Sei sulla pagina 1di 6

TSLA: An Energy Efficient Time Synchronization

and Localization Algorithm for Underwater Acoustic


Sensor Networks

Zhongze Tian, Cailian Chen, Youjie Xia, Xinping Guan


Department of Automation, Shanghai Jiao Tong University
and Key Laboratory of System Control and Information Processing, Ministry of Education of China
Shanghai 200240, P.R.China
E-mails: {tzz2013, cailianchen, youjiexia, xpguan}@sjtu.edu.cn

AbstractTime synchronization is one of the most funda- message is also much higher than receiving a message. These
mental and widely employed services in networks. Because of bring challenges to time synchronization in UASNs.
long propagation delay in underwater acoustic sensor networks
(UASNs), the time synchronization becomes more challenging. To In order to obtain high precision and energy efficiency of
solve these challenges, this paper presents a new synchronization time synchronization, we need to solve these issues at the
and localization protocol (TSLA), which uses anchor nodes to same time. There are various time synchronization algorithms
broadcast one-way packets to synchronize and localize sensor in UASNs, such as TSHL [5], D-Sync [6], Mobi-Sync [7] and
nodes. Sensor nodes are in duty-cycle mode and only need to DA-Sync [8]. Although they all consider the long propagation
receive packets, which makes the algorithm energy efficient. The- delay in UASNs, they cannot track the skew and offset all
oretical analysis and simulation results demonstrate the energy the time. Inspired by the considerations given above, this
efficiency and high precision of TSLA.
paper presents a new synchronization protocol for UASNs.
TSLA requires that every sensor node receives enough packets
I. I NTRODUCTION from different anchors to finish the process of synchronization.
UASNs have attracted significant attention from both Sensor nodes only receive packets but do not need to send.
academia and industry recently [1]. It provided a broad range Meanwhile, a sleep-wake model of sensor nodes is applied to
of applications such as ocean sampling, assisted navigation, minimize energy cost. Compared with other algorithms, the
coastal surveillance, environmental monitoring and mine re- cost of sensor nodes is low. Anchor nodes broadcast their
connaissance. location and time information to sensor nodes. With the help
of useful packets, the sensor will be synchronized with anchor
Time synchronization is one of the most fundamental and nodes. Theoretical analysis and simulation results will show
widely employed services in underwater networks. Most of the TSLA is highly accurate and energy efficient.
protocols and applications in UASNs require a common time
The rest of the paper is organized as follows: related
scale. For example, data fusion and processing need global
works will be introduced in Section 2. Network model and
time stamps. Time synchronization also plays an important
problem formulation is presented in Section 3. In Section 4,
role in most localization algorithms. Underwater localization
we conduct the time synchronization protocol based on one-
algorithms need the time information to calculate ranges in
way exchanges. Theoretical analysis is presented in Section 5.
sensor networks [2]. Furthermore, not only application layer,
Following that, we provide numerical simulation in Section 6.
but also medium access control (MAC) and network layers
Finally, conclusions and future work are offered in Section 7.
get benefits from time synchronization. Some commonly used
MAC protocols, such as FAMA and TDMA often require time
synchronization among sensor nodes. II. R ELATED WORKS
There are various time synchronization algorithms for
In UASNs, the propagation delay among sensor nodes is
terrestrial and underwater sensor networks. These algorithms
not negligible. The signal propagation speed in an underwater
can be divided into two categories: (1) the receiver-receiver
acoustic channel is about 1500m/s, which is quite lower
approach; (2) the sender-receiver approach. In the following
than the radio propagation speed (3 108 m/s) in ground
part, we will give a brief introduction of recent works for time
[3]. Meanwhile, due to ocean current, tides and bathymetry,
synchronization in underwater sensor network.
sensor node mobility cannot be negligible. The mobility of
nodes caused by the mobility of the ocean current with a TSHL [5] is the first protocol that considers high-latency
speed of 3 6km/h (around 0.83 1.67m/s) [4]. Moreover, communication. TSHL compensates for high-latency commu-
the bandwidth of underwater acoustic channels is limited. nication while also minimizing energy. The basic assumption
Furthermore, energy efficiency is a critical issue to consider. of TSHL is that UWSNs are static. Thus the propagation delay
Due to the limited size of underwater sensor node, the capacity is constant. When nodes move fast, TSHL performs badly. D-
of battery is constrained and it cannot be recharged. However, Sync [6] is a protocol which can handle substantial mobility
the power consumed by consumption is on the order of several without making any assumptions about the underlying motion.
watts or less. Meanwhile, the energy consumed by sending a D-Sync estimates the Doppler shift and compensates the error.

