2010 Second International Conference on Networks Security, Wireless Communications and Trusted Computing

Fast and Accurate WSN Positioning in Emergency

Logistics

Lai Xin Li Jun

Software College State Key Laboratory for Novel Software Technology
Tsinghua University Nanjing University
Beijing China Nanjing, China
[email protected] [email protected]

Zhong Wei Cao Yuan

Science Research Institute Software Center, High-Tech Research Institute (CZ)
General Logistics Department, PLA Nanjing University
Beijing China Changzhou, China

Abstract—Accurate and fast nodes locating is requiredin [7]. Recently, the SDS-TW methodproposed to improve the
large-scale emergency logistics with WSN(wireless sensor network) positioning accuracy [8].
deployed,. Under this background, the sensor nodes often move
rapidly and frequently, and the mobility of nodes leads to poor
reachability to the reference nodes. In this paper, we propose a
fast and accurate positioning scheme based on WSN in emergency
II. SYSTEM OVERVIEW
logistics. Two different wireless sensor networks with 20%+ nodes
moving are deployed. Preliminary experimentsdemonstrate that To obtain the accurate and fast locating effects, two requires
the efficiency of the proposed system overperformed other should be met:
schemes.
1) If enough reference nodes can be found in the network,
Keywords-WSN; wireless localization; localization quality;
and the quality of the reference point should be selected
track forecasting; and movement detection
optimized;
I. INTRODUCTION 2) If the mount of reference nodes are insufficient, , the
In recent years, wireless sensor network [1] attracted attentions positioning accuracy should be improved by predicting
in many different fields. In a WSN, localization technique the nodes’ movements.
provides position information of the sensing data, and basis of Following considerations discussed for these two aims:
WSN routing strategy.
(1) For the situation that enough reference nodes can be
Under different applications, different localization defined, the choice of reference point significantly improves
technique are required. In a large-scale emergency action, fast the locating accuracy, meanwhile, the credibility also can be
and accurate localization of supplies, equipments and personals evaluated dependent on positioning quality;
are important to efficient management. Oncedeployed, a large
amount of WSN nodes should be involved and the nodes move (2) Once the positioning quality reaches the predetermined
rapidly. Therefore, the fast and accurate locating system of threshold, the node could be upgraded to an auxiliary reference.
WSN nodes is required. As a result, the stability and credibility of multi-location
operations can be improved;
Generally, localization techniques can be classified into two
major categories, distance-based localization algorithm and (3) In the case that no enough reference nodes can be
distance-independent localization [2]. Theoretically, WSN- defined with definite locations, we may simplify the solution
based applications are not necessary toobtain exact locations of algorithm by removing the unreliable solution, and use
nodes, that is, the positions of the nodes to be located can be historical data and information around the node trajectory
estimated using the distances between them. Several locating prediction, or combine the communication quality signal
algorithms such as vector-hop [3], centroid algorithm [4], strength and other location coordinates of the projection model.
amorphous algorithm, and APIT, etc. have been proposed. For By data fusion, the generation of transition and uncertainty can
the applications,exact locations of nodes are desired, so be reduced so that the localization performance can be
distance-based location algorithm are usually used . For improved as much as possible.
instance, TDOA-based algorithm was applied in Cricket For description, we define some terms as following:
Positioning System [5], AOA-based algorithm is adopted in the
APS [6], RSSI-based algorithm was used in RADAR system

