You are on page 1of 8

Applied Acoustics 151 (2019) 55–62

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

Sound velocity correction based on effective sound velocity


for underwater acoustic positioning systems
Dajun Sun a,b,⇑, Haipeng Li a,b, Cuie Zheng a,b,*, Xiang Li a,b
a
Acoustic Science and Technology Laboratory, Harbin Engineering University, Harbin 150001, China
b
College of Underwater Acoustic Engineering, Harbin Engineering University, Harbin 150001, China

a r t i c l e i n f o a b s t r a c t

Article history: This paper proposes a sound velocity correction method for underwater acoustic positioning systems
Received 29 March 2018 (UAPSs) which suppresses the influence of heterogeneous sound velocity and achieves high-precision
Received in revised form 19 January 2019 positioning for the real-time operations using the effective sound velocity (ESV). The method involves
Accepted 25 February 2019
searching the optimal ESV from a sparse effective sound velocity table (S-ESVT) as-extracted from an
Available online 8 March 2019
effective sound velocity table (ESVT) by genetic algorithm (GA), which minimizes the burden on the hard-
ware of the system and improves efficiency. Simulation results show that the proposed method performs
Keywords:
faster and more accurately than traditional methods. Field trial results show that the reversed RMSE of
UAPSs
Effective sound velocity
localization can be improved from 5.02 m to 2.35 m using the proposed method, where GPS outputs serve
Correction of sound velocity as true values to evaluate the localization performance. The results altogether indicate that the proposed
Genetic algorithm method is well applicable to high-precision real-time localization for UAPSs.
Ó 2019 Elsevier Ltd. All rights reserved.

1. Introduction velocity. To this effect, sound velocity bias is the primary cause
of position error in the UAPSs over a long range [21]. Temperature
Autonomous underwater vehicles (AUVs) or remotely operated variations are also responsible for velocity variations, especially in
vehicles (ROVs) are commonly used in deep water applications deep water [9,24]. Besides, it is found that the temperature varia-
such as geophysical field surveys, offshore industry applications, tion can well explain the velocity variation especially in deep water
environmental measurements, and minefield detection. The sound [15,18]. Thus, the unique characteristics of underwater acoustic
wave is the most effective carrier for underwater information channels necessitate a sound velocity correction algorithm which
transmission. Underwater acoustic positioning systems (UAPSs) properly accounts for sound trajectory bending [9]. Traditional
are regarded as essential positioning and navigation components methods of UAPSs sound velocity correction do not yield suffi-
in these vehicles [9–11,13,16,19,24]. UAPSs including ultra-short ciently accurate results with satisfactory efficiency.
baseline (USBL) positioning systems [21], long baseline(LBL) posi- In recent years, three primary methods have been investigated
tioning systems [29] and short baseline(SBL) positioning systems for UAPSs sound velocity correction: ray-tracing, equivalent sound
[23], etc. are routinely used to locate underwater vehicles with speed profile (ESSP), and effective sound velocity (ESV). The sound
respect to the surface vessels for various operational. The high- trajectory consists of many arcs under the assumption of a strati-
precision measurements of range and bearing are the core of fied sound velocity gradient in ray-tracing theory. Ray-tracing
UAPSs. makes use of standard ray equations to model the trajectories
UAPSs, in which a sonar array is employed to determine the which rays follow in water [2,20,24]. Ray-tracing yields accurate
range and bearing to the vehicles [1], are indeed routinely used results but is a laborious and time-consuming process [2,5,9]. ESSP
today. UAPSs employ two-way message exchange to estimate simplifies ray-tracing theory by using a constant-gradient SSP to
propagation times, thus deducing the ranges between the sonar replace the actual profile. The sound trajectory can be regarded
array and vehicles per a combination of time delay and sound as one arc to be calculated, making ESSP considerably more effi-
cient than ray-tracing [6,7,28]. however, it yields less accurate
sound velocity estimations. ESSP is usually applied to multi-
⇑ Corresponding authors at: Acoustic Science and Technology Laboratory, Harbin beam sonar with operation distance only within tens of meters.
Engineering University, Harbin 150001, China. Effective sound velocity (ESV) theory has been proposed for UAPSs
E-mail addresses: sundajun@hrbeu.edu.cn (D. Sun), lihaipeng@hrbeu.edu.cn that consider distance biases as parameters to be estimated. The
(H. Li), zhengcuie@hrbeu.edu.cn (C. Zheng), lixiang@hrbeu.edu.cn (X. Li).

