© The Author 2010. Published by Oxford University Press on behalf of The British Computer Society. All rights reserved. For Permissions, please email:
[email protected] doi:10.1093/comjnl/bxq002
Efficient Query-Based Data Collection for Mobile Wireless Monitoring Applications Long Cheng1 , Yimin Chen2 , Canfeng Chen3 , Jian Ma3 , Lei Shu4 , Athanasios V. Vasilakos5 and Naixue Xiong6,∗
5
1 State
Key Lab of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China 2 Department of Electrical Engineering, Stanford University, Standford, CA 94305, USA 3 Nokia Research Center, Beijing, China 4 Department of Multimedia Engineering, Osaka University, Osaka, Japan 5 Department of Computer and Telecommunications Engineering, University of Western Macedonia, Kozani, Greece 6 Department of Computer Science, Georgia State University, Atlanta, GA 30303, USA ∗Corresponding author:
[email protected]
10
Considering sensor nodes deployed densely and uniformly a mobile sink moving through the sensing field queries a specific area of interest for monitoring information. The Query packet, injected by the mobile sink, is routed to the specific area and the corresponding Response packet is expected to return via multi-hop communication. In this paper, we analyze such a network model to address the problem of efficient data collection for mobile wireless monitoring applications. We first propose a meeting position-aware routing (MPAR) protocol for routing the Response packet efficiently and then propose an efficient query-based data collection scheme (QBDCS) for mobile wireless monitoring applications based on the MPAR. In order to minimize the energy consumption and packet delivery latency, the QBDCS chooses the optimal query time of injecting the Query packet and tailors the routing mechanism for sensor nodes forwarding packets. Simulation study has verified the analysis and demonstrated that the QBDCS can significantly reduce the energy consumption and end to end delivery latency.
15
20
25
Keywords: wireless sensor networks; mobile sinks; data collection; wireless monitoring Received 26 August 2009; revised 2 December 2009 Handling editor: Yu-Chee Tseng
30
1.
35
40
INTRODUCTION
In recent years, wireless sensor networks (WSNs) for wireless monitoring have been designed and developed in both military and civil areas for a wide variety of applications, including smart battlefield, environment and habitat monitoring, health care, home automation, traffic control etc. [1]. Some of these applications, for example, environment and habitat monitoring, require a large number of tiny sensors while these sensors usually operate on limited non-rechargeable battery power. Furthermore, a WSN is always expected to last over several months or years with limited power supply. Thus, energy saving becomes one of the most critical design issues for WSNs,
while other constraints, such as throughput and fairness, become relatively less important. To extend the lifetime of a WSN for wireless monitoring application, one common approach is to dynamically schedule each sensor’s work/sleep cycle (or duty cycle) [2]. When a sensor is in the sleep mode, the sensor’s processor will be turned off, but a low-power timer or some other triggering mechanism may be running to wake up the sensor at a later time, therefore it consumes only a tiny fraction of the energy, compared to that consumed in the active mode. More specifically, a radio will be turned off when it is not actively sending and receiving, and it will be turned on when communication is expected [3].
The Computer Journal, 2010
45
50
2 55
60
65
70
75
80
85
90
95
100
105
L. Cheng et al.
The duty-cycling operation in WSNs can be loosely categorized into two main types: (1) random duty-cycling, where sensors are turned on and off in a random fashion independent of each other; and (2) coordinated duty-cycling, where sensors coordinate via communication and information exchange to collectively achieve an on/off schedule followed by multiple sensors. With regard to random duty-cycling, its advantage lies in its simplicity and without any additional communication overhead for coordination among neighbors. However, randomly turning off the wireless radio inevitably results in a larger delay for each hop [4]. Energy efficient data collection in wide-deployed WSNs is a well-studied problem. Aiming at solving this problem, researchers have proposed to exploit mobility of sinks for energy efficient data collection in WSNs [5]. It is believed that this can balance the energy consumption of each sensor node and reduce the possibility of routing hot spots, which are introduced by fixed sinks due to the nearby heavy data flow. The mobile sinks can be generally categorized into two kinds: (1) using dedicated mobile robots which are specially designed for collecting sensed data; and (2) employing existing mobile objects such as mobile handsets, vehicles or ships, on which the data collection devices are usually mounted. For the dedicated mobile robots, the motion trajectory of mobile sinks is always carefully designed and controlled to improve the performance of WSNs, such as [6, 7]. However, in real-world scenarios, mobile sinks are more likely to be the combination of data collection devices and existing mobile objects with random mobility, e.g. mobile handsets, notebook computers or vehicles. Some studies have been done for exploiting such mobile objects present in wireless monitoring fields to collect and deliver sensed data, such as Data MULEs [8], TSA-MSSN [9] and [10]. In this case, the mobility of sinks may be uncontrollable, but predictable in a short period to some degree [9]. For example, when a vehicle moves along the road with a digital map available, the prediction of its future mobility becomes quite accurate. Moreover, employing existing mobile objects to collect sensed data is economical by reason of no expenditure of deploying dedicated mobile robots. In this paper1 , we focus our investigation on densely and uniformly deployed sensor networks for wireless monitoring applications. A mobile sink moving through the sensing field at a certain speed querying a specific area or a point of interest to collect sensed data, which we call the query-based data collection in mobile wireless monitoring applications. A Query packet targeted for an interested area is injected by the mobile sink and the corresponding Response packet will be returned via multi-hop communication. Due to the mobility (velocity and direction) of the sink, the Query and Response packets always have different routes. This research problem is encountered in many wireless monitoring scenarios,
1An
abbreviated version of this paper appeared in [11].
e.g. environment monitoring, habitat monitoring and intelligent transportation systems [12]. In such applications for query-based data collection, the requested data packets are expected to be delivered from the source node to the mobile sink energy efficiently and as quick as possible. The routing mechanism consequently becomes the major challenge since the position of the mobile sink is continuously changing. One intuitive solution is to continuously broadcast the sink’s mobility information so that the sensors can be updated periodically. However, frequent broadcasts yield a high communication overhead, and may cause collisions in wireless channels. To solve the problem, a position prediction technique is adopted in this paper. Mobile sinks being studied are usually mounted on reasonable moving objects, such as mobile handsets, vehicles or ships. Although their mobility is considered random, during a short period, in most situations they would follow certain mobility models [13, 14] and their future position can thus be predicted based on the continuity of mobility. The specific contributions of this paper mainly come from the following two points: (i) a meeting position-aware routing (MPAR) protocol for routing the Response packet is proposed. MPAR predicts the position of the mobile sink at the time of meeting the Response packet. Thus, the Response packet can be directly forwarded toward this predicted position from the source node. (ii) an efficient query-based data collection scheme (QBDCS) for mobile wireless monitoring applications is proposed based on MPAR. The QBDCS finds the optimal time to inject the Query packet, in order to complete a query-based data collection cycle with minimum energy consumption and delivery delay. The rest of this paper is organized as follows. Section 2 surveys the related work. Section 3 presents the system model and problem statement. The detailed MPAR protocol and QBDCS design are presented in Sections 4 and 5, respectively. Simulation results are presented in Section 6. Finally, Section 7 concludes this paper.
2.
