag course
paths paths
North Component 0.01489 0.02173 0.01074 0.01318
DGPS POSITION
East Component 0.015 0.02 0.01074 0.01318
DGPS North Component 0.01489 0.021 0.01074 0.01245
VELOCITY East Component 0.01489 0.021 0.01123 0.0127
IMU North Component 0.015 0.021 0.01074 0.012
ACCELERATION East Component 0.015 0.021 0.01074 0.012
The study of the spectrums for the other two trajectories leads to similar
observations as in Figure 50 where the DGPS has most of its significant spectral
content just after the peak of interest corresponding to the cart’s motion. These
observations suggest that the complementary regions of the sensors overlap
around 0.05Hz, which is used as the data fusion point.
4.2.2 Step 2: Validate the Choice for the Data Fusion Frequency
As aforementioned, the IMU acceleration measurement is high-pass filtered
prior to the data fusion process and the data fusion point is selected at 0.05Hz.
Figure 51 shows the diagram of the first data fusion process.
Fig. 51 Data fusion diagram between the IMU acceleration data and the DGPS velocity
measurements in order to obtain the enhanced velocity estimate.
The data fusion frequency validation is applied observing the signals though
key sequential steps, depicted in Figure 51. The frequency domain is selected for
that investigation. Figure 52 shows the results of this analysis for the north
component of the DGPS velocity and the north component of the IMU
acceleration during the square maneuver. The same process is applied for all the
signals and trajectories.
4.2 Data Acquisition System Lab Testing 57
40 40
30 30
20 20
10 10
0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
Frequency[Hz] Frequency[Hz]
50 50
PSD Vx gps,Scaled HP(Ax),Add
c d
30 30
20 20
10 10
0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
Frequency[Hz] Frequency[Hz]
Fig. 52 PSD at particular steps of the data fusion process between the DGPS north
component velocity and the IMU north component acceleration.
Figure 52 shows the Power Spectral Density (PSD) at key steps of the data
fusion process between the DGPS north component velocity and the IMU north
component acceleration. Figure 52a compares the DGPS north component
velocity (blue), the north component of the IMU acceleration (black), and the
same signal high-pass filtered (red). Figure 52b shows the DGPS north component
velocity (blue) next to the high-pass filtered north component of the IMU
acceleration (black), and the same signal scaled using the data fusion frequency
(red). Figure 52c shows the DGPS north component velocity (blue), the scaled
high-pass filtered IMU acceleration (black) and the addition of the two signals in
red. The addition of scaled high-pass filtered IMU acceleration and the DGPS
velocity is noted as a preemphasized signal. Finally Figure 52d shows the north
component of the merged velocity (red) compare to the DGPS north component
velocity (blue) and its addition to the scaled high-pass filtered IMU acceleration.
This study shows the high-pass filter applied to the raw acceleration
measurement has reduced the region of frequency where the DGPS velocity
spectrum has most of its significant spectral content (Figure 52.a) which allows
the merged signal to follow the DGPS velocity spectrum before the data fusion
frequency (Figure 52.d). Figure 52.b demonstrates the advantage of scaling the
58 4 Motion Observation and Experimental Results
2 a
Vx/y GPS,CSAx/y,Fused Vx/y
0
-2
-4
-6
-8
250 300 350 400 450
2
b
0
-2
-4
250 300 350 400 450
Time[s]
Fig. 53 Comparison in the time domain between the merged velocity (red) and the velocity
obtained by direct integration of the raw IMU acceleration signal (black). The blue signal is
the DGPS velocity measurement. The upper panel shows the north component of the signal
(a) and the lower, the east component (b)
Now that the merged velocity estimate of the data acquisition system is
available, the subsequent data fusion process applied is between the merged
velocity and the DGPS position measurement to obtain a full frequency measure
of the position estimate. The choice of the data fusion frequency is done in a
similar fashion to aforementioned and the same data fusion point at 0.05Hz is
selected. Figure 54 shows the diagram of the process of the second data fusion
process between the enhanced velocity signal and the DGPS position
measurement.
4.2 Data Acquisition System Lab Testing 59
Fig. 54 Data fusion diagram between the DGPS position measurement and the merged
velocity estimate obtained by fusing the IMU acceleration data and the DGPS velocity.
a 0.8 c
% of agreement (1=100%)
0.6
% of agreement (1=100%)
0.5
0.4
0 0.2
0
-0.5 -0.2
-0.4
2 4 6 8 10 12 14 2 4 6 8 10 12 14
4 4
x 10 x 10
b d
0.5 0.5
0 0
-0.5 -0.5
2 4 6 8 10 12 14 2 4 6 8 10 12 14
Sample Number 4
x 10 Sample Number 4
x 10
Fig. 55 Crosscorrelation (a) (respectively (b)) between the north, (respectively east)
component of the DGPS velocity and the north (respectively east) component of the merged
velocity estimates. Similarly, (c) (respectively (d)) corresponds to the crosscorrelation
between the north (respectively east) component of the DGPS position and the north
(respectively east) component of the merged position estimates.
Table 8 Estimates of the standard deviation of the merged velocity signal for the three
trajectories of the on shore data acquisition test.
The standard deviation of the enhanced velocity signal averages 0.83 cm/s and
the signal is used in the subsequent section for the correction of the ADCP data
when performing a mission at sea.
Chapter 5
At Sea Experiment of Data Acquisition System
This chapter presents the mission at sea conducted for the observation of the
motion data acquisition system measurements in the field as well as the collection
and correction of unreferenced ADCP data.
Both the motion data acquisition system and TRDI ADCP are installed on a test
vessel, the R/V Oceaneer IV, which performs a series of specifically chosen
maneuvers in open sea while the motion data along with the ADCP data are
simultaneously collected to be later post-processed. The mission is performed off
the southeast coast of Florida where the currents run predominately near shore in a
north-south direction with velocity ranging up to 1m/s (Figure 56). The Florida
Current receives its water from two main sources, the Loop Current and the
Antilles Current. The Loop current is the most significant of these sources and can
be considered the upstream extension of the Gulf Stream System [NOAA].
