2010 8th IEEE International Conference on ThA2.

5
Control and Automation
Xiamen, China, June 9-11, 2010

Accuracy Analysis of DVL/IMU/Magnetometer Integrated

Navigation System using Different IMUs in AUV
Yanrui Geng, Ricardo Martins, Joao Sousa, Member, IEEE

flux-gate compass. Aided with DVL, a DR system can

Abstract— This paper presents an integrated navigation system determine the AUV´s attitude, velocity and position [4].
for autonomous underwater vehicles which is based on a FEUP has been engaged in the research on underwater
Strapdown inertial navigation system (SDINS) accompanying navigation systems for more than 10 years. However, the
Doppler velocity log (DVL), pressure sensor and magnetometer. previous navigation models of AUV were almost based on
The paper focuses on the performance analysis of the bias of LBL system. When cost, space, and weight restraints of
inertial sensors and the error of position using different level AUV become stringent, other solutions need to be used. The
IMU. An extended Kalman filter is employed to estimate the
purpose of this paper is to estimate the feasibility of using
bias of the inertial sensors and then give the ultimate error of
position in about one hour campaign. Numerical simulation was DVL and some low-cost IMU system. Although several
presented and the result shows that the performance of AUV solutions including many redundant navigation sensors have
mainly depends on the precision of DVL and magnetometer. been presented in many papers, the minimal and cheapest set
The simulation also shows that the performances of the position of sensors will be considered to achieve maximum
accuracy with different IMUs are almost same. navigation accuracy in our project for the purpose of
reducing the volume and cost of AUV as possible.
I. INTRODUCTION The paper presents an integrated underwater navigation
system for AUV using inertial sensors with the aid of DVL,
F For decades, GPS and INS are the standard navigation
systems which are widely used in surveying or service-
and-maintenance applications to require the most
pressure sensor and magnetometer. Using this set of sensors,
the proposed navigation system can estimate all of the bias
accurate navigational information [7]. However, underwater of inertial sensors and give us a pretty good navigation
navigation requires different kinds of sensors than common performance with a small drift position error. The object of
in airborne or land vehicle navigation applications due to the this paper is to evaluate the performances of different level
disability of GPS signals in the water [6]. IMUs in AUV, and to choose the eventually solution in real
Conventionally, the acoustic long-base-line (LBL) system, application according to this analysis.
which can precisely measure the time-of-flight of sound The paper is organized as follows: Section 2 defines a
waves through water, is used as a solution in AUV scheme of a hybrid navigation system which composes of
applications [8]. But this kind of system has a limitation that IMU, DVL, depth and heading sensors. A measurement
the beacon array should be mounted on the ocean bottom or model for the integrated navigation system is designed to
fixed to a mooring. Furthermore, the position precision of implement an extended Kalman filter. Section 3 gives the
LBL may be deteriorated by the incorrect assumed sound simulation results of the method. To testify the correctness of
speed profile, especially in shallow water, and also by the algorithm, numerical simulations were conducted using the
motions of the beacons. character of various IMU, from tactical grade IMU to a
MEMS one. Finally section 4 presents the conclusion of the
The Doppler Velocity Log (DVL), emitting pulses from four paper.
transducers, can measure the frequency shift of the reflected
pulses to determine the relative velocity between the II. INTEGRATED NAVIGATION SYSTEMS IN AUV
transducer head and the reflecting surface along each beam
direction. Due to its accuracy and convenience, DVL, aided A. Error model of Strapdown INS
with some dead reckoning (DR) systems, are becoming the A basic set of error model of Strapdown INS for the
most popular solution in the navigation application of AUV integrated navigation system normally only include the
[1], [2], [3]. navigation parameters, the accelerometer and gyroscope
Usually, a dead reckoning system includes a set of three error states such as follows:
orthogonal accelerometers, three orthogonal gyros and a X = [φ E , φ N , φU , δv E , δv N , δvU , δL, δλ , δh,
ε x , ε y , ε z , ∇ x , ∇ y , ∇ z ]T (1)
Y.R. Geng is with the Department of Electrical and Computer
Engineering, University of Porto, Portugal. (phone: 351- 225-081-615 ; fax
351-225-081-443; e-mail:[email protected])
Ricardo Martins is with the Department of Electrical and Computer
Engineering, University of Porto, Portugal. (e-mail: [email protected])
J. Borges de Sousa is with the Department of Electrical and Computer
Engineering, University of Porto, Portugal. (e-mail: [email protected])