https://doi.org/10.1016/j.apacoust.2019.02.027
0003-682X/Ó 2019 Elsevier Ltd. All rights reserved.
56 D. Sun et al. / Applied Acoustics 151 (2019) 55–62

ESV between the two points is the ratio between the slant range The coordinates of the responder are given by Eq. (2). y can also
and transit time [25–27]. The curved propagation trajectory may be calculated by the same principle.
be equivalent to a straight line if ESV theory is introduced to
UAPSs. cs13 R
x ¼ R cos h ¼ ð2Þ
This paper proposes a novel sound velocity correction method d
which is suitable for UAPSs. ESV theory is integrated into UAPSs where R is the range between the center of the array and the
within which range and bearing biases are parameters to be esti- responder.
mated. We convert the acoustic angle (grazing angle) to the geo- The departure angle of the ray with respect to the horizontal
metric angle via ESV which is calculated by ray-tracing off-line to plane is referred to as the acoustic angle, which does not reflect
simplify the positioning process; this simultaneously exploits the the true direction of the target as a result of the curved propagation
accuracy of ray-tracing and efficiency of ESV. We calculate the ESVs trajectory. This phenomenon is illustrated in Fig. 2.
of the entire operating area to constitute an effective sound veloc-
ity table (ESVT), then apply a genetic algorithm (GA) to establish a
sparse effective sound velocity table (S-ESVT) for its simple opera- 2.2. Ray-tracing theory
tion process and highly robust for global optimization; this mini-
mizes the burden on the hardware of the system and makes the Ray-tracing theory, which is based on geometrical acoustics
method applicable to real-time operations. We also developed a deduced by Helmholtz equation in the Cartesian coordinate system
faster iterative algorithm originated in Dichotomy to seek the opti- [3], is commonly used to solve the above problem. The general
mal ESV from the S-ESVT for real-time operations and analyzed the expressions of the horizontal propagation distance and propaga-
efficiency of the algorithm. Simulation and field trial results indi- tion time of sound are as follows [3,20].
cate that the proposed method is efficient, accurate, and practical. Z z
ncðz0Þ 0
The remainder of this paper is organized as follows. The basic rðzÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dz ð3Þ
z0 1  n2 c2 ðz0 Þ
model of UAPSs and ESV algorithm are introduced in Section 2. Sec-
tion 3 describes the procedure of the proposed method, including Z z
extracting a S-ESVT by GA and the iteration algorithm originated 1 0
sðzÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dz ð4Þ
from polynomial interpolation and Dichotomy for searching opti- z0 cðz0 Þ 1  n2 c2 ðz0 Þ
mal ESVs. We also evaluate the accuracy and efficiency of the algo-
rithm and compare it against traditional methods by simulation cosðhðz0 ÞÞ cosðhðz1 ÞÞ cosðhðzn ÞÞ
and numeral calculations. Section 4 discusses the field trial we con- n¼ ¼ ¼  ¼ ð5Þ
cðz0 Þ cðz1 Þ cðzn Þ
ducted to test the method; as mentioned above, we found that it
performs well on high-precision positioning tasks in practice. In where r is the horizontal propagation distance; s is the propagation
Section 5 we discusses some notable limitations of the method. time; cðz0 Þ is the sound velocity at depth z0 ; n are Snell-satisfying
Concluding remarks are provided in Section 6. ray-parameters, and hðzi Þ is the grazing angle at depth zi .
The complex SSP is assumed to be a stratified medium with a
constant sound speed gradient. The propagation time and horizon-
2. Basic principles
tal distance are deduced by the following:

2.1. Basic model of UAPSs X


N 1
ci
r¼ j sin hi  sin hðiþ1Þ j ð6Þ
i¼0
g i cos hi
UAPSs always employ a sonar array to measure range from
time-of-propagation and direction per the phase difference of an
 
acoustic signal emitted by a responder. The basic model of an UAPS X
N1
1  1 þ sin hi 1 þ sin hiþ1 
s¼ ln  ln ð7Þ
with four receiving sensors is shown in Fig. 1.
i¼0
2g  1  sin h
i 1  sin h  i iþ1
The coordinates of the responder are denoted as Tðx; y; zÞ. Tak-
ing into account the planar wave approximation, the time- where g is the gradient of each layer.
difference-of-arrival (TDOA) between receivers 1 and 3 is given by

dcosh
s13 ¼ t1  t3 ¼ ð1Þ
c

where d is the baseline length; c is the speed of sound in water.

Fig. 1. Basic UAPS model. Fig. 2. Influence of curved trajectory.


D. Sun et al. / Applied Acoustics 151 (2019) 55–62 57

2.3. ESV for UAPSs According to the algorithm above, ESVT is a function of za ; zt and
r. Each ðza ; zt ; rÞ has a corresponding ESV in this table. The accuracy
The ESV between two points is the ratio of the slant range to the of ESVT is determined by the step size of zt and r.
transit time. The curved propagation trajectory may be equivalent 2 3
ce11  ce1m
to a straight line if ESV theory is introduced to UAPSs (Fig. 3). We
6 .. 7
bring the ESV theory into UAPSs to optimize the target positioning 6 . 7
6 7
process. Unbiased SSP measurements are necessary to avoid sys- 6 . .. 7
ESVT ¼ 6 . . 7 ð8Þ
tematic errors in the ESV calculation [25,26]. The ESVs of the whole 6 . ceij 7
6 7
operating area must be calculated and built into an ESVT. 6 .. 7
4 . 5
The coordinates of the center of the array and the target are
denoted ðxa ; ya ; za Þ and ðxt ; yt ; zt Þ. The horizontal distance between cen1  cenm
the target and receiver is denoted r. Ray-tracing can be used to cal-
culate any ESV with a certain ðza ; zt ; rÞ; we can set up an ESVT with ceij ¼ ce ðza ; zti ; r j Þ ð9Þ
the known depth of array za by changing the values of zt and r. The
Dichotomy is often used to search for the initial grazing angle
which is the basis of ray-tracing. The algorithm based on ray- 3. Sound velocity correction
tracing and Dichotomy for ESV calculation is shown below.
An ESVT can be constructed with small step size to ensure high
accuracy. Searching the table is very time-consuming, however,
Algorithm 1 Calculate ESV
and occupies an excess of valuable hardware resources in practice.
Input: Here, we sample the ESVT via GA to obtain an S-ESVT. We also
The coordinates of the center of the array: ðxa ; ya ; za Þ; attempted to look up the S-ESVT by a faster iteration algorithm
The coordinates of the target: ðxt ; yt ; zt Þ; to find corresponding ESV for the UAPSs.
Output:
ESV; 3.1. S-ESVT based on GA
1: Initial grazing angle: ak ¼ a1 þ2 an ; ðk ¼ 1; 2; . . . ; nÞ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2: Horizontal distance: r ¼ ðxa  xt Þ2 þ ðya  yt Þ2 ; We use the numerical solution mentioned in Algorithm. 1 to cal-
culate an ESVT with 10 m step size of zt and r. An actual deep-sea
3: Depth range: ½za ; zt ;
SSP and its responding ESVT are shown in Fig. 4, where the blank
4: repeat
area in the upper right corner is the sound shadow area.
5: Use ray-tracing with the initial grazing angle ak to
The ESVT shown in Fig. 4 is a 400  1000 matrix, which repre-
calculate t 0 and r 0 ;
sents a massive amount of data for the hardware of UAPSs to man-
6: Compare jr  r 0 j with the threshold 4r;
age. We need a global optimization method to obtain a sparse
7: if jr  r 0 j 6 4r then
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi matrix that contains most information of the ESVT. Genetic algo-
ðzt za Þ2 þr 0 2
8: ce ¼ t0 ; rithm (GA), Simulated annealing algorithm (SA), Particle Swarm
9: else Optimization (PSO), etc. are commonly used for global optimiza-
10: if r  r 0 > 0 then tion. GA searches for parameters by defining a cost function, which
11: ak ¼ a1 þ2 ak ; makes the search process simple. The search process is potentially
12: else parallel. Multiple individuals can be compared at the same time,
13: ak ¼ ak þ2 an ; making it more robust [12]. However, the disadvantage of GA is
14: until jr  r 0 j 6 4r; that the search efficiency is slow when facing high-dimensional
15: return ce ; problems. SA has strong local search capability but knows less
about the search space making it difficult to get into the best
search area [4]. In each evolutionary process of PSO, only the opti-
mal particles can transfer information to the next generation,
which makes the search speed very fast but it also causes it easy
to fall into a local optimum at the same time. In addition, it is
difficult for PSO to effectively solve optimization problems in
non-rectangular coordinate systems [22].