The ship maneuvers follows two different tracks: an L-shape track (going south
then east) and a straight line roundtrip track along the south-north direction. In the
first section of this chapter, results of the navigational data fusion for each
maneuver are presented. The second section examines the unreferenced ADCP
velocity profiles and their correction, by subtracting the vessel motion from the
measurements. The correction is performed on both its Beam coordinate frame,
C. Gelin: A Low-Cost, High Rate Motion Measurement System, NAMESS 1, pp. 61–84.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
62 5 At Sea Experiment of Data Acquisition System
where the data is recorded, and in its Earth Reference frame (North-East-Up
frame). The last section summarizes the results of the mission at sea.
0
a b
-100 1000
X gps[m]=>North
X gps[m]=>North
-200
500
-300
-400 0
-500
-500
-600
0 200 400 600 800 1000 0 50 100 150
Y gps[m]=>East Y gps[m]=>East
Fig. 57 Trajectory perceived by the DGPS during the first (a) and second (b) maneuver at
sea.
and after the data fusion point to follow the behavior of the merged signal. As
predicted, the figures demonstrate how the enhanced estimate of the vessel’s
velocity supports the DGPS spectra below the data fusion point and the IMU
acceleration measurement above it.
0
10
0 10
Vx GPS Vx GPS
Ax IMU Ax IMU
Vx Fused Vx Fused
-1 -1
10 10
0 b 0
d
10 10
Vy GPS Vy GPS
Ay IMU Ay IMU
Vy Fused Vy Fused
-1 -1
10 10
Frequency[Hz] Frequency[Hz]
Fig. 58 Close ups around the data fusion frequency, 0.05Hz, of the PSD of the velocity
measurement from the DGPS (blue), the acceleration estimate from the IMU (black) and
the enhanced estimate of the velocity obtained by data fusion (red).
The time series of the three components, north, east and down, of the enhanced
velocity estimate for the two maneuvers is plotted in Figure 59. The vessel travels
at 1.1m/s when heading south and 2.06m/s when heading east according to the
enhanced velocity signal, for the first maneuver (Figure 59 a, b and c). During the
second maneuver, (Figure 59 d, e, and f) the boat is traveling at 1.14m/s when
heading south and 2.94m/s when heading north. The enhanced velocity estimate,
updated every 1/128 s, leads to a more accurate measurement of the vessel’s
motion since measured over a larger range of frequency of motion when compared
to the GPS velocity measurement updated every 2 s. The enhanced velocity is later
removed from the ADCP measurements in order to obtain true current
measurements.
Vx [m/s]
Vx [m/s]
1 4 d
0 2
a 0
-1
0 200 400 600 800 1000 1200 0 500 1000
2 1.5
Vy [m/s]
Vy [m/s]
1 e
1 0.5
b 0
0 -0.5
0 200 400 600 800 1000 1200 0 500 1000
2
Vz [m/s]
Vz [m/s]
c 1 f
1
0 0
-1 -1
0 200 400 600 800 1000 1200 0 500 1000
Time[s] Time[s]
Fig. 59 Time series of the vessel’s enhance velocity measurement obtained by data fusion
with its north (east, down) component in blue (red, black) for the first maneuver (a, b, c)
and second maneuver (d, e, f)
64 5 At Sea Experiment of Data Acquisition System
According to the tilt sensor the boat rolled 1.66° and pitched 0.58° during the
first maneuver. It rolled 1.9° and pitched 0.56° during the second maneuver. These
tilts are taken into account in the navigational data fusion and for the corrections
of the ADCP data. Characteristic of the maneuvers at sea, like the distance
travelled, the headings and the enhanced velocity measurement of the ship are
now available to study and correct the water current measured by the ADCP.
Fig. 60 Diagram of the neceessary reference frame transformations to transform the vessell’s
enhanced velocity measureed by the data acquisition system into the ADCP Beaam
coordinate frame.
The results of correctiion of the ADCP data in the beam coordinate frame arre
presented for the first maaneuver, then the second maneuver. For each section, thhe
nd the velocity profiles along the 16 bins are presented.
currents at the first bin an
The last two bins of the water
w profile were discarded by the ADCP internal qualitty
algorithms and are represeented by a white space in the water profile figures.
5.2.1.1 Water Current Measured for the First Maneuver, L-Shape Track
Heading South Then
T East
The following presents, for the L-shape track, an estimate of the water currennt
measured at the first bin then observes the 16 bins of the ADCP velocity profille.
The ship’s velocity alon ng beams 2 and 3 (looking forward) as well as thhe
uncorrected and correctedd water current measure along beam 2 and 3 (Figure 61)
are computed, plotted an nd examined. The same procedure is done for the daata
along beams 1 and 4, lookking aft (Figure 62).
66 5 At Sea Experiment of Data Acquisition System
1.5 a
VShipBm2
1
0.5
0
-0.5
0 200 400 600 800 1000 1200
2 b
VBm2 bn1
1.5 c
1
0.5
0
-0.5
0 200 400 600 800 1000 1200
Time[s]
2 d
VShipBm3
e
1.5
VBm3 bn1
1
0.5
0
-0.5
0 200 400 600 800 1000 1200
VCurrent Bm3 bn1
f
1
-1
0 200 400 600 800 1000 1200
Time[s]
Fig. 61 Ship velocity, in blue, along beam 2 (a), and 3 (d) compare to the contaminated
measurement of the water current, in black, along beam 2 (b) and 3 (e), and to the true
water current, in red, along beam 2 (c) and 3 (f) during the first maneuver while the beams
2 and 3 are looking forward.
5.2 ADCP Unreferenced and Corrected Measurements 67
1 a
VShipBm1
0
-1
0 200 400 600 800 1000 1200
VBm1 bn1 1 b
-1
0 200 400 600 800 1000 1200
VCurrent Bm1 bn1
1 c
-1
-1
1 e
VBm4 bn1
-1
0 200 400 600 800 1000 1200
VCurrent Bm4 bn1
f
1
-1
0 200 400 600 800 1000 1200
Time[s]
Fig. 62 Ship velocity, in blue, along beam 1 (a), and 4 (d) compare to the contaminated
measure of the water current, in black, along beam 1 (b) and 4 (e), and to the true water
current, in red, along beam 1 (c) and 4 (f) during the first maneuver while the beams 1 and
4 are looking aft.