110
115
120
125
130
135
140
145
RELATED WORK
Recently, the demand for WSNs in many monitoring applications has grown tremendously in both military and civil areas, such as surveillance system [15], habitat monitoring [16], environmental observation and forecasting [17], health monitoring [18] and intelligent traffic [12]. Geographic routing is considered to be an efficient and scalable data delivery scheme and is quite commonly adopted for information delivery in large-scale WSNs. It is assumed that the sensor nodes are location aware and the source node
The Computer Journal, 2010
150
Efficient Query-Based Data Collection for Mobile Wireless Monitoring Applications 155
160
165
170
175
180
185
190
195
200
205
knows where the destination is. Sensor nodes may have global positioning system (GPS), or an ad hoc way to figure out their coordinates. The basic idea for geographic routing is greedily forwarding data packets to the neighbor geographically closest to the destination. A number of research works on geographic routing in WSNs have been conducted, e.g. GPSR [19], GEAR [20], TPGF [21] and [22]. In [19], a packet is forwarded to a one-hop neighbor that is closer to the destination than the current node. This process is repeated until the packet reaches the destination, or the packet is blocked at a node whose onehop neighbors are all farther away from the destination, which is defined as a hole. Some variants of geographic routing, such as [23–26], propose different routing metrics for different applications and network models. Many hole-bypassing mechanisms have been studied. As proposed in GPSR [19], when a packet gets stuck at a node, it is routed by the right-hand rule counter-clockwise along a face of the planar graph until either it can be routed by greedy forwarding again, or it goes back to where it enters the perimeter mode, which means that no route to the destination is available. Another approach is to identify hole boundary nodes first, then use the knowledge of these identified boundary nodes to facilitate the hole-bypassing routing [27]. The problem of efficient data collection has been investigated by many researchers. Most researches are based on the assumption that data collection involves all nodes of a WSN. A typical representative is the LEACH protocol [28], which selects a small number of clusters in a self-organized manner. A node in each cluster designated as the cluster head, collects and fuses data from other nodes in its cluster and transmits the result to the base station. In [29], the authors proposed to balance the energy consumption of the entire network, instead of load balance within each cluster in LEACH. In [30], a twotier data dissemination approach was proposed to provide data delivery to multiple mobile sinks. Each data source constructs a backbone structure (partitioning the whole network into several small grids.) for query and data transmission. When the sink requests any data, the query will only be flooded reaching the data source. However, the overhead associated with maintaining and recalculating the grid would be very high [31]. In fact, there is a number of queries that select only a subset of nodes in a network. Upon receiving a user’s query message, each sensor node that is corresponding to the query delivers the sensed data to the sink node. The authors in [32] argued that selective queries (query-based data collection) are an important class of queries that can benefit from algorithms that are tailored for partial node participation of a WSN. In [33], an on-demand localized data collection scheme is proposed. According to the fixed location of the sink node, packet propagation and data collection are localized in a small area. The query message is efficiently unicasted to the forwarding node in the queried area, and is forwarded into the network within a localized area. Sensed data is collected at the forwarding node, at which data is aggregated. The aggregated data is then sent to the
3
fixed sink node. In contrast to the sink with fixed position in [33], we focus on the query-based data collection with mobile sinks. The network model in this paper, which is similar to that of [34], as shown in Figure 1, is inspired by the recent research on WSNs applied in the area of mobile wireless monitoring, collecting data such as temperature, sound, pressure or pollutants. The Query packet is sent by the mobile sink to a WSN and the corresponding Response will be delivered from the queried area to the mobile sink. However, as a case study, the proposed network model in [34] is too application-specific and the algorithm is restricted to the conditions that are initially assumed. For example, some active sensor nodes, or Vise Sink in their work, need to be deployed along the mobile sink trajectory. The introduction of Vise Sinks is not only constraining a lot of applications, but also introducing significant delivery latency when they are too widely separated. In this paper, the network model considers a more general case without such assumptions. The application scenario can be that sensor nodes are densely and uniformly deployed in the sensing field where no Vise Sinks exist. Possible mobile sinks are different kinds of roving vehicles moving through the sensing field to query the WSN. Considering the continuity of mobile sink mobility and for simplicity, the analysis in this paper only focuses on a mobile sink moving along a straight path. In [34], the authors have proposed an adaptive routing algorithm based on the instantaneous position estimation, which can avoid frequent broadcasting of the mobile sink’s position. However, they only considered the Response Propagation phase, in this paper we consider a complete query-based data collection cycle, including choosing the right moment to inject
t=t0
t=t0
FIGURE 1. Example of the considered application scenario in [34]. The mobile sink injects a Query in the WSN through the closest vice sink, which is then forwarded to the queried region. The Response packet is finally delivered through a Vice Sink to the mobile sink. However, the Response packet is forwarded toward the mobile sink’s instantaneous position, where the routing path as a whole deviates from the optimum (if directly forwarding toward the predicted packet-sink meeting position). Besides, the authors in [34] only considered the Response Propagation phase, and did not point out when to inject the Query packet.
The Computer Journal, 2010
210
215
220
225
230
235
4 240
245
L. Cheng et al.
the Query packet. Besides, their method was based on the mobile sink instantaneous position prediction. Each relaying sensor node estimates the instantaneous mobile sink position and then routes the data packet toward the virtual destination node according to a given metric (for instance Euclidean distance or energy-aware distance) calculated by the relaying sensor node. However, the routing path as a whole deviates from the optimum compared with the proposed MPAR protocol, because MPAR can forward the Response packet directly toward the predicted packet–sink meeting position.
3. 250
Sensor nodes
FIGURE 2. Densely and uniformly deployed sensor nodes form a two-dimensional grid, allowing a slight deviation.
SYSTEM MODEL AND PROBLEM STATEMENT
3.1.
Query
Common assumptions
Response
We make the following common and reasonable assumptions.
255
260
265
270
(i) Sensor nodes are homogeneous, immobile, energyconstrainted and expected to run for a long time. They are able to wirelessly communicate with neighbors in a short range. (ii) Each sensor node has a duty cycle DC ∈ [1%, 100%]. It periodically opens the radio to transmit sensed data, but at other times it sleeps to save energy. (iii) Each sensor node can identify its geographic location and maintains a neighbor table for routing packets. The location information can be gained by running a localization algorithm [35]. (iv) The mobile sink has an estimation of its current mobility (velocity, direction and position), which can be obtained from the GPS etc. The mobile sink operates on a rechargeable battery and has much higher computation and communication capabilities than sensor nodes. (v) The mobile sink can obtain the sensor node’s information, such as transmission range and duty cycle. This can be achieved by the initial negotiation procedure. 3.2.
275
280
Sensor deployment strategy
For simplifying the analysis and without loss of generality, we consider that sensor nodes are deployed in the sensing field densely and uniformly, forming a two-dimensional grid [36], where inside each cell a sensor node is placed allowing a little deviation. For example, sensor nodes are manually placed at selected locations and the deployment tolerates a slight deviation, as shown in Fig. 2. 3.3.
Hierarchical network structure
Considering the wireless monitoring applications with sensor nodes densely and uniformly deployed in the sensing field, a mobile sink moving through the sensing field with random speed queries a specific point or an area of interest for monitoring
Upper Tier: Mobile Sink Middle Tier: On Demand Cluster Head Lower Tier: Sensor Nodes Query Propagation Response Propagation
Interested Area
FIGURE 3. An example of query-based data collection for mobile wireless monitoring applications. Sensor nodes, on-demand cluster head and mobile sink constitute the three hierarchical tiers. Obviously, the Query and Response packets have different routes.
information from the physical world. The hierarchical network structure consists of three tiers: sensor nodes, on-demand cluster head and mobile sink. Each Query packet contains location information of the interested area. The sensor node closest to the center of the interested area elects itself as the cluster head, which is called the ‘on-demand cluster head’, and is responsible for gathering and aggregating data in the interested area. The aggregated data (Response packet) is then sent back to the mobile sink via multi-hop communication. As shown in Fig. 3, a vehicle is moving along a straight road, and it queries a WSN deployed around the road for wireless monitoring. The deployed WSN enables vehicles to obtain the conditions of the surrounding environment, e.g. the status of parking slots in a specific parking ground along both sides of the street. 3.4.