0
1540

1000 1530

1520
z (m)

2000

1510

3000
1500

4000
1450 1500 1550 0 2000 4000 6000 8000
Sound Speed (m/s) r (m)

Fig. 3. ESV for UAPSs. Fig. 4. ESVT of sound speed profile in deep sea.
58 D. Sun et al. / Applied Acoustics 151 (2019) 55–62

0.5 0.5
p =0.2 p =0.2
cr mu
p cr=0.4 p mu=0.4
p cr=0.6 p mu=0.6
0.45 0.45
p =0.8 p =0.8
cr mu
fitness

fitness
0.4
0.4

0.35
0 200 400 600 800 1000 0 200 400 600 800 1000
generation generation
(a) (b)
Fig. 5. The fitness of the population. (a) Crossover probability. (b) Mutation probability.

In this paper, the optimization problem is low-dimensional and


1497
can be solved off-line. Therefore we apply GA to obtain the S-ESVT
for its simple operation process and highly robust. The calculation
process can be found in the literature [8]. Crossover probability and 1496
mutation probability are the most important parameters in the cal-
culation process. We calculate the fitness of the population under
different probabilities by simulation (Fig. 5), and finally choose 1495
the appropriate combination with crossover probability of 0.8
and mutation probability of 0.2.
We optimized the S-ESVT with 20 columns as the samples using 1494
the above steps (same operations for rows). The result is shown in
Fig. 6. The gradient(blue line) is a significant characteristic of sound
which yields more useful information when it is steeper. The red 1493
asterisk represents the gradient value at the output of GA, which
indicates a greater probability for GA outputs in the large-
1492
gradient area. 0 2000 4000 6000 8000 10000
It can be seen that the trend of ESV is not complicated according r (m)
to the horizontal slice (Fig. 7) and the vertical slice (Fig. 8) of ESVT.
So we rebuilt the original ESVT matrix from the S-ESVT by cubic Fig. 7. Horizontal slice of ESVT.
spline interpolation to test the performance of the GA and compare
it against the uniform sampling method. Theoretically, the recon-
struction accuracy increases as the number of samples increases. 1550
It gradually becomes constant, however, after the number of sam-
ples is raised to a certain value. The reconstruction results of the
1540
two methods are shown in Fig. 9. We found that the GA can set
up an optimal S-ESVT with higher precision than uniform sampling
using the same number of samples. 1530

1520

Gridient
GA output 1510
0.2

0 1500
Gridient

-0.2 1490
0 500 1000 1500 2000 2500 3000 3500 4000
-0.4 z (m)

Fig. 8. Vertical slice of ESVT.


-0.6

-0.8
3.2. Target localization

-1
0 50 100 150 200 250 300 350 400
In practice, the depth of the array is known and the depth of the
target can be obtained by a pressure sensor built into the respon-
z (m)
der. The propagation time can be measured by an UAPS array
Fig. 6. GA outputs. installed on the surface vessels.
D. Sun et al. / Applied Acoustics 151 (2019) 55–62 59

reduced the consumption of memory. In the following, we analyze


the difference between the proposed method and other sound
velocity correction algorithms from time complexity and time con-
sumption of simulation.