Table 9 Ship’s enhanced velocity measurement, uncorrected and corrected ADCP water
current measurement in beam coordinates, at the first bin, during the first maneuver
Beam Coordinates,
Ship’s velocity, ADCP uncorrected Corrected Water
1st maneuver, bin 1
[m/s] Water current, [m/s] current, [m/s]
L- Shape Track
Beam 2 0.2 then 0.13 0.46 then 0.38 0.26 then 0.25
Forward
Beam 3 0.27 then 0.62 0.39 then 0.37 0.12 then –0.25
Beam 1 -0.22 then -0.21 –0.55 then –0.42 –0.33 then –0.21
Aft
Beam 4 -0.29 then -0.71 –0.48 then –0.41 –0.19 then 0.30
68 5 At Sea Experiment of Data Acquisition System
Using Table 9 and basic trigonometric equations, the water current measured at
the first bin during the first maneuver is estimated at 4.74° North, 0.5m/s when the
ship is heading south and 7.5° North, 0.72m/s when heading east. The subsequent
part studies the ADCP velocity profiles to estimate the water current along
all bins.
There are a number of options for representing time series water current data
depending on the desired analyses. For this investigation of the ADCP velocity
profiles and for the two maneuvers, the Matlab ‘colormap plot’ is the tool used to
conceptualize the current velocity amplitudes. The direction and magnitude of the
water current is then calculated using basic trigonometric equations. Uncorrected
(Figure 63 and Figure 65) and corrected (Figure 64 and Figure 66) ADCP velocity
profiles along the beams 2 and 3 and uncorrected (Figure 67 and Figure 69) and
corrected (Figure 68 and Figure 70) velocity profiles along the beams 1 and 4 are
presented below for the 16 bins of the water profile of the first maneuver. The bin
15 and 16 contain only the ADCP flag for ‘bad data’, i.e. -32768, indicating the
limits of the ADCP range has been reached. Estimations of the water current are
compiled in Table 10.
1 .5
3
5 1
7
0 .5
Bin
9
0
11
-0 .5
13
15 -1
Fig. 63 Uncorrected ADCP velocity profile along beam 2, looking forward, during the first
maneuver going south then east
5.2 ADCP Unreferenced and Corrected Measurements 69
1 1 .2
3 1
0 .8
5
0 .6
7 0 .4
Bin
9 0 .2
0
11
-0 .2
13 -0 .4
-0 .6
15
Fig. 64 Corrected ADCP velocity profile along beam 2, looking forward, during the first
maneuver going south then east.
Figure 64 show the water current along beam 2 is of positive value across the
maneuver with a slight increase in value when the boat is heading east. Peaks in
current velocity occur along beam 2 when the boat is heading south.
1 .5
3
5
1
7
0 .5
Bin
11 0
13
-0 .5
15
Fig. 65 Uncorrected ADCP velocity profile along beam 3, looking forward, during the first
maneuver going south then east
70 5 At Sea Experiment of Data Acquisition System
1
1
3
5
0 .5
7
Bin
0
9
11
-0 .5
13
-1
15
Fig. 66 Corrected ADCP velocity profile along beam 3, looking forward, during the first
maneuver going south then east.
Figure 67 shows the water current along beam 3 has approximately the same
range of value but of opposite signs when the boat heads south (positive) and east
(negative). Peaks in current velocity mostly occur along beam 3 when the boat is
heading south.
3 0 .5
5
0
7
Bin
9 -0 .5
11
-1
13
15
-1 .5
200 400 600 800 1000 1200
E n s e m b le
Fig. 67 Uncorrected ADCP velocity profile along beam 1, looking aft, during the first
maneuver going south then east
5.2 ADCP Unreferenced and Corrected Measurements 71
0 .5
3
5
0
7
Bin
-0 .5
9
11
-1
13
-1 .5
15
Fig. 68 Corrected ADCP velocity profile along beam 1, looking aft, during the first
maneuver going south then east.
Figure 69 shows the water current along beam 1 is of negative value across the
maneuver with a slight increase in value when the boat is heading south. Peaks in
current velocity occur sporadically along beam 1 throughout the maneuver.
3 0 .5
5
0
7
Bin
9 -0 .5
11
-1
13
15 -1 .5
Fig. 69 Uncorrected ADCP velocity profile along beam 4, looking aft, during the first
maneuver going south then east
72 5 At Sea Experiment of Data Acquisition System
1
3
5
0 .5
7
Bin
9 0
11
-0 .5
13
15
-1
200 400 600 800 1000 1200
E n s e m b le
Fig. 70 Corrected ADCP velocity profile along beam 4, looking aft, during the first
maneuver going south then east.
Figure 70 shows the water current along beam 4 has a negative value when the
boat is heading south and a positive value when the boat heads east. Peaks in
current velocity mostly occur along beam 3 when the boat is heading east.
Looking at both of the corrected velocity profiles of the beams 2 and 3, looking
forward, the peaks in current velocity appears in dark red and mostly occurs when
the boat is heading south, i.e. when the water current is coming towards the
beams. Plotting the time series using the colormap method reveals the variations
in current magnitude according to direction and depth. Along the four beams, the
water current is found to be homogeneously distributed over the water column, i.e.
the coloring is mostly the same from the surface to the limit of the ADCP range.
Quantifications of the uncorrected and corrected water current along the four
beams and all bins are summarized in Table 10. In the table the abbreviation ‘Bm’
stands for ‘beam’.