285
290
295
Problem statement
A complete query-based data collection cycle is composed of three phases: (1) Query Propagation phase, where the initially injected Query packet is propagated to the on-demand cluster head within the interested area; (2) Data Aggregation phase, where the queried sensed data is aggregated and the Response packet is generated by the on-demand cluster head; and (3) The subsequent Response Propagation phase, where the
The Computer Journal, 2010
300
305
Efficient Query-Based Data Collection for Mobile Wireless Monitoring Applications
310
315
320
corresponding Response packet is forwarded and then delivered to the mobile sink. The query-based data collection problem is conceptually formulated by the following question: how can the WSNs complete the transmission of the Query and Response packets with minimum energy consumption and delivery delay? The basic technique adopted is the geographical location-aware routing; a scheme of bypassing holes is also adopted during packet transmission. To minimize the energy consumption and delivery latency for transmitting the Query and Response packets, the following specific approaches are taken: (1) the packet delivery velocity is estimated and the position of the mobile sink at the time it meets the Response packet is predicted; thus, the Response packet can be forwarded toward the meeting position via multihop wireless transmission directly with geographical locationaware routing; and (2) the optimal time for the mobile sink to inject the Query packet to a WSN is estimated.
4. 325
330
335
340
345
On-demand cluster head selection
Once a Query packet is sent by the mobile sink, which includes the location information of the interested sensing field (radius and center), it will be routed toward the specific area of interest until reaching a node which we call the on-demand cluster head, i.e. the sensor node closest to the center of the interested area. At each hop, the current forwarding node calculates its distance from the center of the queried area. If the distance is less than a predefined threshold and there is no other neighboring node that has a closer distance, the node will elect itself as the on-demand cluster head. It then broadcasts a data aggregation request within the queried area. Finally, it aggregates2 the sensed data and generates a Response packet. Note that we use the Euclidean distance as the metric in cluster head selection and the routing process. The metric can also be combined with residual battery energy of nodes for the load balance consideration. 4.2.
interference, duty cycling operation and numerous other factors [37]. Let L be the length of the packet, R the data transmission rate of each sensor node, d the distance between the data source and destination and r the communication range of each sensor node. Let dhop be the average distance per hop (dhop ≤ r). Intuitively, the number of hops between a source and a destination is roughly linear with the distance d between them and the variance depends on the actual sensor nodes deployment. Taking the duty-cycled DC operation into account, which introduces an additional delay at each hop, we have the estimated packet delivery velocity under the ideal condition as follows: V =
d r · R · DC ≈γ · , (L/R · DC ) · d/dhop L
350
355
360
0 < γ ≤ 1, (1)
where d/dhop is the estimated hop count and γ (γ = dhop /r) is the modification coefficient in the packet delivery velocity estimation by taking into account the actual deployment of sensor nodes.
365
MPAR PROTOCOL
In this section, we first briefly introduce the on-demand cluster head selection in the Data Aggregation phase. Then, a MPAR protocol is presented for forwarding the Response packet in the Response Propagation phase. 4.1.
5
Packet delivery velocity estimation
The packet delivery velocity V is defined as the delivered distance a packet travels divided by its end-to-end delay. The end-to-end delay of a message propagating in WSNs is related to the node processing speed, overhead, congestion, collision, 2 Data aggregation technique is not described in detail since it is not the focus in this paper. We assume that only one Response packet will be generated toward a query.
4.3. Average distance per-hop estimation The Query packet is basically composed of the following elements: (i) interested types of the sensed data (temperature, humidity etc.); (ii) geographic location information of the interested sensing field (radius and center); (iii) mobile sink mobility status and the timestamp; (iv) routing hop counts. During the Query Propagation phase, each relaying node increases the hop count. Thus, given each propagation distance and the hop counts, dhop can be estimated at the on-demand cluster head. 4.4.
370
375
380
Meeting position estimation
Let t0 denote the query time point, when the mobile sink injects a Query packet to a WSN, let t1 denote the time point of the corresponding Response packet being sent and t2 denote the estimated packet-sink meeting time point. Let Lr and Vr be the length and estimated delivery velocity of the Response packet, respectively. Given the starting point (x1 , 0) of the Query packet, the center of the interested sensing area (a, b) and the velocity of the mobile sink Vs , as depicted in Fig. 4, we have the estimated meeting position (EMP) (x2 , 0) of the Response packet and the mobile sink. (x2 − a)2 + b2 |x2 − x1 | (t1 − t0 ) + = , (2) Vr Vs where r · R · DC Vr = γ · . (3) Lr
The Computer Journal, 2010
385
390
6
L. Cheng et al. t0, (x1, 0) Mobile Sink
y
On-Demand Cluster Head
t2, (x2, 0)
Vs
EMP
t1
x
Vr t1, (a, b)
FIGURE 4. The method of packet–sink meeting position estimation, EMP is the abbreviation for estimated meeting position.
395
4.5.
MPAR design
Based on the packet delivery velocity estimation and packetsink meeting position estimation, the MPAR protocol for the Response Propagation phase is designed as follows.
400
405
410
415
4.5.1. Adaptive EMP modification After the queried sensed data is aggregated and a Response packet is then generated by the on-demand cluster head, the cluster head first calculates the initial EMP. It then attaches this information to the packet and sends it to the next relaying node by adopting a geographic routing algorithm. For each relaying sensor node, before forwarding the Response packet, it first recalculates the EMP and updates the corresponding fields in the Response packet, which is also calculated under the condition that the Response packet is forwarded toward the estimated destination in a straight line with speed Vr , as shown in Fig. 5. The recalculated EMP may deviate from the initial EMP, possible reasons are that the propagation delay exceeds the initial estimation (e.g. due to bypassing ‘holes’), or the propagation delay is overestimated. Then the Response packet will be forwarded toward the new calculated EMP by adopting geographic routing algorithm. This recursive process will last until the Response packet arrives at or goes beyond the ‘notifying node’, as shown in Fig. 6. The ‘notifying node’ is referred as the sensor node nearest to the latest calculated EMP. When the Response packet bypasses a
t1, (a, b)
FIGURE 5. Example of the adaptive routing mechanism in MPAR, recalculating the EMP recursively, EMP 1 is the initial estimation, and EMP 2 is the recalculated estimation.
t1, (a, b)
FIGURE 6. Example of the MPAR protocol under the case that the mobile sink has passed the ‘notifying node’. Node A is the ‘notifying node’; when the Response packet arrives at the Node A, it finds out that the mobile sink has already passed, thus entering into the chasing mode. i+5
FIGURE 7. Example of routing recovery mechanism for bypassing holes; when facing a hole, the relaying node follows the EMP direction, selecting the neighbor with the smallest positive angle between the EMP direction and the next hop direction.