3.3.1. Time complexity


Time complexity is an important indicator to evaluate an algo-
rithm [17]. It analyzes the time consumption of the algorithm by
separating the analysis process from the system’s software and
hardware. Assume that the time required for each statement of
the algorithm to execute once is the unit time. Therefore, the total
time consumption of an algorithm is the sum of the frequency
count (number of executions) of all statements in the algorithm.
The statement with largest frequency in the algorithm determines
the time complexity of the algorithm.
In ray-tracing theory, we need to calculate the horizontal prop-
agation distance of each layer which is the most frequent
statement.
ci  
Fig. 9. Reconstruction accuracy. ri ¼ sin hi  sin hðiþ1Þ  ð10Þ
g i cos hi
The frequency count of this process (Eq. (10)) is 1 and it is not
This paper develops a novel sound velocity correction method
related to the problem scale. The time complexity of the above pro-
which is suitable for UAPSs. ESV theory which convert the acoustic
cess is T ðnÞ ¼ Oð1Þ . The entire ray tracing algorithm needs to loop
angle (grazing angle) to the geometric angle is integrated into
n times according to Eq. (6); so the time complexity of the entire
UAPSs to simplify the process. We calculate the ESVs of the entire
ray tracing algorithm is T 1 ðnÞ ¼ OðnÞ .
operating area and constitute an ESVT, then apply a GA to establish
The time complexity of the algorithm proposed in this paper is
a S-ESVT and store it in system memory. When field operations are
determined by the dichotomy. Per the core principle of Dichotomy,
in progress, we use Dichotomy to seek the optimal ESV from the S-
the interval ½a; b is halved repeatedly until the interval size is less
ESVT. Different from Algorithm 1, the parameters are discrete in
than the threshold value. The accuracy is as follows
this case. We solve this problem with a combination of limited
information and polynomial fitting differences. The process is as ba
follows.
d¼ ð11Þ
2kþ1
The problem size k can be deduced by
Algorithm 2 Proposed method for localization
 
Input: ba
k ¼ log2 1 ð12Þ
SSP d
The depth of target H Thus, the time complexity of the algorithm proposed in this
The propagation time of sound tp paper is T 2 ðnÞ ¼ Oðlog2 nÞ.
Output: The running time of the algorithm is independent of the prob-
The coordinate of target ðxt ; yt ; zt Þ lem size for ESSP, which means the time complexity of the method
1: Calculate ESVT by Algorithm 1; is T 3 ðnÞ ¼ Oð1Þ.
2: Construct the S-ESVT from the ESVT via GA; The time complexity of the four methods is shown in Fig. 10.
3: ifH is not on the node of S-ESVT then We found that the ESSP and the fixed sound speed method have
4: Use polynomial interpolation to calculate all the ESV lower time complexity due to simplified models. The time com-
C e ðhÞ at this depth;
5: C e ðhÞ ¼ ½C e ðhÞ; C e1 ðhÞ; . . . ; C e2 ðhÞ
6: Search the optimal ESV C eopt ðhÞ based on Newton 50
dichotomy(Iteration value is calculated by polynomial Ray-tracing
interpolation) ESSP
40 Proposed algorithm
7: if H is on the node of S-ESVT then
8: Find the optimal ESV C eopt directly by step 6
9: Bring the slant range R ¼ C eopt  tp into the ultra-short
baseline positioning model 30
T(n)

20
3.3. Time consumption analysis

We adjust the UAPS model by integrating ESV theory into the 10


system. We constitute an S-ESVT off-line and store it in system
memory and seek for the optimal ESV from the S-ESVT when field
0
operations are in progress. There are usually three indicators for 0 10 20 30 40 50
evaluating the quality of an algorithm: (1) Ability to solve prob-
n
lems correctly. (2) Memory consumption. (3) Time consumption.
It has been pointed out above that the S-ESVT has effectively Fig. 10. Time complexity.
60 D. Sun et al. / Applied Acoustics 151 (2019) 55–62