5.2 ADCP Unreferenced and Corrected Measurements 73
Beam Coordinates,
Uncorrected ADCP Water Corrected Water profile,
1st maneuver, all bins
Profile, [m/s] [m/s]
L- Shape Track
Going South: Going South:
between 0.3 and 0.7 between 0.1 and 0.5
Bm 2
Going East: Going East:
between 0.2 and 0.5 between 0.1 and 0.3
Forward
Going South: Going South:
between 0.3 and 0.5 between 0.1 and 0.3
Bm 3
Going East: Going East:
between 0.3 and 0.5 between –0.3 and -0.1
Going South: Going South:
between -0.7 and -0.5 between -0.5 and -0.2
Bm 1
Going East: Going East:
between -0.5 and -0.3 between -0.3 and -0.1
Aft
Going South: Going South:
between -0.6 and -0.2 between -0.3 and -0.1
Bm 4
Going East: Going East:
between -0.5 and -0.3 between 0.2 and 0.4
The water current is estimated averaging the water current measurements along
each beam (Table 10) and using basic trigonometric equations. For the first
maneuver, the water current is estimated at 13° north, 0.72m/s when the ship is
heading south and 8.8° north, 0.64m/s when the vessel’s heading east.
The following section presents estimates of the water current during the second
maneuver, using the same method as the study of the first maneuver.
5.2.1.2 Water Current Measured for the Second Maneuver, Linear Track
Heading South Then North
The following presents, for the linear track, an estimate of the water current
measured at the first bin then observes the 16 bins of the ADCP velocity profile.
The ship’s velocity along beams 2 and 3 (looking forward) is computed, plotted
and examined as well as the uncorrected and corrected water current measure
along beam 2 and 3 (Figure 71). The same procedure is done for the data along
beams 1 and 4, looking aft (Figure 72).
74 5 At Sea Experiment of Data Acquisition System
2 a
VShipBm2
1
b
1
VBm2 bn1
-1
1
c
-1
0 200 400 600 800 1000 1200
Time[s]
2 d
VShipBm3
e
1
VBm3 bn1
-1
0 200 400 600 800 1000 1200
VCurrent Bm3 bn1
1
f
-1
0 200 400 600 800 1000 1200
Time[s]
Fig. 71 Ship velocity, in blue, along beam 2 (a), and 3 (d) compare to the contaminated
measure of the water current, in black, along beam 2 (b) and 3 (e), and to the true water
current, in red, along beam 2 (c) and34 (f) during the second maneuver while the beams 2
and 3 are looking forward.
5.2 ADCP Unreferenced and Corrected Measurements 75
1 a
VShipBm1
0
-1
-2
0 200 400 600 800 1000 1200
0.5 b
VBm1 bn1
0
-0.5
-1
-1.5
0 200 400 600 800 1000 1200
VCurrent Bm1 bn1
1 c
-1
0 200 400 600 800 1000 1200
Time[s]
1 d
VShipBm4
-1
-2
0 200 400 600 800 1000 1200
0.5 e
VBm4 bn1
0
-0.5
-1
-1.5
0 200 400 600 800 1000 1200
VCurrent Bm4 bn1
f
0.5
-0.5
0 200 400 600 800 1000 1200
Time[s]
Fig. 72 Ship velocity, in blue, along beam 1 (a), and 4 (d) compare to the contaminated
measure of the water current, in black, along beam 1 (b) and 4 (e), and to the true water
current, in red, along beam 1 (c) and 4 (f) during the second maneuver while the beams 1
and 4 are looking aft.
Figure 71 and Figure 72 demonstrate the influence of the ships’ motion (blue)
on the ADCP measurement of the current (black) and it is only when this
contamination is eliminated that one can interpret the water current measurement
(red). Table 11 compiles the data estimates.
Table 11 Ship’s velocity, uncorrected and corrected ADCP water current measurement in
beam coordinates, for the first bin, during the second maneuver
Beam Coordinates,
Ship’s velocity, ADCP uncorrected Corrected Water
2nd maneuver, bin 1
[m/s] Water current, [m/s] current, [m/s]
Straight Line Track
Bm 2 0.25 then 0.75 0.48 then 0.36 0.22 then –0.36
Forward
Bm 3 0.26 then 0.75 0.39 then 0.42 0.12 then -0.32
Bm 1 -0.24 then -0.65 -0.59 then –0.48 -0.32 then 0.16
Aft
Bm 4 -0.24 then –0.66 -0.51 then -0.53 -0.25 then 0.12
76 5 At Sea Experiment of Data Acquisition System
Using Table 11 and basic trigonometric equations, the water current measured
at the first bin during the second maneuver is estimated at 5.49° north, 0.65m/s
when the ship is heading south and 4.8° north, 0.68m/s when heading north. The
subsequent part studies the ADCP velocity profiles to estimate the water current
along all bins.
As for the first maneuver, the Matlab ‘colormap plot’ is the tool used to
conceptualize the current velocity amplitudes. The uncorrected and corrected
ADCP velocity profiles along the beams 2 and 3 and the uncorrected and
corrected ADCP velocity profiles along the beams 1 and 4 are observed for
the second maneuvers. The bin 15 and 16 contain only the ADCP flag for ‘bad
data’(-32768), indicating that the limits of the ADCP water profile range have
been reached. The estimation of the water current is compiled in Table 12.
The corrected velocity profiles of the beams 2 and 3, looking forward, shows
peaks in current velocity (dark red) mostly occurring when the boat is heading
south, i.e. when the water current is coming towards the beams while peaks in
current velocity appear on the beams 1 and 4, looking aft, when the boat is
heading north. Along the four beams, the water current is considered
homogeneously distributed over the water column although one can note slight
changes in the colors, compare to the first maneuver profiles, showing the water
current amplitude reducing with depth. The estimations of the uncorrected and
corrected water current along the four beams and all bins are summarized in
Table 12 where the abbreviation ‘Bm’ stands for beam.