hole, it may arrive at a sensor node close to the road and beyond the ‘notifying node’, as shown in Fig. 7. 4.5.2. Chasing mode When the Response packet arrives at the ‘notifying node’ and finds that the mobile sink has not passed and is not in its radio range, the ‘notifying node’waits for the mobile sink to come and then delivers the packet. However, it is possible that by the time the Response packet arrives at the ‘notifying node’, the mobile sink has already passed, possibly because the mobile sink’s mobility varies significantly or the packet delivery velocity is overestimated. In this case, the packet enters into the Chasing Mode as depicted in Fig. 6. The packet follows the sink’s moving trajectory until it catches up with the mobile sink. 4.5.3. Bypassing holes When a relaying sensor node finds that its one-hop neighbors are all farther away from the EMP, it means that this sensor
The Computer Journal, 2010
420
425
430
435
Efficient Query-Based Data Collection for Mobile Wireless Monitoring Applications
440
445
450
455
460
465
node is facing a hole. A hole is likely to appear when a set of sensor nodes are dead due to energy exhaustion or damage. The Response packet cannot progress toward the EMP by greedily examining its local neighborhood when facing a hole. In this case, the relaying sensor node will apply the recovery routing mechanism to route the Response packet to the one-hop neighbor that has the smallest positive angle formed between the EMP direction and next hop direction, as shown in Fig. 7. When the Response packet arrives at any node close to the road, although that node is not the ‘notifying node’, it just checks whether the mobile sink has passed or not. If has not passed, the node waits for the mobile sink to enter its radio range before delivering the Response packet. Otherwise, the Response packet enters into the Chasing Mode following the sink’s moving trajectory until it catches up with the mobile sink. 4.5.4. Deadline awareness In order to avoid the case where the packet chases the mobile sink endlessly, there usually exists a limit on the period of propagation time that the user allows between sending out the Query and receiving the Response packet. This time limit is denoted as Tdeadline . Similar to the TTL (time to live) in TCP/IP protocols, if the Response packet cannot arrive at the destination within Tdeadline , it should be discarded. However, determining the remaining time to Tdeadline at each intermediate node is not trivial due to the lack of a globally synchronized clock. To handle this problem, we borrow the mechanism in [38] by piggybacking the elapsed time to the packet so that the following forwarding node can determine the remaining time to deadline without a globally synchronized clock. 4.6.
470
475
480
485
Summary of MPAR
The flowchart of the MPAR protocol is presented in Fig. 8. The on-demand cluster head is responsible for aggregating the queried sensed data and generating a Response packet. It then calculates the EMP for the Response packet. After that, it forwards the Response packet to the next relaying node. For a relaying node, once receiving a Response packet, it first checks whether it exceeds the time limitation Tdeadline . If yes, the Response packet will be discarded. Then it checks whether it is the ‘notifying node’. If yes, the Response packet either enters into the Chasing Mode or waits for the mobile sink if the mobile sink is not in the relaying node’s radio range. If the relaying node is not the ‘notifying node’, it first recalculates the EMP and then forwards the Response packet to the next relaying node recursively. This process will last until the Response packet arrives at the ‘notifying node’ or a sensor node close to the road and beyond the ‘notifying node’. MPAR is more efficient compared with the existing routing protocol [34] for similar mobile wireless monitoring
7
deadline
FIGURE 8. Flowchart of the MPAR protocol.
applications. The feature of MPAR is that it predicts the position of the mobile sink at the time of meeting the data (Response) packet. Thus, the Response packet is then directly forwarded toward this predicted position from the source (ondemand cluster head) during the Response Propagation phase. However, in [34], each individual hop is toward the virtual mobile sink instantaneous position calculated by the relaying sensor node. The routing path in [34] is expected to be an arc line theoretically. The comparison of MPAR with the routing algorithm in [34] is shown in Fig. 9. It is desirable that the routing path of the Response packet in MPAR is shorter than the referenced routing protocol, which may lead to lower energy consumption and consequently shorter delivery latency.
5.
490
495
QBDCS
Intuitively, there must be an optimal query time, at which sending a Query packet is energy efficient and has shorter delivery latency theoretically. To complete a query-based data collection cycle with minimum energy consumption and delivery delay, the QBDCS finds the optimal time to inject the Query packet based on the proposed MPAR protocol.
The Computer Journal, 2010
500
505
8
L. Cheng et al. where
t1
Estimated mobile sink instantaneous position
Lq , R Lr Er = Pt · . R
Eq = Pt ·
Sensor Node Notifying Node
where
EMP Cluster Head
525
r · R · DC , Lq r · R · DC . Vr = γ · Lr
Referenced Protocol
Vq = γ ·
MPAR
t1, (a, b)
FIGURE 9. Example of the comparison of MPAR with the referenced routing protocol in [34].
510
515
QBDCS objective
Let Eq and Er be the average energy consumption per hop of sending the Query packet and Response packet, respectively; Lq be the length of the Query packet; and Vq be the estimated delivery velocity of the Query packet, as shown in Fig. 10. The two objectives of the QBDCS are to minimize the energy consumption and at the same time shorten the delivery latency when transmitting the Query and the corresponding Response packets. These two objectives are inherently unified; and minimizing energy consumption is chosen as the QBDCS objective in this paper. Thus, the optimization problem can be formulated as follows:
min Eq ·
(x1 − a)2 + b2 + Er · dhop
(x2 − a)2 + b2 dhop
y
x1 ,x2 ,λ L(x1 , x2 , λ) = 0.
5.2.
On-Demand Cluster Head
535
(11)
When the mobile sink is located at the critical point (xoptimal , 0), where the objective function attains a global minimum, we get the optimal query time which can minimize the overall energy consumption to complete the transmission of Query and corresponding Response packets.
Vq d2 Response Propagation phase
α
530
540
x
Query Propagation phase
d1
(9)
Here tg is the estimated data aggregation time within the interested sensing area, which is proportional to the size of the interested area. The Lagrangian multiplier method is applied to solve this problem. Incorporating the condition into (4), we have the auxiliary function (x1 − a)2 + b2 (x2 − a)2 + b2 + Er · L(x1 , x2 , λ) = Eq · dhop dhop (x1 − a)2 + b2 (x2 − a)2 + b2 +λ· + Vq Vr |x2 − x1 | , (10) + tg − Vs
, (4)
Vs
Mobile Sink
(8)
and solve
t2, (x2, 0)
t0, (x1, 0)
520
(6)
Here Pt denotes the transmission power of a packet. The constraint condition is (x2 − a)2 + b2 |x2 − x1 | (x1 − a)2 + b2 + tg + = , (7) Vq Vr Vs
Mobile Sink
5.1.
(5)
β
Vr t1, (a, b)
Data Aggregation phase
FIGURE 10. Three phases in a query-based data collection cycle: (1) Query Propagation phase, (2) Data Aggregation phase and (3) Response Propagation phase.
Optimal query time
5.2.1. A result Generally, the Query packet is much shorter than the Response packet in length. In this case, the optimal query time is when the mobile sink is located at (xoptimal , 0), which the mobile sink can meet with the Response packet at (a, 0), as shown in Fig. 11. 5.2.2. Proof Let Lr /Lq = k; we have Er /Eq = k, Vr /Vq = 1/k. To simplify the analysis, we assume that tg ≈ 0, which will not affect the
The Computer Journal, 2010
545
550
9
Efficient Query-Based Data Collection for Mobile Wireless Monitoring Applications t0, (xoptimal, 0)
Optimal query time when K>1
t2, (a, 0)
Vs
Mobile Sink
Optimal query time when K>1
x
EMP d2
Vq d1
Vr On-Demand Cluster Head
y
On-Demand Cluster Head
t1, (a, b)
(a, b)
K>1 K>1
FIGURE 12. Two cases (k > 1 and k < 1) of the optimal query time.