plexity of the proposed algorithm is higher than ESSP but signifi- in Songhua Lake, China, in September of 2017. The SSP of the oper-
cantly lower than ray-tracing. ation area is shown in Fig. 12. The SSP of the same area collected in
June 2017 is also included in Fig. 12.
3.3.2. Time consumption of simulation Location estimation efficiency and location estimation error are
We compare the running time of the algorithm by the simula- often used to evaluate the ability to solve problems of UAPSs. The
tion under the same conditions. The simulation conditions are as number of successfully estimated locations can be affected by the
follows. transmitter signals reflected (multipath) from the water surface
Computer configuration: Intel Core i7-4790, 16G memory, Win- and the uncontrolled directivity of the transmitters that prevent
dows 10. their detections. Location estimation errors are the best metrics
Simulation software: Matlab 2016b. to evaluate the location estimation performance of UAPSs [14].
Target depth is set to 3000 m. The horizontal distance between Location estimation error accuracy is evaluated by the distance
the target and the array is 5000 m. Ray-tracing and ESVT constitu- error. As the depth of target can be known via the pressure sensor.
tion are operated with 10 m step size. We performed 1000 Monte The distance error is defined as the distance difference between the
Carlo simulations of the three methods (Fig. 11). GPS x  y plane position coordinates from the reversed boat x  y
The results indicate that ray-tracing algorithm’s runtime is plane location estimated.
much larger than the other two algorithms. What we need is high
refresh rate real-time targets positioning. In the following, we will Distanceerror ¼ kGPS  Tracked boat locationk ð13Þ
verify the solver performance of three low-cost algorithms through
field tests. A responder was deployed at a depth of nearly 36 m. An USBL
array was installed vertically on a boat equipped with GPS. The
whole system is shown in Fig. 13 and the USBL array is shown in
4. Field test result
Fig. 14.The working frequency of the setup ranges from 8 kHz to
16 kHz. The GPS is differential and has an horizontal-dilution-of-
To evaluate the positioning performance of the method in a
precision(HDOP) of 1.2, which is an ideal output. The S-ESVT was
real-world scenario, we conducted a USBL localization field test
built off-line and stored in the system.
We used the USBL to locate the stationary responder, then
1 reversed the position to track the boat and compared the tracking
Ray-tracing trajectory against the GPS locations to evaluate the positioning
ESSP performance of the proposed method. Fig. 15 shows a comparison
0.8 Proposed method of a widely-used method(blue line) in field trials which uses the
mean value of the SSP as a fixed sound speed, ESSP method (green
line), and proposed method (red line); the black line indicates GPS
0.6 outputs in a horizontal projection. The red line is nearly identical to

0.4

0.2

0
0 200 400 600 800 1000
Number of simulations

Fig. 11. Comparison of simulation time.

5
September
10 June

15

20 Fig. 13. Mounted USBL system.


Depth (m)

25

30

35

40

45
1430 1440 1450 1460 1470 1480 1490 1500
Sound Speed (m/s)

Fig. 12. Sound speed profile. Fig. 14. USBL array.


D. Sun et al. / Applied Acoustics 151 (2019) 55–62 61

the black dotted line, which indicates that the tracking trajectory of
the proposed method approximates the GPS output. The distance
RMS error of the blue line reaches 5.02 m while that of the green
line is 4.39 m and the red line is 2.35 m. These results altogether
indicate that the proposed method well outperforms the others
we tested.

5. Discussion

We collected a set of SSPs (Fig. 12) for the same lake region in
another trial in June 2017. The two SSPs differed considerably in
the same place at different times in which we collected them.
The sound speed changes nearly 40 m/s in the range of 5–10 m
(Fig. 12), which can reduce the performance of the proposed
method.
We calculated the ESV between each element of the USBL and
the transponder in a fixed position (Fig. 16) based on the SSPs
shown in Fig. 12. The results (Table 1) indicate that the June SSP
Fig. 15. Comparison of localization results by different methods with ideal GPS
output. yields a 1:2  103 m=s MSE of ESV while the September SSP leads
to 4:8  106 m=s. The USBL system can barely afford such an error
due to the small scale of its baseline.
The positioning results shown in Fig. 17 also indicate that the
June SSP comes with lower positioning accuracy than the Septem-
ber SSP. The variance of the localization results reached 0.0062 m,
i.e., system performance was relatively poor.
We analyzed the reasons for this phenomenon. As the temper-
ature in June is higher, the temperature in the shallow water area is
increased, causing the speed of sound in the water to become lar-
ger. Besides, the depth of the operation area is only about 45 m
which makes the influence of temperature on the sound speed
more obvious.