978-1-4244-5196-8/10/$26.00 ©2010 IEEE 516

Authorized licensed use limited to: Indian Institute of Technology Patna. Downloaded on May 03,2023 at 07:31:57 UTC from IEEE Xplore. Restrictions apply.
 ThA2.5

where φ E , φ N , φU are the attitude errors; δv E , δv N , δvU are the C. Kalman filter
velocity errors; δL, δλ , δh are the position errors; ε x , ε y , ε z The IMU/DVL/Pressure Sensor/Magnetometer integrated
navigation system can be modelled as equation (2), and the
are the gyro constant drifts, and ∇ x , ∇ y , ∇ z are the
equations (11) to (15) are the detail of the model.
accelerometer biases.
The error state equation is  φ&  Fφφ Fφv Fφr − Cbn 0   φ   wε 
      
X& (t ) = F (t ) X (t ) + G(t )W (t ) (2) δv& Fvφ Fvv Fvr 0 Cbn  δv w∇ 
δr& =  Frφ Frv Frr 0 0  δr  +  0  (11)
The detail of equation (2) is shown in equations (3-7), and       
the attitude error equation is as follows  ε&   0 0 0 0 0  ε   0 
φ& = φ × ωinn + δωinn − Cbnωibb − ε n ∇ &  0   ∇   0 
(3)   0 0 0 0
The velocity error equation is
δv& n = −φ × f n + Cbn f b The measurement equation is as follow
(4)
( ) ( )
+ δv n × 2ωien + ωenn + v n × 2δωien + δωenn + ∇ n z ( k ) = H ( k ) x ( k ) + v( k ) (12)
And the error equation of position is The detail of measurement are shown in equations (13)
and (14)
 δvn vn 
 − δh   ψ INS − ψ Mag 
δL&   Rm + h (Rm + h )2
 z (k ) = d INS
n
− d Prn essure  (13)
 &   δve ve vn sec L 
δλ =
  R +h sec L + δL tan L sec L − δh  v INS
b b

δh&   n Rn + h (Rn + h)2  − v DVL
   δvu 
   01×2 1 01×3 01×3 01×3 01×3 

H (k ) =  01×3 01×3 0 0 1 01×3 01×3 
(5) (14)
The error of gyros and accelerometers of IMU can be − Cn (v×) Cnb
b
0 01×2 01×3 01×3 
modelled as a random constant with white noise. So the
output errors of inertial sensors can be expressed as The discrete time implementation of the Kalman filter
ε g = ε gf + wg (6) requires a discrete time state propagation matrix φk / k −1 , and a
∇ a = ∇ af + wa (7) discrete time process noise covariance matrix, Qk −1 , where
where ε g is the gyro drift and ∇ a is the accelerometer drift. φ k / k −1 = exp( Fk ∆t ) and Qk −1 = GQGT ∆t .
ε gf and ∇ af are random constants. wg and wa are
assumed Gaussian white noise which have zero mean and Basically, the Kalman filtering estimation algorithm
the error variances Q g and Qa , respectively. comprises two steps, namely prediction and updating with
external measurements. The main Kalman filtering equations
B. Measurement Equations are given below [5]:
The measurements of yaw, depth and body velocities Prediction:
provided by magnetometer, pressure sensor and DVL xˆ k / k −1 = φ k / k −1 xˆ k −1 (15)
compose the Kalman filter measurement equation. Pk / k − 1 = φ k / k − 1 Pk − 1φ kT / k − 1 + Q k − 1 (16)
zψ = ψ INS − ψ mag = δψ = φU (8)
Updating:
n
z d = d INS − d Prn essure = δh (9) K k = Pk / k − 1 H T
k [ H k Pk / k − 1 H k + R k ]−1 (18)
b b b
zv = v INS −v DVL = δv xˆ k = xˆ k / k − 1 + K k [ z k − H k xˆ k / k − 1 ] (19)
= Cˆ nb vINS
n
− Cnb vDVL
n
= Cnb (I + φ ×)vINS
n
− Cnb vDVL
n P k = ( I − K k H k ) Pk / k − 1 (20)
( )
= C nbδv n − C nb v n × φ n (10) where xˆ k / k − 1 is the predicted state vector; Pk / k − 1 is the
Where variance matrix for xˆ k / k − 1 ; K k is the gain matrix; x̂ k is
b the estimation of filtering; and Pk is its variance matrix.
vINS velocity of INS in body frame
v b
velocity of DVL in body frame With a Kalman filter, a schematic of the navigation system is
DVL
n
shown in Fig 1.
d INS depth of INS in navigation frame
n
d depth of pressure sensor in navigation frame
Pr essure