FIGURE 11. Theoretical optimal query time based on MPAR.
final result. The objective function can be written as Eq · min{d1 + k · d2 }, dhop 555
(12)
where Eq /dhop can be regarded as a constant. Let Vq /Vs = m; the constraint is d1 + k · d2 = m · (13) d12 − b2 + d22 − b2 .
5.2.3. The general result The theoretical optimal query time depends on Lr /Lq = k, which is the ratio of the length of the Response and Query packets. As shown in Fig. 12, when k > 1, the optimal query time is when the mobile sink can just meet the Response packet at (a, 0); when k < 1, the optimal query time is just when the mobile sink is located at (a, 0). When k = 1, the optimal query time is when the mobile sink is located at such a position where the Response and Query packets can have symmetrical routes (α = β).
580
585
Applying the Lagrange multipliers method yields k·
560
d1 d12
−
− b2
d1 + k · d2 − m ·
d2 d22
−
5.3.
= 0, b2
d12 − b2 +
d22 − b2
= 0.
(14)
Let d1 = b sec α and d2 = b sec β, where α, β ∈ (0, π/2). Then (14) can be written as k 1 − = 0, sin α sin β sec α + k sec β − m tan α − m tan β = 0, 565
−2mk sin β + (m + k + 1) sin β = 0. Thus,
575
2
2
(16)
π m2 + k 2 + 1 , β ∈ 0, . (17) 2mk 2 Obviously, (17) has no solution. Therefore, the objective function attains the optimal value when α = 0 or β = 0. When k > 1, it can be proved that the objective function has optimal value when β = 0, which is equivalent to the mobile sink and the Response packet meeting at (a, 0). Based on the above analysis, when the Query packet is shorter than the Response packet in length, the QBDCS finds the optimal query time when the mobile sink is located at (xoptimal , 0), where (a − xoptimal )2 + b2 a − xoptimal b + tg + − = 0. (18) Vq Vr Vs sin β =
570
2
The QBDCS is summarized in Table 1. It is noteworthy that the dhop is estimated by the on-demand cluster head in MPAR. However, the estimation of dhop is conducted by the mobile sink when injecting the Query packet in the QBDCS, which can be estimated after the initial negotiation procedure. A scheme of bypassing holes is adopted when designing the MPAR; in fact, this scheme will be adopted during packets transmission, not limited to the Response Propagation phase.
590
595
(15)
which has boundary values when α = 0 or β = 0. Solving (15), when k = 1, we have 3
Summary of QBDCS
6.
SIMULATION
To evaluate the performance of the proposed QBDCS, we simulate the QBDCS in ns-2 simulator. We first consider regular network topology (without ‘holes’), where sensor nodes are deployed in a 1000 m × 500 m sensing field densely and uniformly. A rectangular grid (50 × 25) is generated, with the side length of each cell being 20 m. Then we test QBDCS for random topologies, which are all generated by the setdest tool in ns-2. 6.1.
600
605
Simulation settings
The key communication parameters generally applies to IEEE 802.15.4, as listed in Table 2. The velocity of mobile sinks varies from 20 to 30 m/s. For simplicity, we assume that the data aggregation time could be neglected. Sensor nodes communicate at a rate of R = 250 kB/s, the transmission range is set at r = 40 m. The CSMA/CA mechanism for collision management is adopted. Taking the data-link layer
The Computer Journal, 2010
610
10
L. Cheng et al.
TABLE 1. Summary of the QBDCS for mobile wireless monitoring applications.
Given Vs , R, r, Lq , Lr , DC , (a, b), dhop . Calculate the EMP and the optimal query time (Xoptimal , 0) by (18). Step 1: Querying at the right time The mobile sink checks whether it is at the optimal query time. If yes, it injects the Query packet toward the interested area. Step 2: Routing the Query packet The Query is routed toward the interested area by adopting the geographical location-aware routing. Step 3: Data aggregation Once the Query packet arrives at the targeted sensing area, the sensor node closest to the center of the interested sensing area elects itself as the cluster head, gathers and aggregates the sensed data, and then generates a Response packet. Step 4: Routing the Response packet The Response packet is then routed toward the estimated packet–sink meeting position until it arrives at the ‘notifying node’ or a sensor node close to the road and beyond the ‘notifying node’. The sensor node checks whether the mobile sink has passed or not. If it has passed, the Response packet enters into the Chasing Mode until it catches up with the mobile sink. Otherwise, the sensor node waits for the mobile sink to enter its radio range before delivering the Response packet. During this process, if exceeding the time limit Tdeadline , the Response packet should be discarded. Step 5: Response packet delivery The Response packet is delivered to the mobile sink.
TABLE 2. Communication parameters setup.
Parameter Range R Lq Lr CT DC
Unit m kB/s bytes bytes s
Values 40 250 30 120 1 1%
denote the maximum delay of the Query and Response packets, respectively; we have Dq ≈ 160 ms and Dr ≈ 640 ms. 6.2.
Performance metric
6.2.1. End-to-end delivery delay This metric measures the amount of time from injecting a Query packet by the mobile sink to the corresponding Response packet returning to the mobile sink. 6.2.2. Normalized sending cost The cost of a transmission consists of the sending cost of the sender, and the receiving cost of one-hop neighbors. Given a fixed number of neighbors, the transmission cost is proportional to the sending cost. Hence, the sending cost metric measures the communication overhead and energy efficiency. We define the normalized energy cost for sending a Response packet as one cost unit. Therefore, the energy cost for sending a Query packet is 0.25 cost unit. 6.2.3. Deviation from the initial EMP This metric measures the deviation of the final meeting position from the initial EMP. 6.2.4. Packet delivery ratio This is the ratio of the successful end-to-end transmissions (the mobile sink receives the Response packet) to the total number of tests. 6.3.
620
CSMA/CA mechanism into consideration, about 40% overhead is introduced [34]. Thus, the data transmission rate under saturation condition is R = 150 kB/s. Sensor nodes have a cycle time CT = 1 s and DC = 1%. The lengths of the Query and Response packets are 30 bytes and 120 bytes, respectively. Therefore, the estimated delivery velocities Vq (Vq = γ · (r · R · DC /Lq )) and Vr (Vr = γ · (r · R · DC /Lr )) are 250 · γ m/s and 62.5 · γ m/s, respectively. Let Dq and Dr
630
635
640
645
Evaluating QBDCS
When evaluating the proposed the QBDCS, three other schemes are also simulated for comparison as follows: (1) injecting the Query 4s earlier than the QBDCS; (2) injecting the Query 4s later than the QBDCS; and (3) the ‘Naive’ scheme, which the mobile sink sends the Query packet when locating at the nearest point from the interested area, as shown in Fig. 13. All schemes adopt the MPAR protocol to forward the Response packet. The simulation starts from the mobile sink injecting a Query packet
Injecting a Query
615
625
FIGURE 13. Schematic of different Query injecting times for the mobile sink.
