3-D Localization of an Underwater Sound Source

(Pinger) using a Passive SONAR System

Prateek Murgai
Department of Electrical and Electronics Engineering
Delhi Technological University, Shabad Daulatpur, Main Bawana road, New Delhi-110042, India

pr at eekmur gai 64@gmai l . com

Abstract- The paper introduces a passive sound source
localization algorithm using an array of 4 hydrophones
(underwater piezoelectric sensors). The algorithm is based on
TDOA (Time difference of arrival) estimation employing
Cross-correlation and GCC-PHAT(Generalized Cross
Correlation using Phase Transform) algorithms .To estimate the
position of the pinger intersecting spherical surfaces provide a
closed form solution given the time difference of arrival
measurements when the distance of the source to any arbitrary
reference is unknown.
I. INTRODUCTION
The problemof passive localization of sound sources using
the time difference of arrival from multiple sensors is an
active research area in the field of underwater acoustics, radar,
bio-acoustics, avalanche detection, unmanned vehicles, etc.
Application techniques includes use of hydrophones along a
line, surface arrays like a square array and even 3-D arrays
like tetrahedron, pyramidal or spherical arrays , Analog to
Digital converters (Data Acquisition Cards [5] , TDOA
estimations techniques
There are many techniques to estimate the position of a sound
source based on energy densities, intensities of signals being
received but the most common approach for passive source
localization is to use time delays between pair of sensors to
define curves of constant time difference or path difference
which are hyperbolas of revolution also called as hyperboloids
[5], [6] . Intersection of these hyperboloids yields the position
of the source. In the next section it will be shown that
minimumof 4 sensors are required to locate a source [6]. The
solution is typically found numerically since closed form
solutions to simultaneous hyperbolic algebraic equations are
difficult to find. In most of the applications the source to be
located is at a distance much greater than the distance between
the sensors, so the direction of the source is given by the
asymptotes of the hyperboloids. In 2-D it is given by
intersection of 2 lines whereas in 3-D the problemis reduced
to the intersection of 3 sets of cones (asymptote lines rotated
about axis of each pair of sensors).
One of the basic problems with numerically solving
hyperboloid equations is that it is difficult to forma closed
formsolution, but this is not true for intersecting spheres. This
is because hyperboloids are defined as surfaces of constant
difference of distances between pair of sensors, while spheres
define surfaces of constant distances fromsingle sensor.
Thus keeping in mind about difficulties in solving
hyperboloids, we employ the method involving intersection of
spheres [1], [2],[3] .
II. HYPERBOLOID EQUATIONS
We start with a set of N hydrophones at positions R
i
=
( X
i
,Y
i
, Z
i
). Designate source localization as R
S
=( X
S
,Y
S
,Z
S
)
we arbitrarily set R
0
= (0,0,0) choosing 0
th
hydrophone as
origin.
Each hydrophone receives sound at a time T
i
, Each
hydrophone pair gives a time difference of arrival as Tij=T
i
-
T
j
and the corresponding range difference D
ij =V*
T
ij
(V is the
speed of sound).
Thus

( 1 ) .

Arbitrarily selecting J=0. We get:-

( 2 ) .

Equation (2) represents a hyperboloid with a constant path
difference between a specific pair of hydrophones [6] .
Thus it is clear that a pair of hydrophones yields one equation
so to localize a source in 3-D we need three equations i.e. a
minimum of 4 hydrophones are required to successfully
locate the source.

Fig 1. Square array of hydrophone made in Solid Works
.

Fig 1. Shows a square array of hydrophones which has been
used for localization of the pinger [5]. A symmetrical array
such as the above helps in noise cancellation while making the
time difference of arrival measurements.

III. TIME DIFFERENCE OF ARRIVAL MEASUREMENTS

Two methods of TDOA estimation are presented, they are as
follows:

A)Cross Correlation
In signal processing cross correlation is a measure of
similarity between 2 signals as a function of time lag between
them.
Let there be 2 signals:
S1 (t) =sig1 (t) +n1 (t)
S2 (t) =sig2 (t) +n2 (t)
where sig1 (t) and sig2 (t) are the original signals at each input
port and n1 (t) and n2 (t) are the respective noise signals at
each port.

For continuous signals S1 (t) and S2 (t) cross correlation is
defined as:


where * is the complex conjugate of a function

TDOA is given as follows:



As the signals we will be using are sampled discrete signals
cross correlation for discrete signals is defined as:



where * is the complex conjugate of a function


where Fs is defined as the sampling rate.


Fig 2. Typical Cross Correlation function simulated in MATLAB
B)GCC-PHAT ( Generalized Cross-Correlation using
Phase transform)






where F is the Fourier transformof a function.