ψ INS heading of INS

ψ mag heading of magnetometer

517

Authorized licensed use limited to: Indian Institute of Technology Patna. Downloaded on May 03,2023 at 07:31:57 UTC from IEEE Xplore. Restrictions apply.
 ThA2.5

Magnetometer 5Hz 2.0deg

Table III Specifications of aided sensors
Sensor Update Random Bias
IMU
Rate Noise Error
Type
(std)
ωibb Accelerometer 100Hz 0.5e-3 g 1.0e-3 g
HG1700
Gyro 100Hz 0.2 deg/hr 1.0
level
deg/hr
f ibb
Micro Accelerometer 100Hz 0.2e-2 g 1.0e-2 g
Strain Gyro 100Hz 0.2 1.0
level deg/sec deg/sec
Table IV Specifications of IMU in simulation

The kinematic trajectory used in the simulation started at

north latitude 40 degrees, east longitude 116 degrees and
ended at north latitude 40.01degrees, west longitude 116.01
degrees. The trajectory altitude was zero m. The initial
velocity was 1.5 m/s with heading north 0 degrees, and the
maximum velocity is no more than 3.5 m/s which is similar
to the actual situation. The trajectory lasts about 4,000
III. NUMERICAL SIMULATIONS seconds. Fig 2 shows the trajectory of simulation. The
Numerical simulations were implemented based on the velocity of the trajectory is shown in fig. 3.
IMU, DVL, pressure sensor and a magnetometer used in this
study. To compare different performance with different kind 40.025
of IMU, two specifications of IMU in simulation are close to 40.02
Honeywell HG1700 series and Micro Strain 3DM-GX1,
respectively. First, the specifications of both HG1700AG12 40.015
Latitude(deg)

and MicroStrain 3DM-GX1 are shown in Table I and II [9], 40.01

[10]. To simplify the calculation, only white noise and
random constant are considered in simulation. The 40.005

specifications of DVL, Pressure sensor and magnetometer in 40

simulation are shown in Table III, and the specifications of
39.995
two IMUs in simulation are shown in Table IV.
39.99
115.99 116 116.01 116.02 116.03 116.04
Specification Value Longitude(deg)
Gyro Rate Bias 1.0 deg/hr to 10 deg/hr
Fig. 2 Trajectory of the Simulation
Angular Random Walk 0.125deg/√hr to
0.3 deg/√hr
Accelerometer Bias 1.0 mg (980 mGal) Case 1: High precision IMU
Accelerometer Linearity 500 ppm The first case is to simulate the HG1700 series and the
Table I Specifications of HG1700AG12 simulation results are shown in from fig. 4 to fig. 6. The bias
estimations of inertial sensors are shown in fig. 7 and fig. 8.
It is shown in fig. 5 and fig 6 that the estimated position
Specification Value error of the integrated navigation system is confined with 6.0
Gyro Rate Bias-In Run stability, 0.1 deg/sec meters after 4000 seconds. However, the position error of the
fixed temp navigation system becomes larger and larger when the time
Gyro Rate Bias-In Run stability, 0.7 deg/sec goes. This reason is that the position measurement cannot be
over temp observable directly without any position sensor in this kind
Angular Random Walk 3.5deg/√hr of integrated navigation system. From fig. 13, it can be seen
Accelerometer Bias-In Run stability, 10mg that the attitude error can be estimated thoroughly and be
over temp corrected using DVL and magnetometer. The bias of both
Accelerometer noise 0.4mg/√Hz rms three gyros and three accelerometers also can be estimated
Table II Specifications of MicroStrain 3DM-GX1 perfectly, which means this navigation system is a one with
complete inertial sensor´s observability with current aided
sensors in the situation of limited turns.
Sensor Update Rate Random Noise
(std)
DVL 1Hz 0.05 m/s
Pressure sensor 10Hz 0.01m