978-1-4673-7687-7/15/$31.00 2015 IEEE


Therefor the synchronization accuracy can be improved. Mobi- Node Clock (y=ax+b) y=x+b
Sync [7] is a novel time synchronization scheme for mobile un-
derwater sensor networks. Different from previous approaches, Clock
Mobi-Sync considers temporal and spatial correlation among time Ideal Clock (y=x)
the mobility patterns of neighboring UASNS nodes. This
enables Mobi-Sync to estimate the moving velocity of ordinary
nodes and the dynamic propagation delays. DA-Sync [8]
is a fundamental pairwise, cross-layer time synchronization
scheme for mobile underwater sensor networks. DA-Sync
Offset(b)
considers a novel Doppler estimation strategy using a bank of
autocorrelators with a well designed preamble. Then utilizing
Real time
Doppler scale estimation in UASNs to gather sensor nodes
relative moving velocities. It considers skew and Doppler shift Fig. 2: Effect of time skew and offset.
together. Thus the accuracy of DA-Sync can be improved.
However the process of Doppler scale estimation is complex.

III. N ETWORK MODEL AND PROBLEM FORMULATION is the average number of anchors per m3 . In probability
theory, a Poisson point process is a particular kind of random
A. Network model
process by which a set of isolated points are scattered about a
In this paper, the network architecture of underwater sensor line or a plane or a three-dimensional space or any of various
networks is shown in Fig. 1. It consists of three types of nodes. other sorts of spaces. We can set = M Vr , where Ma is the
a

total number of anchor nodes in the region of interest, and Vr


1) Buoys. Buoys are equipped with GPS. They can get is the volume of the region. The probability that there are m,
the location and time information of themselves. They m = 1, 2, ..., anchor nodes in the circle of sensor node (R) is
can be viewed as satellite in UASNs. given by:
2) Anchor nodes. Anchor nodes are the nodes which 3
m 4R
3

have been synchronized. They also have the location ( 4R


3 ) e
3
P (m) = (1)
information of themselves, which makes them serve m!
as reference nodes in the underwater sensor network. A message from the anchor node is detected if its distance
3) Sensor nodes. Sensor nodes are ordinary nodes which from an active sensor is less than distance R, which is the
can get the information of the environment and need detecting range.
to be synchronized regularly. They can communica-
tion with anchor nodes directly. Sensor nodes cannot B. Pairwise Synchronization
be recharged, which necessitates the energy efficiency
algorithm. Through estimating the clock offset and skew, we can
perform time synchronization for pairs of clocks. TSLA also
Anchors Buoys Sensors
uses pairwise synchronization approach. Time is measured by
a clock, which is determined by a crystal oscillator and a
Water surface counter. In different hardwares, the crystal oscillator may run
at different frequencies. We give the clock of sensors as:
T = t + , (2)
where is the skew of the clock, is the offset of the clock
and t is the ideal time. The effect of time skew and offset
is shown in Fig. 2. As time progresses, the skew will cause
error increasing. The offset is the initial value between clock
Fig. 1: Network architecture of UASNs. and ideal time. Different nodes may have different offsets. We
need to estimate the skew and offset of the clock accurately
to avoid the need for frequent resynchronization.
To conserve energy, each sensor node follows a (q, p) duty-
cycle schedule [9]. It is active for q time over a period of IV. T IME SYNCHRONIZATION PROTOCOL BASED ON
length p time (p > q) and inactive for the remaining p q ONE - WAY EXCHANGES
time of the period. We assume sensor nodes are randomly
distributed to cover an area for capturing stochastic events. In this section, we will describe TSLA in detail. In order
Each sensor node has three main functional features, such as to carry out synchronization process, every sensor node needs
sensing, communication, and computation. When sensor nodes to receive synchronization packets from different anchors. We
are active, they have the three functions simultaneously. But will introduce the time synchronization scheme in details.
when it is inactive, the three functional modules are turned off
TSLA is an energy efficient algorithm which based on
immediately. Hence, only when it is awake, it detects signals
one-way message exchanges. Sensor nodes are in sleep-awake
sent by the anchor nodes.
mode. The scheme exchanges number of broadcast packets
Assuming the spatial distribution of the anchor nodes is between one anchor and one sensor to compute skew and
followed a Poisson point process of intensity [10], where then performs several exchanges between several anchors and
Reference
t11
time Anchor 1 Anchor 2 Anchor 3 Anchor 4 Sensor node senor node, Ti is the time of the senor node B. In order to
t2
1
simplify the linear regression of skew, we denote:
T11
T12 ij , ij + , j {1, 2, ..., N } (4)