Beam Coordinates,
Uncorrected ADCP Water Corrected Water
2nd maneuver, all bin
Profile, [m/s] profile, [m/s]
Straight Line Track
Going South: Going South:
between 0.35 and 0.6 between 0.1 and 0.3
Bm 2
Going North: Going North
between 0.25 and 0.6 between -0.5 and -0.2
Forward
Going South: Going South:
between 0.35 and 0.5 between 0.1 and 0.2
Bm 3
Going North: Going North
between 0.3 and 0.65 between -0.5 and -0.1
Going South: Going South:
between -0.65 and -0.5 between -0.4 and -0.2
Bm 1
Going North: Going North:
between -0.55 and -0.45 between 0.1 and 0.2
Aft
Going South: Going South:
between -0.65 and -0.35 between -0.4 and -0.1
Bm 4
Going North: Going North:
between -0.55 and -0.4 between 0.1 and 0.2
5.2 ADCP Unreferenced an
nd Corrected Measurements 777
The water current is esstimated averaging the water current measurements alonng
each beam (Table 12) an nd using basic trigonometric equations. For the seconnd
maneuver, the water currrent is estimated 1.25° north, 0.64m/s when the ship is
heading south and 3.2° noorth, 0.67m/s when the vessel’s heading north.
The following section presents the ADCP data correction in the North-East-U Up
Frame, this time for compparison purposes.
Fig. 73 Diagram of the neceessary reference frame transformations to transform the ADC
CP
data into the North-East-Up coordinate frame where the enhanced velocity measurement of
the vessel is available.
5.2.2.1 Water Current Measured, at the First Bin, in the NEU Frame for the
L-Shape Track and the Linear Track
This section quantifies the north and east component of the corrected water current
measurement at the first bin in the North-East-Up reference frame for the two
maneuvers. The north (respectively east) component of the ship’s velocity is
shown in Figure 74.a (Figure 75.a) for L-shape track and in Figure 74.d (Figure
75.d) for the linear track. The north (east) component of the uncorrected and
corrected currents, at the first bin, is presented in Figure 74.b (Figure 75.b) and
Figure 74.c (Figure 75.c) for the L-shape track and in Figure 74.e (Figure 75.e)
and Figure 74.f (Figure 75.f) for the linear track.
4
VShipN
VShipN
1
0 2
-1 a 0 d
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
2 b 4 e
VCurrent N bn1 VN bn1
1 2
0 0
-2
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
2 c 6 f
1 4
2
0 0
-2
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time[s] Time[s]
Fig. 74 Time series of the north component of the ship (blue), of the contaminated water
current measured by the ADCP in the middle of the first bin (black) and of the water
current resulting from its correction (red) in the NEU during the first (a, b an c) and the
second maneuver (d, e, and f).
2 1.5
VShipE
d
VShipE
1
1 0.5
a 0
0 -0.5
0 200 400 600 800 1000 1200 0 500 1000
1 2 e
b
VCurrent E bn1 VE bn1
VCurrent E bn1 VE bn1
0 1
-1 0
-1
-2
0 200 400 600 800 1000 1200 0 500 1000
c 2 f
1 1
0 0
-1
-1
0 200 400 600 800 1000 1200 0 500 1000
Time[s] Time[s]
Fig. 75 Time series of the east component of the ship (blue), of the contaminated water
current measured by the ADCP in the middle of the first bin (black) and of the water
current resulting from its correction (red) during the first (a, b an c) and the second
maneuver (d, e, and f).
5.2 ADCP Unreferenced and Corrected Measurements 79
The boat changes direction after 600s for the first maneuver and 630s for the
second maneuver. The impact can be seen on the east component of the data
(Figure 75a and Figure 75b) where there is a temporary inversion in the velocity
sign. These brief reversals are due to the ship’s dynamic response to waves.
Another perturbation on the ADCP data can be noted after 950s on the second
maneuver (Figure 74 and Figure 75 d, e and f). Each figure is accounted for and
the data estimations are presented in Table 13 for the first and second maneuver.
Table 13 Ship’s velocity, uncorrected and corrected ADCP water current measurement in
North-East-Up coordinates, for the first bin, during the first and second maneuver
The water current measured has an estimated 0.93m/s north component and a
0.14m/s east component (Table 13). Hence the water current measured, at the first
bin, in the NEU frame and on average over the two maneuvers, is estimated at 8.6°
north, 0.94m/s. The next part examines the ADCP velocity profiles to estimate the
water current along all bins.
The next step is to observe and quantify the north and east component of the
uncorrected (Figure 76, Figure 78, Figure 80 and Figure 82) and corrected (Figure
77, Figure 79, Figure 81 and Figure 83) ADCP velocity profiles along all bins for
the two maneuvers.
There are a number of options for representing time series water column
current data, depending on the desired analyses. For this investigation of the
ADCP velocity profiles and for the two maneuvers, the Matlab ‘colormap plot’ is
the tool used to conceptualize the current velocity amplitudes. The direction of the
water current is then calculated using basic trigonometric equations. Bin 15 and 16
contain only the ADCP flag for ‘bad data’, i.e. -32678, indicating that the limits of
the ADCP range have been reached.
80 5 At Sea Experiment of Data Acquisition System
2
2.5
4
2
6 1.5
8 1
Bin
0.5
10
0
12
-0.5
14
-1
16
-1.5
200 400 600 800 1000 1200
Ensemble
Fig. 76 Uncorrected north component of the ADCP velocity profile during the first
maneuver of the mission at sea, creating an L-shape track going south then east.
2
2.5
4 2
6 1.5
8 1
Bin
0.5
10
0
12
-0.5
14
-1
16
-1.5
200 400 600 800 1000 1200
Ensemble
Fig. 77 Corrected north component of the ADCP velocity profile during the first maneuver
of the mission at sea, creating an L-shape track going south then east.
Figure 77 shows the north component of the water current is of positive value
across the maneuver with a peak in current velocity when the boat changes
directions.
5.2 ADCP Unreferenced and Corrected Measurements 81
2
1
4
0.5
6
0
8
-0.5
Bin
10
-1
12
-1.5
14
-2
16
Fig. 78 Uncorrected east component of the ADCP velocity profile during the first maneuver
of the mission at sea, creating an L-shape track going south then east.