978-0-7695-4011-5/10 $26.00 © 2010 IEEE 277

DOI 10.1109/NSWCTC.2010.70
 (1) Depending on whether the nodes are kinematic, the In open outdoor environment, the weights in equation (1)
nodes are divided into Fixed-node and kinematic-node. Fixed can be given as f1 : f 2 : f 3 : f 4 = 0.4: 0.1: 0.2: 0.3. From the
node, also referred to as Anchor, has strong processing positioning quality defined in (1), it is observed that
capability and may transmit location data from other nodes positioning accuracy depends on whether or not the node is
from network. For the purpose of cost saving, there is only a moving , and is closely related to the quality of reference point.
small amount in the network. Kinematic node, also referred to
as Tag, has a very limited processing capability and storage
B. Motion Detection
capacity. They need the assist of adjacent nodes to figure out
the location itself; According to the motion history of nodes of WSN, we may
divide the entire WSN area into different zones, which allows
(2) Depending on whether the nodes are location-aware, the refreshment of node information in different zones and thus
nodes are divided into known-node (Knode) and unknow-node will benefit the power-saving scheme of the WSN. For a
(Unode). Knode is a kind of node aware of real-time location dynamic WSN with moving nodes, we suppose that the
such as Anchor, which is located by the assist of adjacent
movement of nodes is not very fast and the nodes move in
nodes. Unode is kind of node un-aware of location currently.
asynchronous way. To determine whether or not the node is
(3) Depending on the count of intermediate-nodes on the moving, we propose a strategy as below.
path to the target node, the adjacent nodes are distinguished as
N-hop Neighbor (where N refers to a natural number). 1-Hop Suppose p is the node whose location is to be determined.
Neighbor (direct adjacent node) is in the signal range of target U 0 denotes the set of neighbor nodes which can be viewed as
node. N-Hop Neighbor is adjacent node who needs to go static relative to node p, U1 represents nodes which are
through at least n-1 intermediate nodes to the target node. If not moving relative to node p, U indicates all of the neighbor
explicitly specified, the Neighbor refers only to the 1-Hop
nodes around node p.
Neighbor nodes.
Before localizing node p, we first judge whether or not
III. ALGORITHM AND PROCESS FLOW node p is in motion. Node p launches a location request to
neighbor nodes in time interval Δt , measures the distance
A. Positioning Quality twice and records the ranging difference between two
Positioning Quality is used to measure the attributes and measured distance Δd i = g (d i1 ) − g (d i 0 ) , for all neighbor
credible scale of position of the node, which is defined as nodes.
4
v = ∑ wi ⋅ fi (q) (1) To detect the motion of nodes, we have the following
i =1 two theorems:
where W1 is the weight of the sub-factor, q denotes the Theorem 1: If count (U 0 ) / count (U1 ) > 1 , where
property vector of the node i, f1 (*) is the amplitude factor, U 0 = {ui | ui ∈ U , Δdi ≤ ε } , U1 = {ui | ui ∈ U , Δd > ε } , and ε
f 2 (*) is the hops factor, f 3 (*) is the count of adjacent nodes indicates a given threshold of Δd , node p can be
factor, f 4 (*) is the reference factor. The four factors are distinguished as moving.
defined as That’s to say, judging from the distance measured
⎧ 1 | d |≤ δ between node p and its surrounding nodes, if the WSN detects
⎪ 1 more static nodes than moving nodes, we determine that node
f1 ( d ) = ⎨
| d |> δ p is in a quiescent state.
⎪⎩| d − δ + 1 |
,
Theorem 2: If node p is in moving state, we have
1
f2 = count (U 0 ) / count (U1 ) << 1 , i.e., judging from the distance
K,
measured between node p and its surrounding nodes, the WSN
π detects far fewer static nodes than moving nodes.
f3 = arctan( L − 1)
2 ,
M
1 C. Trace Prediction
f4 =
M
∑υ i
i =1 . Trace prediction is introduced to improve further the
quality of localization. As raw positions contain noises, we
In the definition, δ indicates the threshold between state
can smooth the raw data by an average smoothing filter.
and moving, d indicates the moving distance, K refers to the
hops from itself, L indicates the number of neighbor nodes For a given node, there is its historical data
around the current node, M indicates the amount of reference H = { pi | i = 1, 2,..., n} Using average smoothing filter to
nodes selected.
smooth the original historical data with a window size of M,

278
 1 M integration of multi-mode solver positioning strategy
we have ∑ pi + j , i = 1, 2,..., n − M + 1} .
H ' = { pi' | pi' =
M j =1 to determine positions;
Obviously, the values of M will affect the characteristics of  Divide the WSN area into different zones, perform
historical data. If the M value is too large, historical data will motion detection zone by zone. If the network is in
be a bit too flat. In the smoothing, not only the random error not in motion, return to step 1) after a long time
has been offset, but also the node position change is smooth interval, restart step 2) to step 6), refresh the position
out at the same time. information of the entire WSN, and optimize the
positioning results;
Historical data of a certain moment can be denoted by
vectors (r, v, a), where r is the position vector ( rx , ry , rz ), v is  If a moving node in WSN is detected, position the
the velocity vector (vx, vy, vz), a is the acceleration vector (ax, node separately. If there are a few moving nodes
ay, az), then the position can be estimated as detected and enough reference nodes can be defined,
return to step 5); If there are a lot of moving nodes
~ ∂rt '−τ ∂ 2 r ' τ 2 ∂3r ' τ 3 and reference nodes in WSN is not enough, return to
r t = rt '−τ + τ + t2−τ * + t3−τ * (2)
step 1) for entire WSN re-positioning;
∂t ∂t 2 ∂t 3
In procedure 7), the entire WSN refreshment is
where τ is the time interval of historical location points; rt'−τ
implemented by combining the previous positioning results
refers to the nearest historical point; with re-positioning results. As a result, the WSN positioning
∂rt '−τ rn' − rn' −1 refers to the nearest historical results are further optimized. In procedure 8), the entire WSN
vn = = '
∂t t (rn ) − t (rn' −1 ) re-positioning is implemented by omitting all the previous
2 '
positioning results, updating all the node positioning
velocity; an = ∂ rt −τ = vn − vn −1 refers to the nearest information including the positioning quality.
∂t 2 t (vn ) − t (vn −1 )
3 '
IV. RESULT
historical acceleration; en = ∂ rt −τ = an − an −1 refers to
∂t 3 t (an ) − t (an −1 ) To verify the effectiveness and efficiency of the
the trend of nearest historical acceleration. proposed positioning procedure, we deployed two
experimental WSN in several typical environments (open
space, sparse forest, grassland and etc.). The WSN includes 30
D. Localization Procedure nodes to composite 3 multi-hop network in 500m X 1000m
region and 50 nodes to composite 6 and more multi-hop
Combining the above definitions and criteria, localization network in 1000m X 2000m region, respectively. In the test,
procedures of the WSN can be outlined as follows: we recorded the initialization time, positioning accuracy,
 Smooth the ranging data by several smoothing refresh time, positioning effective rate and so on.
techniques to eliminate points with abnormal The WSN node we deployed is shown in Fig. 1. The
ranging data; node has an operating range of more than 1000 meters and a
 Set several reference nodes as the original reference ranging precision less than 1 meter. Besides, the node supports
nodes in the WSN for positioning, and start the RSSI and may operate at communication speed of 2Mbps.
network traffic routing update;
 Based on the position information of reference nodes,
determine adjacent nodes’ positions with modified
maximum likelihood estimation method and Gauss-
Newton method. Then setting the localized nodes as
new reference nodes, determine the positions of new
adjacent nodes’ positions, and so on. All nodes in
WSN can be positioned;
 Calculate the positioning quality defined in (1), and
endow those moving nodes with positioning quality
by specific algorithm;
Figure 1. Node used in the WSN for test.
 For nodes in motion, if neighboring reference points
is enough, use Newton's iteration method and update
the positioning quality;
 For nodes in motion, if neighboring reference points
is insufficient, use trajectory prediction in (2) and