The Computer Journal, 2010
650
11
Efficient Query-Based Data Collection for Mobile Wireless Monitoring Applications
665
670
675
680
685
Deviation from the initial EMP (%)
6.3.1. Choosing γ The estimation of the packet delivery velocity has a direct effect on the performance of the QBDCS. An overestimate may lead to sending the Query packet later than the optimal query time. An underestimate may result in sending the Query packet earlier than the optimal query time. For both cases, the Response packet has to frequently change the estimated packet–sink meeting position, and consequently deviates from the initial EMP. We simulate a mobile sink querying the area with QD = 500 m. Fig. 14 depicts the change of end-to-end delivery delay by increasing the value of γ with different mobile sink velocities. From Fig. 14, it is shown that the optimal value for the modification coefficient γ is around 0.95, at which the QBDCS shows the least end-to-end delivery delay. Figure 15 shows the changes of deviation from the initial EMP as the γ changes. The deviation increases as the γ decreases when it is smaller than 0.95, where the packet delivery velocity is underestimated. When γ is larger than 0.95, where the packet delivery velocity is overestimated, the QBDCS will send the Query packet later than the optimal query time, which also causes the deviation. It is found that the optimal value of the modification coefficient γ depends on the sensor deployment strategy. In our simulations, the optimal value of γ is about 0.95, whereas when the side length of each cell in the deployed sensing filed is 30 m and the transmission range is set at 50 m, the optimal value of γ is about 0.7.
0.25
Vs=25 m/s Vs=30 m/s
0.2
0.15
0.1
0.05
0.65
0.75
0.8
0.85
0.9
0.95
1
FIGURE 15. Deviation from the initial EMP vs. modification coefficient γ .
6.3.2. The impact of QD We vary the QD from 260 to 500 m, comparing the differences when the mobile sink velocity is set at 25 m/s. The modification coefficient γ is taken to be 0.95. Obviously, the end-to-end delivery delay and energy consumption depend on the QD. The longer the QD, the larger is the delivery delay and more the energy consumption. As shown in Fig. 16, with an increase of the QD, the advantage of the QBDCS becomes more significant. Compared with other querying times, the QBDCS outperforms these schemes. This is because the QBDCS can guarantee the least hop counts during the Response Propagation phase, which is the bottleneck of the end-to-end delivery delay and the energy consumption. From Figs 17 and 18, we observe that the QBDCS has the minimum Response packet hop count. Although the number of Query packet hop counts in the QBDCS is more than the ‘Naive’
18
OBDCS
Vs=25 m/s
13.5
0.7
Modification coefficient γ
Vs=20 m/s
14
Vs=20 m/s
0
Vs=30 m/s
End-to-end delivery delay (s)
660
toward the interested area, and ends when the mobile sink receives the Response packet or when exceeding the limit of the predefined propagation time. The vertical distance from the interested area to the road where the mobile sink is moving along is defined as the query distance (QD).
End-to-end delivery delay (s)
655
13 12.5 12 11.5 11
4s earlier
16
4s later Naive scheme
14
12
10
8
10.5
6 0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
250
Modification coefficient γ
FIGURE 14. End-to-end coefficient γ .
delivery
delay
300
350
400
450
Query distance (m)
vs.
modification FIGURE 16. End-to-end delivery delay vs. QD.
The Computer Journal, 2010
500
690
695
700
12
L. Cheng et al. 22
Query packet hop counts
22
OBDCS
OBDCS
4s earlier
4s earlier
End-to-end delivery delay (s)
24
4s later
20
Naive scheme
18 16 14 12 10 8 6
20
4s later Naive scheme
18
16
14
12
10
250
300
350
400
450
20
500
22
4s earlier
Normalized sending cost (unit)
Response packet hop counts
20
4s later Naive scheme
18 16 14 12 10 8
250
4s earlier 30 28
4s later Naive scheme
26 24 22 20 18
300
350
400
450
500
20
FIGURE 18. The Response packet hop counts vs. QD.
scheme and the ‘4s later’ scheme, the QBDCS achieves the minimum end-to-end delivery delay.
715
6.3.3. The impact of mobile sink velocity We then vary the mobile sink velocity from 20 to 30 m/s with QD = 500 m. The modification coefficient γ is taken as 0.95. Figs 19 and 20 depict the effect of increasing mobile sink velocity on end-to-end delivery delay and normalized sending cost, respectively. We can see that the performance of the QBDCS does not show much dependance upon the mobile sink velocity. This is because of the fact that completing a query-based data collection mainly depends on the hop counts of routing the Response packet. The QBDCS always provides a minimum Response packet hop count during the Response Propagation phase compared with other schemes. For the ‘Naive’ scheme, as the mobile sink velocity increases, the number of Response packet hop counts increases as well. Thus,
22
24
26
28
30
Mobile sink speed (m/s)
Query distance (m)
710
30
16
6
705
28
OBDCS
32
OBDCS
22
26
FIGURE 19. End-to-end delivery delay vs. mobile sink velocity.
FIGURE 17. The Query packet hop counts vs. QD.
24
24
Mobile sink velocity (m/s)
Query distance (m)
FIGURE 20. Normalized sending cost vs. mobile sink velocity.
the performances of the ‘Naive’ scheme decreases rapidly with the increase of the mobile sink velocity. Next, we look into the impact of the variation of mobile sink velocity for the QBDCS. We set a mean mobile sink velocity and a variation range; the mobile sink moves with the mean velocity, and changes its speed every 2 s to another random value within the range; the results have been averaged over 100 runs. Figs 21– 23 show the end-to-end delivery delay, normalized sending cost and deviation from the initial EMP as the variation of the mobile sink velocity increases, respectively. We can see that even when the variation of mobile sink velocity is significant, for example, the variation ranges in [−15 m/s, 15 m/s], it has little influence on the end-to-end delivery delay and normalized sending cost performance, with 12% end-to-end delivery delay increase and only a 4% normalized sending cost increase. This is partially because of the sensor node deployment in our simulation. From Fig. 23 we can see that the distance of routing the Response packet
The Computer Journal, 2010
720
725
730
735
13
Efficient Query-Based Data Collection for Mobile Wireless Monitoring Applications 20.5
Avg. end-to-end delivery delay (s)
Avg. end-to-end delivery delay (s)
13 12.5 12 11.5 11 10.5 10
Mean Vs=20 m/s Mean Vs=25 m/s
9.5 9
0
2
4
6
8
10
12
14
19.5 19
[Packet Delivery Ratio] [30%]
18
[60%]
17.5
[70%]
17
[60%]
16.5 16
[60%]
[80%]
15.5
[80%]
15
14 16
[10%]
Naive scheme
18.5
14.5
Mean Vs=30 m/s
OBDCS
20
[80%]
[80%] [80%]
[100%] 20
22
24
26
28
30
Mobile sink speed (m/s)
Variation range of mobile sink velocity (m/s)
FIGURE 21. End-to-end delivery delay vs. mobile sink velocity variation range.
increases a lot, however, the number of Response packet hop counts does not increase much due to the variation of mobile sink velocity. It is notable that the changing trend of Figs 21 and 22 are not completely the same, the reason being that due to the mobile sink velocity variation, there are some cases of the Response packet waiting for the mobile sink, which only increases the end-to-end delivery delay.