2
1
4
0.5
6
0
8
-0.5
Bin
10
-1
12
-1.5
14
-2
16
Fig. 79 Corrected east component of the ADCP velocity profile during the first maneuver
of the mission at sea, creating an L-shape track going south then east.
82 5 At Sea Experiment of Data Acquisition System
4 4
3
6
2
8
Bin
1
10
0
12
-1
14
-2
16
Fig. 80 Uncorrected north component of the ADCP velocity profile during the second
maneuver of the mission at sea, following a straight line track going south then north.
4 4
3
6
2
8
Bin
1
10
0
12
-1
14
-2
16
Fig. 81 Corrected north component of the ADCP velocity profile during the second
maneuver of the mission at sea, following a straight line track going south then north.
Figure 81 shows the water current is of positive value across the maneuver with
a decrease in value when the boat is changing direction.
5.2 ADCP Unreferenced and Corrected Measurements 83
2
2
4
1.5
6
1
8 0.5
Bin
10 0
12 -0.5
14 -1
16 -1.5
Fig. 82 Uncorrected east component of the ADCP velocity profile during the second
maneuver of the mission at sea, following a straight line track going south then north.
2
2
4
1.5
6
1
8 0.5
Bin
10 0
12 -0.5
14 -1
16 -1.5
Fig. 83 Corrected east component of the ADCP velocity profile during the second
maneuver of the mission at sea, following a straight line track going south then north.
The figures are analysed individually and estimations of the water current
measurements are presented in Table 14.
84 5 At Sea Experiment of Data Acquisition System
The average water current measured from the two maneuvers is composed of a
longshore current in the northerly direction, estimated at 1.07m/s and a weaker
offshore current of 0.1m/s (Table 14). The water current measured, observing all
bins and averaging the two maneuvers, is estimated at 5.3° north, 1.07m/s in the
NEU frame.
C. Gelin: A Low-Cost, High Rate Motion Measurement System, NAMESS 1, pp. 85–92.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
86 6 Conclusion
The data fusion techniques developed in this work combine the complementary
outputs of sensors measuring a related state to eliminate the drift of integrating the
measurement , while increasing the rate and resolution of where
and are measurements of some state output by two distinct
sensors (position sensor and velocity sensor for example). The pre-emphasized
signal, , is obtained by summing the signal with the derivative signal
, such that (Mudge and Lueck 1994):
, (38)
Ω
where the scaling factor Ω , denoting the cutoff frequency, is a real positive
constant. The choice of Ω is determined by the characteristics of the
complementary region of the two sensors. For frequencies that are small compared
to the cutoff frequency (Ω Ω , the signal portion of the spectrum comes
predominantly from , while for Ω Ω , the signal is predominately from
. The enhanced version of the signal is then obtained by
convolving with a single-pole, low-pass filter. The enhanced signal of
the signal contains low-frequency information from the sensor measuring
, and the high-frequency information from the sensor measuring .
The first data fusion method is applied in finding the Euler angles. Considering
the characteristics of each instrument, a data acquisition system is developed that
synchronously decodes data from all the instruments and converts them into a
consistent format using the Euler angles, which are not directly measured. The
Euler angles, , are obtained by fusing the low-frequency Euler angles, , and
, calculated from the tilt measurements and , and the compass heading, ,
with the high-frequency IMU angular rates, , , T.
Fig. 84 Diagram of the data fusion IMU / TCM2/ Tilt sensor to obtain Euler angles, .
The experiment applied to find a suitable data fusion frequency between the
IMU, TCM2 and tilt sensor to calculate the Euler angles consists in mounting the
sensors on the same rigid plate and turning the sensor system simultaneously
Conclusion 87
about multiple axes and rotation at different speed. The IMU rate gyros are found
to have a low drift rate and the data fusion point is chosen to be at 1/30Hz. This
data fusion frequency is selected so that at frequencies lower than the cutoff
frequency, the tilt sensors and compass heading, provide accurate and stable
measures of the Euler angles, , and and at frequencies above the cutoff
frequency, the rate gyros in the IMU provide accurate measures of the Euler rates,
, and .
The estimated Euler angles are then used to convert the IMU acceleration from
body-fixed frame, , , T , to the NED frame, , , T , where
gravity is removed. The enhanced velocity of the ship, , is obtained by directly
fusing the high frequency (128Hz) acceleration measurement from the IMU, ,
and the low frequency (0.5Hz) velocity obtained from the speed and course
T
overground output from the GPS, , , . The position is then
T
obtained by fusing the enhanced velocity, , , , with the latitude and
longitude measured with the GPS. Two experiments are conducted to choose the
best data fusion point (Ω ) to be later used to obtain the full frequency measure of
the ships velocity, , and its position, .
The first experiment investigates the properties of the vertical NED
acceleration and the different methods available to obtain the merged vertical
velocity, and the merged position, . The experiment takes place in a machine
shop and consists of mounting the IMU, the tilt sensor and the TCM2 compass on
a level plate. The plate is leveled and tethered to the extremity of a 1.03m rigid
lever. The middle of the lever is attached to a gearbox that is attached to a rotating
engine. The extremity of the lever describes circular trajectories of 0.515m radius
at different speeds. The test consists of six sets of vertical roundtrip periods of
approximately 5, 10, 15, 20, 25 and 35 s, each lasting about 10 minutes. The
speeds are manually set using the speed variator of the rotating engine.
The IMU, which is assumed accurate only at high frequencies, is merged with a
null signal at low frequency and a data fusion point at 1/100Hz is found to be the
best compromise to obtain a full frequency measure of the vertical velocity. The
vertical velocity is then merge with a null signal to obtain the full frequency
measure of the vertical position signal and a cutoff frequency of 1/50Hz is found
to be the best choice for that data fusion method.
The second experiment investigates the properties of the data acquisition
system on shore, without the ADCP in order to find the best data fusion points of
the horizontal velocity and position signals. The experiments takes place in an
open parking lot to ensure the GPS system has a clear and unimpeded signal. The
experimental setup consists in mounting the IMU, the tilt sensor and the compass
on a rigid plate that is fixed to a cart where the rest of the data acquisition system
(without the ADCP) is mounted. Because of the lack of automatic motion control,
the cart is moved manually between four spots on the ground that mark the corners
of a square with 7.88m legs with the corners pointing towards the four cardinal
points. Three trajectories are selected: a square path, a zigzag course and a circle.