GCC-PHAT ,[4] is defined as:



where * is the complex conjugate of a function.



and Rphat (t) is defined as:



where F
-1
is the inverse Fourier transformof a function.


Fig 3. Typical GCC-PHAT simulation in MATLAB

In environments of high levels of reverberation GCC-PHAT
helps to improve robustness and accuracy in calculating the
time difference of arrival [3], [4]. We can see fromthe above
MATLAB simulation that GCC-PHAT enhances the peak and
whitens the region around it, whereas in the Cross Correlation
simulation the region around the peak has some disturbance,
also the peak is not that well defined as it is in the GCC-
PHAT .
Both the methods show similar performance under noise,
GCC-PHAT is better in case of added reverberations.

IV. PASSIVE LOCALIZATION USING INTERSECTION
OF SPHERES

For four sensors, we can define the path difference from
source to sensor i and source to sensor j as d
ij
, in terms of
absolute distances D
i
fromsource to each of the sensors as
follows:

d
ij
= D
i
- D
j

d
ij
= V*T
ij
, ( 1 )

V is the speed of sound.

where T
ij =
T
i
T
j
,which is also equal to the time difference
of arrival.

Take in consideration the distance of the source to the sensors,
which give us four basic equations,

( 2 )

where ( x
i
, y
i
, z
i
)

are the positions of the four sensors and
( x
s
, y
s
, z
s
) is the position of the sound source , D
i
is the
distance of the source to each sensor.
Simplifying equation (2) gives


( 3 )


( 4 )

Solving for the distance of source fromthe origin modifies the
above equation as following:

( 5 )

( 6 )

( 7 )

Substituting Rsin (3) into (6)




( 7 )

without loss of generality, we may take sensor 4 at
origin ,which gives us ( x
4
, y
4 ,
z
4
) =( 0 , 0 , 0 ) .



( 8 )


In terms of symbols above equation can be written as:

( 9 )

where










Fromthe above equations the source position matrix can be
written as following:

( 10 )

The above equation describing the source stands valid only if
is a non-singular matrix. The above equation still includes
Rs which is an unknown, thus we need to reduce the above
equation which has only one set of unknown .To solve this
problemwe carry out following computation.

( 11 )
Substituting ( 10 ) into ( 11 ) yields the following equation:





In the above equation the resultant of terms in the brackets
are 1x1 matrices, i.e. they represent the scalar values of the
1x1 matrices. Solving the quadratic equation gives us the
following results:



where



We consider only the values for which R
s
(range)

is greater
than zero
.
If we get two positive values of R
s
, the actual value
of the range is selected based upon some criterion separating
the two values ,usually the difference between the two values
is large so that they can be separated easily.
Once the range is achieved we can substitute this value in
( 10 ) to achieve the source position or coordinates w.r.t a
reference origin i.e. the sensor 4.

V. MATLAB SIMULATION RESULTS
In this section we construct a simulation model in MATLAB
to forecast the system performance. We assume a 3-D
microphone array consisting of 4sensors as displayed in the
figure below.
To simulate a localization situation we do perform the
following task

Let the pinger position be (20, 16, 5).
Let the sensor to sensor spacing =28 cm=0.28m=d.

Sensor 1 position =(d, 0, 0)
Sensor 2 position =(0, d, 0)
Sensor 3 position =(0, 0, d)
Sensor 4 position =(0, 0, 0)

Speed of Sound =1484 m/s
Path difference between sensor 4 to source and sensor 1 to
source =0.2291083 m


Path difference between sensor 4 to source and sensor 2 to
source =0.15208991m

Path difference between sensor 4 to source and sensor
3 to source =0.046521 m

TDOA1 =(0.2291083/1484) =1.543e-4
TDOA2 =(0.15208991/1484) =1.024e-4
TDOA3 =(0.046521/1484) =0.3134e-4

N1=Number of delayed samples 1 =TDOA1/Fs ~46 samples
N2=Number of delayed samples 2=TDOA2/Fs ~31 samples
N3=Number of delayed samples 3=TDOA/Fs ~9 samples

where Fs is the sampling rate .

To simulate signals we have taken chirps as our input signals.
These chirps are generated within MATLAB itself .These
chirps have an initial frequency f0=22000 Hz and final
frequency f1=27000Hz with a central frequency of 25000Hz.

We have assumed the speed of sound underwater is
approximately 1484m/s and sampling rate is 200000KS/s

The chirps which are assigned to sensor 1, 2, and 3 are
delayed with N1, N2, N3 samples respectively w.r.t the chirp
assigned to the sensor 4.

As we have our 4 input signals in the formof chirps we can
reverse engineer by first applying TDOA methods and then
apply the Spherical Intersection algorithmto estimate position.
The results of the simulation for both the TDOA methods are
stated below.