18
Avg. normalized sending cost (unit)
17.5 17 16.5 16 15.5 15 14.5 14 13.5 13 12.5 12 11.5
Mean Vs=20 m/s
11
Mean Vs=25 m/s
10.5 10
Mean Vs=30 m/s 0
2
4
6
8
10
12
14
16
Variation range of mobile sink velocity (m/s)
FIGURE 22. Normalized sending cost vs. mobile sink velocity variation range. 0.8
Mean Vs=20 m/s Mean Vs=25 m/s
Deviation from EMP (%)
0.7
FIGURE 24. End-to-end delivery delay vs. mobile sink velocity.
Mean Vs=30 m/s 0.6 0.5
6.3.4. Practical consideration In this test, we consider more realistic scenarios with irregular network topologies (even with holes) to better investigate the QBDCS. Ten different random topologies are generated by the setdest tool in ns-2. The modification coefficient γ is taken as 1. The value of the Tdeadline is set at 20 s, which means that the Response packet will be discarded if it cannot be delivered to the mobile sink within 20s. Fig. 24 reports the average end-to-end delivery delay and the packet delivery ratio of different schemes as the mobile sink velocity increases. It is shown that the ODBCS outperforms the ‘Naive’ scheme even with irregular network topologies. For a random topology, the ODBCS may experience larger end-to-end delivery delay due to the existence of ‘holes’; however, the ODBCS provides a higher packet delivery ratio, shorter delivery latency and, consequently, lower energy consumption on average.
740
745
750
755
0.4
6.4.
Discussion
760
0.3 0.2 0.1 0 0
2
4
6
8
10
12
14
16
Variation range of mobile sink velocity (m/s)
FIGURE 23. Deviation from the initial EMP vs. mobile sink velocity variation range.
In this paper, we assume that the MAC layer is not optimized (no synchronized duty cycles). Thus, the one-hop delay would be relatively large, in which our proposed scheme has significance. In our future work, we are planning to investigate the dutycycle scheduling problem in query-based WSNs. Besides, considering the limitation of sensor devices on processing power, the computation load of sensor nodes should be as low as possible. The investigation of efficient data aggregation for the on-demand cluster head is also our future work.
The Computer Journal, 2010
765
14 7. 770
775
780
785
790
L. Cheng et al. CONCLUSION
This paper deals with the problem of query-based data collection for mobile wireless monitoring applications, which is inspired by the recent research on WSNs applied in the area of wireless monitoring. It has two unique characteristics: (1) only partial sensor nodes are involved in a query-based data collection cycle; and (2) the mobile sink position is continuously changing during packet propagation. These two characteristics lead to the fact that existing routing protocols are not suitable for the mobile wireless monitoring applications. To address this problem, we tailored the routing protocol so as to minimize the energy consumption and delivery latency in a complete querybased data collection cycle. We made use of the predictable mobile sink mobility, and adopted the idea of estimating the velocity of packets and the packet–sink meeting position. We presented an MPAR protocol which can route the Response packet toward the predicted meeting position directly. We then proposed an efficient QBDCS based on MPAR. The QBDCS makes the decision on when to inject the Query packet for the mobile sink. Extensive simulations have been carried out for performance evaluation. The experimental results show that the proposed QBDCS achieves the least energy consumption, delivery latency and higher packet delivery ratio to complete a query-based data collection cycle.
FUNDING
795
800
This work was supported by National Natural Science Foundation of China (NSFC) under grants No. 60673178 and No. 60873241, and Hi-Tech Research and Development Program of China under grants No. 2008AA01Z217 and No. 2009AA01Z210. The work of L.S. was partially supported by the Lion project supported by Science Foundation Ireland under grant no. SFI/08/CE/I1380 (Lion-2), and partially supported by the Grant-in-Aid for Scientific Research (S)(21220002) of the Ministry of Education, Culture, Sports, Science and Technology, Japan.
REFERENCES
805
810
815
[1] Romer, K. and Mattern, F. (2004) The design space of wirel. sensor networks. IEEE Wirel. Commun., 11, 54–61. [2] Wang, L. and Xiao, Y. (2006) A survey of energy-efficient scheduling mechanisms in sensor networks. Mobile Netw. Appl., 11, 723–740. [3] Deng, J., Han, Y.S., Heinzelman, W.B. and Varshney, P.K. (2005) Scheduling sleeping nodes in high density cluster-based sensor networks. Mobile Netw. Appl., 10, 825–835. [4] Hsin, C. and Liu, M. (2006) Randomly duty-cycled wireless sensor networks: dynamics of coverage. IEEE Trans. Wirel. Commun., 5, 3182–3192. [5] Ma, J., Chen, C. and Salomaa, J.P. (2008) MWSN for large scale mobile sensing. J. Signal Process. Syst., 51, 195–206.
[6] Wang, B., Xie, D., Chen, C., Ma, J. and Cheng, S. (2008) Deploying Multiple Mobile Sinks in Event-Driven WSNs. Proc. Int. Conf. Communications (ICC’08), Beijing, CN, China, May 19–23, pp. 2293–2297. IEEE, Washington, DC, USA. [7] Ekici, E., Gu, Y. and Bozdag, D. (2006) Mobility-based communication in wireless sensor networks. IEEE Commun. Mag., 44, 56–62. [8] Jain, S., Shah, R.C., Brunette, W., Borriello, G. and Roy, S. (2006) Exploiting mobility for energy efficient data collection in wireless sensor networks. Mobile Netw. Appl., 11, 327–339. [9] Song, L. and Hatzinakos, D. (2007) Architecture of wireless sensor networks with mobile sinks: Sparsely deployed sensors. IEEE Trans. Veh. Technol., 56, 1826–1836. [10] Jayaraman, P.P., Zaslavsky, A., and Delsing, J. (2007) Sensor Data Collection Using Heterogeneous Mobile Devices. Proc. Int. Conf. Pervasive Services (ICPS’07), Istanbul, Turkey, July 15– 20, pp. 161–164. IEEE, Washington, DC, USA. [11] Cheng, L., Chen, Y., Chen, C. and Ma, J. (2009) Query-Based Data Collection in Wireless Sensor Networks with Mobile Sinks. Proc. 5th Int. Wireless Communications and Mobile Computing Conference (IWCMC’09), Leipzig, Germany, June 21–24, pp. 1157–1162. ACM, New York, NY, USA. [12] Tubaishat, M., Zhuang, P., Qi, Q. and Shang, Y. (2009) Wireless sensor networks in intelligent transportation systems. Wirel. Commun. Mob. Comput., 9, 287–302. [13] Jardosh, A., Belding-Royer, E.M., Almeroth, K.C. and Suri, S. (2003) Towards Realistic Mobility Models for Mobile Ad Hoc Networks. Proc. 9th Annual Int. Conf. Mobile Computing and Networking (MobiCom’03), San Diego, CA, USA, September 14–19, pp. 217–229. ACM, New York, NY, USA. [14] Camp, T., Boleng, J. and Davies, V. (2002) A survey of mobility models for ad hoc network research. Wirel. Commun. Mob. Comput., 2, 483–502. [15] He, T., Krishnamurthy, S., Stankovic, J.A., Abdelzaher, T., Luo, L., Stoleru, R., Yan, T., Gu, L., Hui, J. and Krogh, B. (2004) Energy-Efficient Surveillance System Using Wireless Sensor Networks. Proc. 2nd Int. Conf. Mobile Systems, Applications, and Services (MobiSys’04), Boston, MA, USA, June 6–9, pp. 270–283. ACM, New York, NY, USA. [16] Mainwaring, A., Culler, D., Polastre, J., Szewczyk, R. and Anderson, J. (2002) Wireless Sensor Networks for Habitat Monitoring. Proc. 1st ACM Int. Workshop on Wireless Sensor Networks and Applications (WSNA’02), Atlanta, GA, USA, September 28, pp. 88–97. ACM, New York, NY, USA. [17] Steere, D.C., Baptista, A., McNamee, D., Pu, C. and Walpole, J. (2000) Research Challenges in Environmental Observation and Forecasting Systems. Proc. Annual Int. Conf. Mobile Computing and Networking (MobiCom’00), Boston, MA, USA, August 6– 11, pp. 292–299. ACM, New York, NY, USA. [18] Jafari, R., Encarnacao, A., Zahoory, A., Dabiri, F., Noshadi, H. and Sarrafzadeh, M. (2005) Wireless Sensor Networks for Health Monitoring. Proc. 2nd Annual Int. Conf. Mobile and Ubiquitous Systems: Networking and Services (MobiQuitous’05), San Diego, CA, USA, July 17–21, pp. 479–481. IEEE, Washington, DC, USA. [19] Karp, B. and Kung, H.T. (2000) Gpsr: Greedy Perimeter Stateless Routing for Wireless Networks.