Phase1
t1N

T1N where j means the jth packet of the ith anchor node. Con-
t2

t2 ! 2
T2 necting the equation (3) and (4), we can obtain the equations
below:
t3 Tij = tji + ij , j {1, 2, ..., N }, (5)

Phase2
t3 ! T3
3
N =LM (6)
t4
h iT h iT
t4 ! 4 T4 where N = Ti1 , Ti2 , . . . , TiN , M = , ,

t1i 1
Fig. 3: Progress of TSLA. 2
ti 1
L= . (7)

.. ..
.
one sensor to compute a skew-corrected offset. In some sense, tN 1
i
They only need to receive but not send packets, which makes
TSLA energy efficient. By doing twice linear regression, we Because the progress of synchronization is short. We assume
can estimate the value of skew and offset accurately. the skew is unchanged in the progress of received only
The time synchronization scheme can be divided into two synchronization scheme. Therefor ij = 1. In order to
phases. The whole progress of the algorithm is shown in Fig. minimize the error of estimation, we represent the error using
3. In the first phase, one of the anchors broadcasts one way equations below:
packets to the sensor until the sensor has received enough min{E} = min{eT e} = min{[LM N ]T [LM N ]} (8)
packets to perform linear least squares method to figure out the
skew. These packets only contain the time-stamps information
[LM N ]T [LM N ] = 0 (9)
in the first phase. The error of skew will effect the accuracy M
of our scheme, we need sufficient packets to make sure the where E is the square error of the estimation. According to
high accurate estimation of skew. In the second phase, because the equations above, we can figure out M , or we can use the
the skew has been obtained, we use the skew to figure out equation below to obtain M .
some parameters. These packets contain the time-stamps and
location information in the second phase. With the help of M = (LT L)1 (LT N ) (10)
time and location information of different anchors, we perform
linear least squares to figure out the offset and location of In the second phase, we will estimate the offset and the
sensor. location of the sensor node. The local time includes error due
In our algorithm, although the skew will change as time to clock skew and offset. Therefor after estimating the skew,
progresses, we assume the skew is unchanged in the progress there is a need to estimate the offset to improve the accuracy of
of received only synchronization scheme. Because the progress the local time of sensor node. The skew has been obtained in
of synchronization is short. This assumption allows us to model the first phase and we can use it to figure out other parameters.
the clock skew using linear regression and use it for predicting With the help of time and location information of different
the future time accurately as well. anchors, we perform linear regression to figure out the offset
and location of sensor.
In the first phase, one of the anchor nodes broadcasts one- di
way time packets until the sensor node has received enough Ti = (ti + + wi ) + (11)
v
packets. We can model the drift of the local clock with respect
to the Beacons reference clock. By doing a linear regression on Ti
di = ( ti )v = kpA pB k (12)
the difference between receive time-stamp and the time-stamp
in the message, the value of skew can be estimated. Because In equations above, k.k denotes the Euclidean norm throughout
the error of skew will effect the accuracy of our scheme. The the paper. di is the distance between anchor A and sensor B.
estimation of second phase relys on the estimation of skew pB is the position of sensor B, which can be expressed as
in the first phase. In an underwater acoustic channel, due to [x, y, z]. pA can be obtained in the same way. v is the signal
slow signal propagation speed, another parameter is needed, propagation speed in underwater acoustic channels. The depth
the long propagation delay . The equations of the algorithm of anchor nodes (zi ) and sensor node (z) are known, because
we can use piezometers to obtain depth information of anchor
are given below: nodes and sensor nodes. xi and yi are also known. With the
help of location and time information of anchor nodes, we can
Ti = (ti + i ) + , i {1, 2, ..., N } (3) get i . Putting them into equations (11), equations (13) will
be obtained:
where and are the skew and the offset of the sensor,
Ti
q
respectively, ti is the ideal time of ith anchor node, i is the ( ti )v = (xi x)2 + (yi y)2 + (zi z)2 , i {1, 2, ..., N }