TABLE 1: Cross-Correlation Results

Pinger
Position
(meters)
Cross-Correlation
|%Error|

Estimated Position

xs

ys

zs

xc

yc

zc

x

y

z

24


16


5


23.75


16.19


4.77


1


1.18


4.6


40.2



10.5



6.3



40.23


10.557
5


6.095


0.07


0.54


3.2


16.6



5.7



3.9



16.47



5.80



3.97


0.78


1.75


1.18


10.6



3.5


3.1


10.69

3.413

3.209

0.85

2.48

3.51

33.9

10.3

6.4

33.76


10.316
0

6.502


0.41


0.16

1.6
TABLE II: GCC-PHAT Results .


V. HARDWARE IMPLEMENTATION OF PASSIVE
SONAR SYSTEM.



Fig 4. Underwater localization task flowchart.

The A/D being used is aNI-DAQ PCI 4462 which is a 24-Bit,
204.8KS/s data acquisition module which has 4 input ports.
The hydrophones being used are TC-4013 which have a
frequency range of 1Hz to 170 KHz and provide
omnidirectional sensitivities. For onboard processing a Single
Board Computer (SBC) or a laptop is used.




VI. CONCLUSION
In this paper, we proposed a method for localization of an
acoustic source using an array of hydrophones (passive
system).Technology based on TDOA method is studied and
the theory and simulation of 2 such methods namely cross-
correlation and GCC-PHAT are presented. Further an
algorithmbased on spherical intersection is used to estimate to
position of the pinger along with range fromthe sensor array.
Simulated results show that the maximumpercentage error in
(x, y, z) coordinates is respectively and the maximum
percentage error in x
s
is 1%, in y
s
is 3.14% and in z
s
is 5%.
We also observe that the error in position due to both TDOA
methods is quite similar as we have not created a situation of
artificial reverberation, thus both methods work equally good.
In conclusion, the implemented method in the paper can be
applied to source localization and achieve fairly good location
accuracy.


REFERENCES
[1] H.C. SCHAU and A.Z. ROBINSON, Passive Source
Localization employing intersecting spherical surfaces from Time of Arrival
Differences , IEEE Transaction , Acoustics ,Speech , Signal Processing ,Vol
ASSP-35 , No .8 ,AUGUST 1987 .
[2] Brian .G .Ferguson, Member, IEEE, KamW. Lo, Senior Member
IEEE ,and Ranjit A .Thuraisingham, Sensor Position Estimation and Source
Ranging in a Shallow Water Environment , IEEE J ournal of Oceanic
Engineering ,Vol 30, No. 2 , April 2005 .
[3] Hamid Reza Abutalebi and Hossein Momenzadeh , Performance
Improvement of TDOA Based Speaker Localization in J oint Noisy
Reverberant Conditions , EURASIP J ournal on Advances in Signal
Processing Volume 2011 ,Article ID 621390 .
[4] C Knapp, G Carter, The generalized correlation method for estimation of
time delay, IEEE Transaction of Acoustics Speech and Signal processing,
Vol 24(4), pp. 320-327, 1976.
[5] Sinha, R.K. Delhi Technological. University. (Formerly Delhi College.
of Eng.), Delhi, India , Ahmed, S. ; Tutwani, M. ; Bahuguna, P. ; Kumra,
R. ; Agarwal, V. ; Saxena, V. ; Mittal, V. ; Gupta, P. , Vehicle for
Automation Research and Underwater Navigation .
[6] Leo Singer, University of Maryland, SONAR cookbook An Underwater
Primer for TORTUGA II , J uly 28 ,2008 .
.

ACKNOWLEDGEMENT
The author gratefully acknowledge Ms. Hemlata Choudhary,
Mr. .Vidush Saxena for their contribution to this journal
paper and for making the concepts more realizable. The author
would also like to appreciate the support of DTU-AUV( Delhi
Technological University Autonomous Underwater Vehicle)
teamfor providing hardware inputs of National Instruments
DAQ ( Data Acquisition Card ) , Hydrophones and a SBC .












Pinger
Position
(meters)
GCC-PHAT
|%Error|

Estimated Position

xs

ys

zs

xc

yc

zc

x

y

z

24


16


5


23.75


16.1


4.75


1


0.6


5


40.2



10.5



6.3



40.22


10.55


6.09


0.05


0.47


3


16.6



5.7



3.9



16.47



5.82



3.93


0.78


2.1


0.77


10.6



3.5


3.1


10.65

3.39

3.19

0.47

3.14

2.9

33.9

10.3

6.4

33.74


10.31

6.49


0.47


0.01

1.4