These trajectories are each repeated at least three times at different speeds. The
path of the trajectories, speed and periodicity are selected to test the system’s
ability to accurately measure the cart motion.
88 6 Conclusion
The first data fusion process, involving the IMU acceleration and the GPS
velocity measurement leads to a full frequency measure of the velocity
measurement. Pre-filtering of IMU data are found to be necessary before the data
fusion process (Figure 85) and the experiment observations suggest the
complementary region of the sensors intersect around 0.05Hz, which is used as the
data fusion point. This data fusion point is selected so that at frequencies lower
than 0.05Hz the GPS provide an accurate measure of the velocity of the system,
and at frequencies above 0.05Hz the IMU provides an accurate estimation of the
velocity.
Fig. 85 Diagram of the data fusion between the IMU acceleration data and the DGPS
velocity measurements in order to obtain the enhanced velocity estimate.
The subsequent data fusion process applied is between the merged velocity and
the DGPS position measurement to obtain a full frequency measure of the position
estimate. The choice of the data fusion frequency is done in a similar fashion to
aforementioned and the same data fusion point at 0.05Hz is selected. Figure 86
shows the diagram of the process of the second data fusion process between the
enhanced velocity signal and the DGPS position measurement.
Fig. 86 Diagram of the data fusion process between the DGPS position measurement and
the merged velocity estimate obtained by fusing the IMU acceleration data and the DGPS
velocity.
Conclusion 89
Table 15 Estimates of the standard deviation of the merged velocity signal for the three
trajectories of the on shore data acquisition test.
The standard deviation of the enhanced velocity signal averages 0.83 cm/s
(< 1cm/s) and the signal is used for the correction of the ADCP data when
performing a mission at sea. The mission at sea is conducted for the observation of
the motion data acquisition system measurements in the field as well as the
collection and correction of unreferenced ADCP data. Both the motion data
acquisition system and TRDI ADCP are installed on a test vessel, the R/V
Oceaneer IV, which performs a series of specifically chosen maneuvers in open
sea while the motion data along with the ADCP data are simultaneously collected
to be later post-processed. The mission is performed off the southeast coast of
Florida where the currents run predominately near shore in a north-south direction
with magnitudes ranging up to 1m/s. The ship maneuvers follows two different
tracks: an L-shape track (going south then east) and a straight line roundtrip track
along the south-north direction.
For this mission, the ADCP is a 600 kHz Teledyne RDI Broadband Workhorse
Sentinel. Its reference beam, beam 3, is mounted 45o counter-clockwise from the
90 6 Conclusion
centerline of the ship in order to increase noise rejection and the effective ADCP
measured velocity by a factor of 1.4. As a result, beams 2 and 3 are pointing
forward, and beams 1 and 4 are pointing aft. The water profile will be composed
of 16 bins, each 4m in length. A default blanking distance of 88cm is used in order
to avoid measuring currents when the ADCP is ringing; resultantly the center of
the first bin is located 5.05m away from the ADCP which puts the center of the
last bin 65.05m away from the ADCP.
The first maneuver, L-shape track, is performed by heading south for 623.56m
at approximately 1.04m/s and then east for 1274.8m at approximately 2.04m/s
(according to the GPS measurements). The trajectory of the boat exhibits a slight
drift to the east when heading south and a more noticeable drift to the north when
heading east. This indicates the presence of a water current, as expected, mainly
along shore in the south-north direction with a secondary transverse east
component. The water currents’ influence on the vessel’s motion can also be
observed during the second maneuver, where the straight line maneuver goes
687.6m south-east at approximately 1.07m/s, and then goes 1988m north-east at
approximately 2.92m/s.
The water current is quantified when observing the corrected ADCP
measurements. The correction of the ADCP data is performed in two different
reference frames for comparison purposes. The first correction occurs in the
ADCP radial beam coordinate frame, which allows us to manipulate ADCP data
in its rawest form, i.e. no internal ADCP corrections applied. The second
correction occurs in the North-East-Up Frame, the Earth coordinate frame of the
ADCP. In addition to looking at the ADCP velocity profiles, the raw velocity at
the first bin, where the water current is its strongest in the middle of the bin (~5m
depth relative to the ADCP), is compared to the corrected ADCP velocity and the
merged ship velocity. The estimates of the water current for the all missions are
compiled in Table 16 and represented using Google earth in Figure 87.
Table 16 Summary of water current estimates obtain by correcting the ADCP data during
the two maneuvers at sea.
Water current
Reference Water current
Maneuver Bin direction
frame magnitude [m/s]
[°N]
L-shape Bin 1 4.74° then 7.5° 0.5 then 0.72
Beam
track All Bins 13° then 8.8° 0.72 then 0.64
going south
then east Bin 1 8.6° 0.94
Earth
Straight line
track All Bins 5.3° 1.07
going south Bin 1 5.49° then 4.8° 0.65 then 0.68
then north Beam
All Bins 1.25° then 3.2° 0.64 then 0.67
Conclusion 91
Fig. 87 Google Earth visualization of the mission at sea with the two maneuvers, first goes
south-east then south-north
The correction of the ADCP data can occur accurately since the standard
deviation of the enhanced velocity of the ship (0.83 cm/s) is lower than the
standard deviation of the water current measurement from the ADCP (Table 15).
Table 17 Estimated standard deviation of the ADCP velocity during the first and second
maneuver at sea in correlation to the bin size, using the standard deviation of the error
velocity
This book has presented a low-cost, high rate motion measurement system
developed for an unmanned surface vehicle with underwater navigation and
oceanographic applications. The system integrates a motion measurement package
to aid in navigation and control while correcting data from an Acoustic Doppler
Current Profiler (ADCP), providing oceanographic measurements.