propagation delay of ith anchor node sending packet to the (13)
We denote: V. T HEORETICAL ANALYSIS
Ti In this section, the influence of possible errors on the prop-
Ti , ti , i {1, 2, ..., N }, , (14) agation delay is analyzed. All observations of physical quan-

tities are uncertainty, because the instruments measurement
has been obtained in the first phase. With the help of , Ti can bring errors inevitable. In our synchronization algorithm,
can be obtained. However, x, y and is still unknown. Then since the signal propagation speed in an underwater acoustic
we will obtain equations below: channel is very slow, the propagation delay estimation is the
p main source of error in underwater synchronization.
(xi x)2 + (yi y)2 + (zi z)2 = (Ti )v (15)
Time jitters during the synchronization process will affect
The equation above is nonlinear. We need to estimate x, y and the synchronization accuracy. We can obtain the equation
. Therefor we need at least three other anchors to provide at below:
least three equations. The more equations the anchors provide, = (1 ) (23)
the more accurate the estimation result will be. In order to
The calculated is different from real skew . is the
using linear least square to solve the problem, we need to relative skew error which is effected by first phase packets
make a transformation. Linear least square solutions are given
numbers and jitters. is sensitive to the number of messages
below:
in the first phase. The error of skew dominates the error of
AH =B (16) offset in the second phase.
iT
According to the equation (3), we can see there are three
h
where H = x, y, ,
sources of errors on the propagation delay estimation, which
are from the measurement of x, y and z coordinates. The
2(x1 x2 ), 2(y1 y2 ), 2(T1 T2 )v 2 average error in the measurements of x, y and z coordinates

A= 2
33 are dx , dy and dz . Define d as the propagation distance
2(x1 x3 ), 2(y1 y3 ), 2(T1 T3 )v C (17)
in propagation delay . We can get the equation as follows:

2(x1 x4 ), 2(y1 y4 ), 2(T1 T4 )v 2 q
d = v = d2x + d2y + d2z (24)
2 2

x2 x21 + y22 y12 + z22 z12 2(z1 z2 )z + (T1 T2 )v2
2 Adhering to the theory of error propagation [11] [12], the error
2 2 2

B= 2 2 2 2 2 2
x3 x1 + y3 y1 + z3 z1 2(z1 z3 )z + (T1 T3 )v of the propagation delay is defined as follows.
2 2
x24 x21 + y42 y12 + z42 z12 2(z4 z2 )z + (T1 T4 )v2 d
(18) = (25)
The equations above represent the case that the sensor can v
s
receive three packets from three different anchors. If there are d2x dx 2 + d2y dy 2 + d2z dz 2
more anchors, the row numbers of A and B will increase. d = (26)
As a result, the accuracy of estimation of x, y and will be d2x + d2y + d2z
improved at the same time. According to equations from (15) Equation (25) describes the error in the estimate of propagation
to (17), In order to minimize the error of estimation result, we delay due to errors in dx , dy and dz . The error in
represent the error using equations below: propagation delay will transfer to skew and offset. According
to equation (3), we can get the equation as follows:
min{eT e} = min{[AH B]T [AH B]} (19) s