6. Conclusion

This paper presented a sound velocity correction method which


eliminates the influence of sound trajectory bending, thus improv-
ing the accuracy and efficiency of real-time operations for UAPSs
Fig. 16. ESV between each hydrophone and the responder. provided that the vertical distribution of sound velocity does not
change dramatically over a small range. We introduce an ESV into

Table 1
ESV of each unit based on different SSP.

ce1 ðm=sÞ ce2 ðm=sÞ ce3 ðm=sÞ ce4 ðm=sÞ ce5 ðm=sÞ MSEðm=sÞ
SSP(Sep) 1465.843 1465.843 1465.843 1465.843 1465.843 4:8  106
SSP(Jun) 1449.648 1449.648 1449.646 1449.645 1449.647 1:2  103

10-4
0.2

0.15
5
0.1

0.05
0
0

-0.05
-5
-0.1

-0.15

-10 -0.2
-1.5 -1 -0.5 0 0.5 1 1.5 2 -0.6 -0.4 -0.2 0 0.2 0.4 0.6
-3
10

(a) (b)
Fig. 17. Localization results. (a) September SSP (MSE = 4:8  106 m=s). (b) June SSP (MSE = 1:2  103 m=s).
62 D. Sun et al. / Applied Acoustics 151 (2019) 55–62