518

Authorized licensed use limited to: Indian Institute of Technology Patna. Downloaded on May 03,2023 at 07:31:57 UTC from IEEE Xplore. Restrictions apply.
 ThA2.5

3.5 8

X-Y error (meter)

6

3 4
Velocity (m/s)

2
2.5
0
0 1000 2000 3000 4000
Time t, seconds
2
Fig. 6 Simulation position errors in X-Y plane
1.5
0 1000 2000 3000 4000 1.5
Time t, seconds

gyro x(deg/h)
1
Fig. 3 The velocity of the trajectory
0.5

4 0
Yaw error (deg)

-0.5
2 0 1000 2000 3000 4000
Time t, seconds
0 1.5

gyro y(deg/h)
1
-2
0 1000 2000 3000 4000
Time t, seconds 0.5

0.4 0
Pitch error (deg)

0.2 -0.5
0 1000 2000 3000 4000
0 Time t, seconds

-0.2 2
gyro z(deg/h)
-0.4 1
0 1000 2000 3000 4000
Time t, seconds
0
0.5
Roll error (deg)

-1
0 0 1000 2000 3000 4000
Time t, seconds
-0.5
Fig. 7 Bias estimation of gyros
-3
x 10
-1 2
0 1000 2000 3000 4000
Time t, seconds
accl x(g)

1
Fig. 4 Estimated attitude errors of AUV
0

6
Latitude error (meter)

-1
0 1000 2000 3000 4000
4 Time t, seconds
-3
2 x 10
1.5
0
accl y(g)

1
-2
0 1000 2000 3000 4000
0.5
Time t, seconds
Longitude error (meter)

4 0
0 1000 2000 3000 4000
2 Time t, seconds
-3
0 x 10
1.1
-2
accl z(g)

-4 1.05
0 1000 2000 3000 4000
Time t, seconds
1
0.04
Height error (meter)

0.95
0.02 0 1000 2000 3000 4000
Time t, seconds
0
Fig. 8 Bias estimation of accelerometers
-0.02
0 1000 2000 3000 4000
Time t, seconds

Fig. 5 Estimated position errors of AUV Case 2: MEMS IMU

The second case is to simulate the MicroStrain 3DM-GX1
series and the simulation results are shown in from fig. 9 to

519

Authorized licensed use limited to: Indian Institute of Technology Patna. Downloaded on May 03,2023 at 07:31:57 UTC from IEEE Xplore. Restrictions apply.
 ThA2.5

fig. 10. The bias estimations of inertial sensors are shown in 1.5

gyro x(deg/sec)
fig. 11 and 12.
1
4
Yaw error (deg)

0.5
2

0
0 0 1000 2000 3000 4000
Time t, seconds
-2 1.5
0 1000 2000 3000 4000

gyro y(deg/sec)
Time t, seconds 1
2
0.5
Pitch error (deg)

1 0

-0.5
0 0 1000 2000 3000 4000
Time t, seconds
-1 1.5
0 1000 2000 3000 4000

gyro z(deg/sec)
Time t, seconds 1
3
0.5
Roll error (deg)

2
0
1
-0.5
0 1000 2000 3000 4000
0
Time t, seconds
-1
0 1000 2000 3000 4000
Fig. 12 Bias estimation of gyros
Time t, seconds

Fig. 9 Estimated attitude error of AUV accl x(g)

0.02

0.01

10
Latitude error (meter)

5 -0.01
0 1000 2000 3000 4000
Time t, seconds
0
0.015
-5
0 1000 2000 3000 4000 0.01
accl y(g)

Time t, seconds
Longitude error (meter)