= ( )2 + ( )2 (27)
[AH B]T [AH B] = 0 (20)
H
According to the equation (20), we can figure out H, or we = (1 ) (28)
can use the equations below to obtain H.
is effected by the jitters and the number of messages from
AT AH = AT B (21) different anchors. The linear regression in the second phase
determine the error of . On the basis of error analysing of
H = (AT A)1 AT B (22) delay estimation, we can obtain the propagation delay. In the
help of propagation delay and skew, the offset error will be
So far, through using linear regression, x, y and have been further determined.
obtained. According to equation (14), as the skew is already
known, then the offset will be obtained. As a result, the VI. N UMERICAL SIMULATION
sensor node can be synchronized and localized. By means
In this part, we will evaluate the performance of the algo-
of two linear regression of first and second phase of our
rithm. With the help of the observed values of received only
algorithm, the estimations of and are highly precise. On
time synchronization scheme, we will show its high precision
the other hand, the whole process of TSLA only based on
and energy efficiency compared with other algorithms. The
one-way packets exchange, therefor the energy consumption of
numerical simulation environment is MATLAB.
TSLA is very low. Although the algorithm needs many packets
to perform linear regression, however the sensor only need to In TSLA, anchor nodes broadcast one-way packets to
receive, therefor TSLA is energy efficiency. synchronize the sensor nodes. Sensor nodes are in sleep-awake
mode, they only receive packets but do not need to send. In
0.25
one run of synchronization progress, sensor nodes only need MUSync
TSLA
to receive four packets. In other algorithms, sensor nodes not
only receive packets, but also send packets. It is obvious that 0.2
the cost of sending packets is much higher than receiving.
Compared with other algorithms, the cost of LA-Sync is low.
0.15
In order to perform numerical simulation of TSLA, some

Errors (s)
parameters need to be initialized such as number of packets
broadcast by anchor nodes, the Gaussian receive jitter and the 0.1
individual clock skew and offset. Without loss of generality,
these parameters are initialized as follows:
0.05
Initial skew=40 ppm.
Initial offset=80us.
0
0 5 10 15 20 25 30 35 40 45 50
Number of beacon messages= 25. Number of beacon message

Granularity = 1us.
Fig. 5: The effect caused by varying the number of beacon
Receive Jitter = 15us. messages.

6
MUSync
MobiSync 2500
DASync MUSync
5 MobiSync
TSLA
TSLA
2000
4
Errors(s)

Number of runs

1500
3

2 1000

1
500

0
0 1 2 3 4 5
10 10 10 10 10 10
Time after sync(s) 0
0.01 0.02 0.03 0.04
Error torlence (s)
Fig. 4: Errors of different algorithms with one run synchro-
nization. Fig. 6: Number of runs when error tolerance is changing.

Fig.4 confirms how error grows after one run synchroniza- 0.12
tion. As time progresses, the skew will cause increasing error. No sync
Under the same conditions, It shows that TSLA grows more MUSync
Error at 10s after synchrization completes (s)

0.1 TSLA
slowly than other algorithms. TSLA takes the propagation de-
lay into account and estimates the delay accurately. Compared
with other algorithms, namely, MU-Sync, Mobi-Sync and DA- 0.08

Sync, it is obvious that TSLA achieves higher accuracy than


other algorithms. The higher accuracy will make TSLA more 0.06

energy efficiency on the other hand.


0.04
As can be seen in Fig.5, TSLA performs better than
the Mu-Sync in most cases. When the number of reference
messages is less than 25, TSLA performs bad than the Mu- 0.02