UAPSs and develop an accurate ESVT to implement the ray-tracing [5] Cinar T, Orencik MB. An underwater acoustic channel model using ray tracing
in ns-2; 2010. p. 299–304.
method off-line, which minimizes computational complexity, fol-
[6] Geng X. Precise acoustic bathymetry. Victoria, B.C: Department of Electrical
lowed by an S-ESVT built via GA to reduce the burden on the hard- and Computer Engineering, University of Victoria; 1997.
ware. We use an efficient iterative algorithm originated from [7] Geng X, Zielinski A. Precise multibeam acoustic bathymetry. Marine Geodesy
polynomial interpolation and Dichotomy is with the S-ESVT to 1999;22(3):157–67.
[8] Han Y, Zheng C, Sun D. Signal design for underwater acoustic positioning
determine the optimal ESV. We conducted a series of simulations systems based on orthogonal waveforms. Ocean Eng 2016;117:15–21.
to find that the proposed algorithm is highly efficient and accurate. [9] Isik MT, Akan OB. A three dimensional localization algorithm for underwater
We also used actual USBL positioning system data to validate the acoustic sensor networks. IEEE Trans Wireless Commun 2009:4457–63.
[10] Lambrinos D, Moller R, Labhart T, Pfeifer R, Wehner R. A mobile robot
proposed method’s accuracy in underwater target localization. employing insect strategies for navigation. Rob Auton Syst 2000;30(1):39–64.
Reversed tracking trajectory based on different methods were [11] Lanbo L, Shengli Z, JunHong C. Prospects and problems of wireless
compared to GPS outputs as true values throughout our field trial. communication for underwater sensor networks. Wireless Commun Mobile
Comput 2010;8(8):1530–8669.
The simulation and experimental results altogether indicate that [12] Leung YW, Wang Y. An orthogonal genetic algorithm with quantization for
the proposed is highly accurate, efficient, and suitable for real- global numerical optimization. IEEE Trans Evol Comput 2002;5(1):41–53.
time target localization. [13] Li S, Wang X, Zhang L. Finite-time output feedback tracking control for
autonomous underwater vehicles. IEEE J Oceanic Eng 2015;40(3):727–51.
In the future, we plan to conduct further statistical experiments [14] Li X, Deng ZD, Sun Y, Martinez JJ, Fu T, Mcmichael GA, Carlson TJ. A 3d
and test the performance of this method for all UAPSs in deep approximate maximum likelihood solver for localization of fish implanted
water to achieve more accurate localizations. with acoustic transmitters. Sci Rep 2014;4:7215.
[15] Li Z, Dosso SE, Sun D. Joint inversion for transponder localization and sound-
speed profile temporal variation in high-precision acoustic surveys. J Acoust
Acknowledgment Soc Am 2016;140(1):EL44.
[16] Li Z, Dosso SE, Sun D. Motion-compensated acoustic localization for
This work was supported by the Key Program of the National underwater vehicles. IEEE J Oceanic Eng 2016;41(4):840–51.
[17] Oliveto PS, Witt C. Improved time complexity analysis of the simple genetic
Natural Science Foundation of China under grant 61531012 and algorithm. Theoret Comput Sci 2015;605(15):21–41.
61701132; The Fundamental Research Funds for the Central [18] Osada Y, Fujimoto H, Miura S, Sweeney A, Kanazawa T, Nakao S, Sakai SI,
Universities under grant HEUCFM180504; The authors would like Hildebrand JA, Chadwell CD. Estimation and correction for the effect of sound
velocity variation on GPS/acoustic seafloor positioning: an experiment off
to thank the editor and several anonymous reviewers for their Hawaii island. Earth Planets Space 2003;55(10):e17–20.
helpful suggestions and comments. [19] Paull L, Saeedi S, Seto M, Li H. AUV navigation and localization: a review. IEEE J
Oceanic Eng 2014;39(1):131–49.
[20] Porter MB, Bucker HP. Gaussian beam tracing for computing ocean acoustic
Appendix A. Supplementary data fields. J Acoust Soc Am 1987;82(4):1349–59.
[21] Reis J, Morgado M, Batista P, Oliveira P, Silvestre C. Design and experimental
validation of a USBL underwater acoustic positioning system 16 (9):2016.
Supplementary data associated with this article can be found, in
[22] Shi Y, Eberhart RC. Empirical study of particle swarm optimization. In:
the online version, athttps://doi.org/10.1016/j.apacoust.2019.02. Evolutionary Computation, 1999. CEC 99. Proceedings of the 1999 Congress
027. on, vol. 1; 2002. p. 320–324.
[23] Smith SM, Kronen D. Experimental results of an inexpensive short baseline
acoustic positioning system for AUV navigation. In: Oceans. p. 714–20. vol 1.
References [24] Urick RJ. Principles of underwater sound. 3rd ed. McGraw-Hill Book Company;
1983.
[1] Allotta B, Costanzi R, Meli E, Pugi L, Ridolfi A, Vettori G. Cooperative [25] Vincent HT, Hu SLJ. Geodetic position estimation of underwater acoustic
localization of a team of auvs by a tetrahedral configuration. Rob Autonom sensors. J Acoust Soc Am 1998;102(5):3099–100.
Syst 2014;62:1228–37. [26] Vincent HT, Hu SLJ. Geodetic position estimation for underwater acoustic
[2] Ameer PM, Jacob L. Localization using ray tracing for underwater acoustic sensors. US; 2000.
sensor networks. IEEE Commun Lett 2010;14(10):930–2. [27] Vincent HT, Hu SLJ. Method and system for determining underwater effective
[3] Cerveny V, Popov MM, Psencik I. Computation of wave fields on sound velocity. US; 2002.
inhomogeneous media-gaussian beam approach. Geophys J Int 1982;70 [28] Yang F, Lu X, Dang Y, Liu Z. Accurate and rapid localization of an AUV in an
(1):109–28. absolute reference frame using the iterative resection. In: Oceans. p. 1–6.
[4] Chiu Y, Chang L, Chang F. Using a hybrid genetic algorithm-simulated [29] Zhang J, Han Y, Zheng C, Sun D. Underwater target localization using long
annealing algorithm for fuzzy programming of reservoir operation. Hydrol baseline positioning system. Appl Acoust 2016;111:129–34.
Process 2010;21(23):3162–72.

You might also like