2 0.005

0
0
-2 0 1000 2000 3000 4000
Time t, seconds
-4 x 10
-3

10.5
-6
0 1000 2000 3000 4000
accl z(g)

Time t, seconds 10

0.04
9.5
Height error (meter)

0.02
9
0 0 1000 2000 3000 4000
Time t, seconds
-0.02
Fig. 13 Bias estimation of accelerometers
-0.04
0 1000 2000 3000 4000
Time t, seconds Similar to the case of high precision IMU, it can also be
Fig. 10 Estimated position error of AUV seen in fig. 10 and fig 11 that the estimated position error of
the MEMS IMU integrated navigation system is confined
with 8.0 meters after 4000 seconds. However, the attitude
8 error of the navigation system is much larger than case 1.
X-Y error (meter)

6 The reason is that the poor quality MEMS inertial sensors

4 used in this case. The bias of both three gyros and three
2
accelerometers also can be estimated perfectly.
From the result of simulation, it seems that the simulated
0
0 1000 2000 3000 4000 position error of the proposed navigation system is
Time t, seconds
independent of the type of IMU. If it is true in real
Fig. 11 Simulation position errors in X-Y plane application, it means that the performance of the integration
navigation system of AUV can be improved by only using

520

Authorized licensed use limited to: Indian Institute of Technology Patna. Downloaded on May 03,2023 at 07:31:57 UTC from IEEE Xplore. Restrictions apply.
 ThA2.5

some cheap MEMS IMU. The future work will be done to

testify this conclusion.

IV. CONCLUSIONS
The navigation algorithm described here is able to
estimate the errors of navigation system. Because the
purpose of this report is only to testify the correctness of
algorithm presented, the noise attributes of navigation
sensors are simplified.
The specifications of IMU in simulation are close to
HG1700 series and MicroStrain 3DM-GX1. After 4000sec,
the position error is no more than 8 meters. From the result,
it seems that the position error is insensitive to the precision
of inertial sensors. If it is true in real application, a solution
with cheap sensors can be implemented. The ultimate
performance of experiment will give us the distinct answer.

REFERENCES

[1] Gabriel Grenon, P. Edgar An, Samuel M. Smith, etc. Enhancement of

the inertial navigation system for the Morpheus autonomous
underwater vehicles. IEEE Journal of Oceanic Engineering, Vol 26,
No. 3, pp. 548 - 560
[2] Pan-Mook Lee, Bong-Huan Jun, etc. Simulation of an inertial
acoustic navigation system with range aiding for an autonomous
underwater vehicle. IEEE Journal of Oceanic Engineering, Vol. 32,
No. 2, pp. 327 - 345
[3] Pan-Mook Lee, Bong-Huan Jun, etc. An integrated navigation
system for underwater vehicles with range measurement. Proceeding
of the fifteenth international offshore and polar engineer conference,
Seoul Korea, June 19 - 24, 2005.
[4] Jay A. Farrell. Aided Navigation: GPS with High Rate Sensors.
McGraw-Hill Professional Publishing, 2008, USA
[5] Simon Haykin. Kalman filter and neural networks. Jogn Wiley &
Sons, Inc. 2001, USA
[6] John Leonard, Andrew Bennett, etc. Autonomous Underwater Vehicle
Navigation. MIT Marine Robotic Laboratory Technical Memorandum
98-1
[7] Eun-Hwan Shin. Accuracy Improvement of Low Cost INS/GPS for
Land Applications. Ph.D thesis. The university of Calgary, 2001
[8] Jerome Vaganay, James G. Bellingham, etc. Comparison of fix
computation and filtering for autonomous acoustic navigation.
International Journal of Systems Sciences, 1998, vol(29), No. 10, pp.
1111-1122
[9] MicroStrain 3DM-GX1 Detailed Specifications. MicroStrain, Inc.,
2006
[10] User´s Manual for the HG1700G12 Inertial Measurement Unit.
Honeywell Inc., 2000

521

Authorized licensed use limited to: Indian Institute of Technology Patna. Downloaded on May 03,2023 at 07:31:57 UTC from IEEE Xplore. Restrictions apply.