Sync. Because the TSLA applies linear regression to estimate


skew and offset over a set points of time interval between 0
0 20 40 60 80 100
adjacent timestamp sequences, therefor the TSLA need more The initial skew (ppm)
timestamps at first to guarantee the accuracy. When the number
of beacon message is high enough, the accuracy of skew can Fig. 7: Effect of the initial clock skew.
be improved.
Fig.6 illustrates the energy consumed by resynchronization
(represented by runs of resynchronization) over a period of
105 s of operating time for error tolerance is changing from [6] F. Lu, D. Mirza, and C. Schurgers, D-sync: Doppler-based time
0.01 s to 0.04 s. It shows that TSLA only need less than synchronization for mobile underwater sensor networks, in Proc. Fifth
100 runs when the error tolerance is 0.01 s. In contrast, other ACM International Workshop on UnderWater Networks, New York, NY,
USA, Sep. 2010, p. 3.
algorithms need far more packets. We can conclude that TSLA
[7] J. Liu, Z. Zhou, Z. Peng, J.-H. Cui, M. Zuba, and L. Fiondella, Mobi-
is more energy efficient than other algorithms. It is obvious sync: efficient time synchronization for mobile underwater sensor
that the cost of TSLA is low. The result also shows that for networks, IEEE Transactions on Parallel and Distributed Systems,
all algorithms the number of resynchronizations decreases as vol. 24, no. 2, pp. 406416, 2013.
error tolerance decreases. [8] J. Liu, Z. Wang, M. Zuba, Z. Peng, J.-H. Cui, and S. Zhou, DA-sync:
A doppler-assisted time-synchronization scheme for mobile underwater
Fig.7 illustrates the effect of the initial clock skew on sensor networks. IEEE Trans. Mob. Comput., vol. 13, no. 3, pp. 582
different synchronization algorithms by varying the initial 595, 2014.
clock skew from 10 ppm to 100 ppm. It shows that the [9] S. He, J. Chen, D. K. Yau, H. Shao, and Y. Sun, Energy-efficient
TSLAs performance is independent of the nodes initial skew. capture of stochastic events under periodic network coverage and
coordinated sleep, IEEE Transactions on Parallel and Distributed
It also shows that both MU-Sync and TSLA maintain a Systems, vol. 23, no. 6, pp. 10901102, 2012.
constant average error but have nothing to do with the initial [10] M. Franceschetti, O. Dousse, D. N. Tse, and P. Thiran, Closing the
skew. Because both they estimate clock skew and offset to gap in the capacity of wireless networks via percolation theory, IEEE
obtain high accuracy. However MU-Sync neglects the one way Transactions on Information Theory, vol. 53, no. 3, pp. 10091018,
propagation delay different from the round trip time, therefor 2007.
it obtains a higher average error in the error bars than TSLA. [11] R. K. Bock, H. Grote, D. Notz, and M. Regler, Data analysis techniques
for high-energy physics experiments. Canbridge Univ. Press, UK, 1993.
[12] J. Taylor, Introduction to error analysis, the study of uncertainties in
VII. C ONCLUSIONS AND FUTURE WORK physical measurements. Univ. Science books, USA, 1997.

In the paper, we present a received-only synchronization


protocol, which uses anchor nodes to synchronize the sensor.
A sleep-wake model is applied to make TSLA more energy
efficient. Through changing the duty-cycle, we can balance
the synchronization efficiency and energy efficiency at the
same time. By doing twice linear regression, we can estimate
the value of skew and offset accurately. Extensive simula-
tions demonstrate the energy efficiency and high precision of
synchronization algorithm. Future works will include how to
extend the ideas to more complicated models. The motion
control of the actor nodes will be brought in to make the
performance of the algorithm better.

ACKNOWLEDGEMENT
This work was supported in part by the National Basic
Research Program of China under Grant 2010CB731803, in
part by the NSF of China under Grants U1405251, 61221003,
61290322, 61174127, and 61273181, in part by the Min-
istry of Education of China under Grants NCET-13-0358,
20110073130005, and 20110073120025, and in part by the
Science and Technology Commission of Shanghai Municipal-
ity (STCSM), China under Grant 13QA1401900.

R EFERENCES
[1] S. Shahabudeen, M. Motani, and M. Chitre, Analysis of a high-
performance MAC protocol for underwater acoustic networks, IEEE
Journal of Oceanic Engineering, vol. 39, no. 1, pp. 7489, 2014.
[2] H.-P. Tan, R. Diamant, W. K. Seah, and M. Waldmeyer, A survey of
techniques and challenges in underwater localization, Ocean Engineer-
ing, vol. 38, no. 14, pp. 16631676, 2011.
[3] J.-H. Cui, J. Kong, M. Gerla, and S. Zhou, The challenges of building
mobile underwater wireless networks for aquatic applications, IEEE
Network, vol. 20, no. 3, pp. 1218, Jun. 2006.
[4] P. Xie and J.-H. Cui, An FEC-based reliable data transport protocol
for underwater sensor networks, in Proc. 16th International Conference
on Computer Communications and Networks (ICCCN), Honolulu, HI,
USA, Aug. 2007, pp. 747753.
[5] A. A. Syed and J. S. Heidemann, Time synchronization for high latency
acoustic networks, in Proc. IEEE INFOCOM, Barcelona, Spain, Apr.
2006, pp. 112.

Potrebbero piacerti anche