The Computer Journal, 2010
820
825
830
835
840
845
850
855
860
865
870
Efficient Query-Based Data Collection for Mobile Wireless Monitoring Applications
875
[20]
880
885
[21]
[22]
[23] 890
895
900
[24]
[25]
[26] 905
910
[27] [28]
915
Proc. Annual Int. Conf. Mobile Computing and Networking (MobiCom’00), Boston, MA, USA, August 6–11, pp. 243–254. ACM, New York, NY, USA. Yu, Y., Govindan, R. and Estrin, D. (2001) Geographical and Energy Aware Routing: A Recursive Data Dissemination. Technical Report UCLA/CSD-TR-01-0023, UCLA Computer Science Department, Los Angeles, CA, USA. Shu, L., Zhou, Z., Hauswirth, M., Phuoc, D.L., Yu, P. and Zhang, L. (2007) Transmitting Streaming Data in Wireless Multimedia Sensor Networks with Holes. Proc. 6th Int. Conf. Mobile and Ubiquitous Multimedia (MUM’07), Oulu, Finland, December 12–14, pp. 24–33. ACM, New York, NY, USA. Shu, L., Zhang, Y., Zhou, Z., Hauswirth, M., Yu, Z. and Hynes, G. (2008) Transmitting and gathering streaming data in wireless multimedia sensor networks within expected network lifetime. Mobile Net. Appl., 13, 306–322. Seada, K., Zuniga, M., Helmy, A. and Krishnamachari, B. (2004) Energy-Efficient Forwarding Strategies for Geographic Routing in Lossy Wireless Sensor Networks. Proc. 2nd ACM Conf. Embedded Networked Sensor Systems (SenSys’04), Baltimore, Maryland, USA, November 3–5, pp. 108–121. ACM, New York, NY, USA. Lee, S., Bhattacharjee, B. and Banerjee, S. (2005) Efficient Geographic Routing in Multihop Wireless Networks. Proc. 6th ACM Int. Symp. Mobile Ad Hoc Networking and Computing (MobiHoc’05), Urbana-Champaign, IL, USA, May 25–28, pp. 230–241. ACM, New York, NY, USA. Taha, Z. and Liu, X. (2006) Investigation of Geographic Routing Metric in Sensor Networks. Proc. Int. Conf. Communications (ICC’06), Istanbul, Turkey, June 11–15, pp. 1071–1076. IEEE, Washington, DC, USA. Li, C., Hsu, W., Krishnamachari, B. and Helmy, A. (2005) A Local Metric for Geographic Routing with Power Control in Wireless Networks. Proc. Second Annual Conf. Sensor and Ad Hoc Communications and Networks (SECON’05), Santa Clara, CA, USA, September 26–29, pp. 229–239. IEEE, Washington, DC, USA. Fang, Q., Gao, J. and Guibas, L.J. (2006) Locating and bypassing holes in sensor networks. Mobile Netw. Appl., 11, 187–200. Heinzelman, W., Chandrakasan, A. and Balakrishnan, H. (2000) Energy-Efficient Communication Protocol for Wireless Microsensor Networks. Proc. 33rd Hawaii Int. Conf. System Sciences (HICSS’00), Hawaii, USA, January 4–7, pp. 3005– 3014. IEEE, Washington, DC, USA.
15
[29] Mandala, D., Du, X., Dai, F. and You, C. (2008) Load balance and energy efficient data gathering in wireless sensor networks. Wirel. Commun. Mob. Comput., 8, 645–659. [30] Ye, F., Luo, H., Cheng, J., Lu, S. and Zhang, L. (2002) A Two-Tier Data Dissemination Model for Large-Scale Wireless Sensor Networks. Proc. 8th annual Int. Conf. Mobile Computing and Networking (MobiCom’02), Atlanta, GA, USA, September 23–26, pp. 148–159. ACM, New York, NY, USA. [31] Al-karaki, J.N. and Kamal, A.E. (2004) Routing techniques in wireless sensor networks: a survey. IEEE Wirel. Commun., 11, 6–28. [32] Kulik, L., Tanin, E. and Umer, M. (2008) Efficient Data Collection and Selective Queries in Sensor Networks. Proc. 2nd Int. Conf. Geosensor Networks (GSN’06), Boston, USA, October 1–3, pp. 25–44. Springer, Heidelberg, Berlin. [33] Teng, R., Zhang, B. and Tan, Y. (2007) A Study of Localized OnDemand Data Collection In Sensor Networks. Proc. 15th IEEE Int. Conf. Networks (ICON’07), Adelaide, Australia, November 20–21, pp. 431–436. IEEE, Washington, DC, USA. [34] Tacconi, D., Carreras, I., Miorandi, D., Chlamtac, I., Chiti, F. and Rantacci, R. (2007) Supporting the Sink Mobility: A Case Study for Wireless Sensor Networks. Proc. Int. Conf. Communications (ICC’07), Glasgow, Scotland, June 24–28, pp. 3948–3953. IEEE, Washington, DC, USA. [35] Battelli, M. and Basagni, S. (2007) Localization for Wireless Sensor Networks: Protocols and Perspectives. Proc. Canadian Conf. Electrical and Computer Engineering (CCECE’07), British Columbia, Canada, April 22–26, pp. 1074–1077. IEEE, Washington, DC, USA. [36] Kumar, S., Lai, T.H. and Balogh, J. (2004) On k-coverage in a Mostly Sleeping Sensor Network. Proc. Tenth Annual Int. Conf. Mobile Computing and Networking (MobiCom’04), Philadelphia, PA, USA, September 26–October 1, pp. 144–158. ACM, New York, NY, USA. [37] David C. Steere and Antonio Baptista and Dylan McNamee and Calton Pu and Jonathan Walpole (2006) Real-Time Power-Aware Routing in Sensor Networks. Proc. 14th IEEE Int. Workshop on Quality of Service (IWQoS’06), New Haven, CT, USA, June 19–21, pp. 83–92. IEEE, Washington, DC, USA. [38] Emad, F., Chang-Gun, L. and Eylem, E. (2006) Mmspeed: Multipath multi-speed protocol for qos guarantee of reliability and timeliness in wireless sensor networks. IEEE Trans. Mob. Comput., 5, 738–754.
The Computer Journal, 2010
920
925
930
935
940
945
950
955
960