279
 Table 1
Performance Index Network 1 Network 2 Remarks

30 nodes, 50 nodes, Communication Distance of each

500m X 1000m X hop is 75~100meters;
1000m 2000m More than 5 nodes in the same
3 hops 6 hops hop;
network network At least three reference points in
WSN WSN the peer communication range
Initialization Time <90s <120s
Positioning Accuracy 1,2 <2m <2m Positioning Accuracy of nodes in
Hop 1 and 2
Positioning Accuracy 3,4 <4m <4m Positioning Accuracy of nodes in
Hop 3 and 4
Positioning Accuracy 5,6 <5m Positioning Accuracy of nodes in
Hop 5 and 6
Network Refresh Interval 30seconds 40seconds
Positioning Effective Rate 94% 89%
Edge Effective Rate 65%<4m 65%<4m Positioning Effective Rate when
90%<10m 90%<10m reference nodes is not enough
Mobile nodes refresh interval <10seconds <15seconds When nodes are in motion.
Figure 2. Static Positioning Results

In the test, we find that that the environments with WSN

deployed have no significant influence on the localization of V. CONCLUSION
WSN, if one can handle the shadow effects around the In this paper, we proposed a fast and accurate positioning
buildings properly. scheme for WSN in large-scale emergency logistics. To
When there are multi-hops in the WSN, the positioning demonstrate the effectiveness and efficiency of the proposed
quality can be guaranteed as well if adjacent nodes’ positioning scheme, preliminary experiment test is reported. It is shown
quality is good and stable. Though the data communication that the proposed positioning scheme works effectively and
process is not well managed and optimized, the data efficiently.
transmission can still meet the requirements for node Future work may be carried out in several aspects. First, the
positioning and text transmission. impacts of WSN topology on the network positioning accuracy
The test also indicates that the deployed nodes can organize should be considered. Second, how to stabilize the hops and
a WSN and perform preliminary WSN positioning in 1~2 routes of reference nodes deserve more study. Third, more
minutes. When the WSN is set in stationary state (nodes keep efforts must be paid to improve the reliability of positioning
still more than 10 minutes), the positioning accuracy of the nodes at the edge of a large-scale WSN such as in emergency
entire WSN is about 1 meter, as shown in Fig 2. logistics.

When more than 20% nodes in the WSN are moving, the REFERENCES
entire WSN positioning information refreshes every 3 minutes,
the positioning results are listed in Table 1. Form Table 1, we
have the observation that the proposed positioning scheme [1] Sun Li-ming, Wireless Sensor Network, 2005: 3. (In Chinese)
works effectively and efficiently. [2] Wang Fu-Bao, Shi Long, Ren Feng-Yuan, Self-Localization Systems
and Algorithms for Wireless Sensor Networks, Journal of Software
vol.16 No5, 2005: 857-878. (In Chinese)
[3] Niculescu,D. Badri Nath. Ad-hoc Position System[J]. IEEE
Communications Societies, 2003,(3):1734-1743.
[4] Stephan Schuhmann, Klaus Herrmann, Kurt Rothermel, Jan Blumenthal,
Dir Timmermann, Improved Weighted Centroid Localization in Smart
Ubiquitous Environments, Ubiquitous Intelligence and Computing, pp.
20-34, 2008.
[5] Nissanka B. P, Anit C. Hari B. The Cricket Location-Support System[J].
ACM, 2000:32-43.
[6] Niculescu, D. Badri Nath. Ad Hoc Positioning System(APS) Using
AOA[C]. IEEE Communications Socienties, 2003(3):1734-1743.
[7] Pahl P, Padmanabhan V.N. RADAR:An In-Building RF-based User
Location and Tracking System[C]. IEEE Communications Socienties,
2000(2): 775-784.
[8] Andreas S, Chih-Chieh H, Mani B. Dynamic fine-grained localization in
ad-hoc networks of sensors[C]. ACM, 2001: 166-179.

280