92 6 Conclusion
Recommendations
Another test at sea as well as a complete calibration of the data acquisition system
are the first two recommendations. An integration of the system for different
control system is also preferable as well as a formal design method for the data
fusion process so we can determine the optimal filter shape. Finally a fuzzification
of the data fusion process is suggested.
References
[1] Leonessa, A., Beaujean, P.-P., Driscoll, F.R.: Development of a small, multi-purpose,
autonomous surface vessel. Florida Atlantic University, Department of Ocean
Engineering (2002)
[2] Sousa, J., Cruz, N., Matos, A., Lobo Pereira, F.: Multiple AUVS for coastal
oceanography. In: OCEANS 1997, MTS/IEEE Conference Proceedings, October 6-9,
vol. 1, pp. 409–414 (1997)
[3] Bane, G., Ferguson, J.: The evolutionary development of the military autonomous
underwater vehicle. In: Proceedings of the 5th International Symposium on
Unmanned Untethered Submersible Technology, vol. 5, pp. 60–88 (June 1987)
[4] Grenon, G., An, E., Smith, S., Healey, A.: Enhancement of the Inertial Navigation
System for the Morpheus Autonomous Underwater Vehicles. IEEE, Journal of
Oceanic Engineering 26(4) (October 2001)
[5] Vickery, K., Sonardyne, Inc.: Acoustic Positioning Systems, a practical overview of
current system. In: IEEE, Proceedings of the Autonomous Underwater Vehicle (1998)
[6] Babb, R.J.: Navigation of unmanned underwater vehicles for scientific surveys. In:
Proceedings of the Symposium on Autonomous Underwater Vehicle Technology,
AUV 1990, June 5-6, pp. 194–198 (1990)
[7] Rayes, R.: Characterization study of the Florida current at 26.11 North latitude, 79.50
West longitude for ocean current power generation. Thesis submitted to the faculty of
the College of Engineering, Florida Atlantic University, Boca Raton (May 2002)
[8] BEI MotionPak Low Cost Multi-Axis Inertial Sensing System technical manual.
Systron Donner Inertial Division (1998)
[9] TCM2 Electronic Compass Module – User’s Manual. Precision Navigation, Inc. (July
2003)
[10] Garmin GPS 76 owner’s manual and reference guide, GARMIN Corporation (2001)
[11] Navigator ADCP/DVL Technical Manual. RD Instruments, Second Edition for
Broadband ADCPs, P/N 951-6069-00 (1996)
[12] Shafer, S.A., Stentz, A.: An Architecture for Sensor Fusion in a Mobile Robot. IEEE
(1986)
[13] Luo, R.C., Kay, M.G.: Multisensor Integration and Fusion in Intelligent Systems.
IEEE SMC-19(5) (1989)
[14] Cvetanovs, J.K.: Autonomous Submersible Robot: Sensor Characterization and
Testing. RSL Australian National University (2000)
[15] Gustafson, E.I.: A Post-Processing Kalman smoother for Underwater Vehicle
Navigation. FAU Ocean Engineering Dept. (2001)
[16] Welch, G., Bishop, G.: An Introduction to the Kalman Filter. University of North
Carolina, Department of Computer Science (2001),
http://www.cs.unc.edu/~welch/kalman/
94 References
[17] Chaumet-Lagrange, M., Loeb, H., Ygorra, S.: Design of an Autonomous Surface
Vehicle (ASV). University of Bordeaux I, France, Automatic and production
Laboratory, IEEE (1994)
[18] Manley, J.E.: Development of the Autonomous Surface Craft ‘ACES’. Massachusetts
Institute of Technology, Department of Ocean Engineering, Sea Grant College
Program, Cambridge MA 02139, IEEE (1997)
[19] CARAVELA Development of a Long-Range Autonomous Oceanographic Vessel,
Dynamic Systems and Ocean Robotics lab (DSOR) (1998-2000)
[20] Advanced System Integration for Managing the Coordinated Operation of Robotic
Ocean Vehicles (ASIMOV). ISR-IST, Lisbon, Portugal, ORCA Instrumentation,
France; System Technologies, United Kingdom; ENSIETA, France (1998-1999-2000)
[21] Oliveira, P., Pascoal, A., Rufino, M., Sebastiao, L., Silvestre, C.: The DELFIM
Autonomous Surface Craft. Report (December 1999)
[22] Oliveira, P., Pascoal, A., Kaminer, I.: A Nonlinear Vision Based Tracking System for
Coordinated Control of Marine Vehicles. IST/DEEC, Lisbon, Portugal, Naval
Postgraduate School, Monterey, USA (2002)
[23] Mudge, T.D., Lueck, R.G.: Digital Signal Processing to Enhance Oceanographic
Observations. Journal of Atmospheric and Oceanic Technology 11(3) (June 1994)
[24] Fossen, T., Lane, B.: Guidance and Control of Ocean Vehicles. John Wiley and Sons,
Ins., England (1994)
[25] Etkin, B.: Dynamics of Atmospheric Flight. Wiley, New York (1972)
[26] Driscoll, F.R., Lueck, R.G., Nahon Retkin, M.: The motion of a deep-sea remotely
operated vehicle system, Part 1: Motion observations. Ocean Engineering 27, 29–56
(2000)
Appendix A - Native Output of the Instruments
1. GPS
The native representation of the GPS is of NMEA output format with the
following NMEA messages available:
2. COMPASS
The TCM2 standard output format is of NMEA format:
$C<compass>P<pitch>R<roll>
Appendix B Setup and Acquisition
of the ADCP
Base
DLAB Read/Write Abr. Register Name
Address
=0 Write - Transmitter Holding Buffer
+0 =0 Read - Receiver Buffer
=1 Read/Write - Divisor Latch Low Byte
=0 Read/Write IER Interrupt Enable Register
+1
=1 Read/Write - Divisor Latch High Byte
Interrupt Identification
- Read IIR
+2 Register
- Write FCR FIFO Control Register
+3 - Read/Write LCR Line Control Register
The data, preceded by the ID code 7F7F, contains header data. The fixed and
variable leader data is preceded by ID codes 0000 and 8000.