You are on page 1of 6

Delay-Efcient Data Aggregation Scheduling in Duty-cycled Wireless Sensor Networks

Nguyen Phan Khanh Ha


College of Information and Communication Engineering Sungkyunkwan University, Korea

Vyacheslav Zalyubovskiy
College of Information and Communication Engineering Sungkyunkwan University, Korea

Hyunseung Choo

npkha@skku.edu

slava@ece.skku.ac.kr

College of Information and Communication Engineering Sungkyunkwan University, Korea

choo@ece.skku.ac.kr

ABSTRACT
Data aggregation is an essential operation in wireless sensor networks (WSNs) in which sensed data are aggregated and transmitted to the sink. In many applications, reducing the latency of data aggregation is an important target. In addition, one of the primary challenges in WSNs is energy scarcity and reducing energy consumption is a problem. Recently, duty cycling, i.e., periodically switching on and o communication and sensing capabilities, has been considered to signicantly reduce the sensors energy consumption and extend a network lifetime. In this paper, we consider the minimum-latency aggregation scheduling problem in dutycycled WSNs. We propose a Delay-Ecient Data Aggregation Scheduling (DEDAS) scheme to generate a collisionfree schedule and minimize the delay for data aggregation in duty-cycled WSNs. Our analysis and comprehensive simulation results indicate that our solution performs better than existing schemes.

Categories and Subject Descriptors


C.2.1 [Computer-communication Networks]: Sensor Network

General Terms
Algorithms

Keywords
Data Aggregation Scheduling, Duty Cycle, Wireless Sensor Networks, Minimum-latency

1.

INTRODUCTION

Wireless sensor networks (WSNs) consist of a large number of sensor nodes with sensing and communication capabilCorresponding author.

ities. In many applications of WSNs such as military surveillance [1], habitat monitoring [2], or structure maintenance [3], data aggregation is a key functionality in which a sink collects only critical data from other nodes and intermediate nodes combine all received packets with their own packets into a single packet according to some aggregation functions (e.g. SUM, MAX, MIN, AVERAGE, and so on) [4]. The minimum-latency aggregation scheduling (MLAS) problem that nds the fastest collision-free schedule for data aggregation has been investigated in many previous literatures because in many applications, such as real-time monitoring and battleeld surveillance, the aggregation delay is the crucial performance metric [5]. Recently, the dutycycling technique where each sensor works for a time duration and sleeps for the remaining time in each working period has been examined as an ecient solution to save a large amount of energy because this method can signicantly decrease the active time of sensors. However, sleep latency [6], i.e., the time spent waiting for receivers to wake up, brought by the duty cycle, results in a substantial increase of data aggregation latency. Therefore, the MLAS problem in duty-cycled WSNs has emerged as a new problem. However, not much research has been done on this problem. In this paper, we consider the MLAS problem in dutycycled WSNs and propose a Delay-Ecient Data Aggregation Scheduling (DEDAS) scheme. The scheme contains two phases. In the rst phase, a balanced shortest path tree is built as the aggregation tree. In the second phase, based on this tree, we propose a scheduling algorithm to generate a collision-free schedule for data aggregation in dutycycled WSNs. Analysis and extensive simulation results show that our proposed scheme outperforms the previous works in terms of data aggregation latency. The rest of this paper is organized as follows. Section 2 introduces related work. In section 3, we present the network model and problem formulation. Our proposed scheme is shown in Section 4. In Section 5, simulation results are provided. Finally, we conclude the paper in Section 6.

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for prot or commercial advantage and that copies bear this notice and the full citation on the rst page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specic permission and/or a fee. RACS12 October 23-26, 2012, San Antonio, TX, USA. Copyright 2012 ACM 978-1-4503-1492-3/12/10 ...$10.00.

2.

RELATED WORK

Data aggregation in sensor networks has been well studied in recent years. There are many studies on data aggregation with dierent targets such as minimizing energy consumption, maximizing fairness, or minimizing latency [5]. In this section, we review the related work that considers the minimizing data aggregation latency problem.

Minimizing the latency of data aggregation is an important objective in WSNs. The MLAS problem is to design the transmission schedule for all sensors such that there is no conict between any two concurrent transmissions and the total number of time slots for all data to reach the sink is minimized. The MLAS problem is proved NP-hard and an algorithm Shortest Data Aggregation with the latency bound ( 1)R is proposed in [7], where is the maximum degree of nodes in a network graph and R is the network radius. Huang et al. [8] propose the First-Fit Scheduling algorithm based on the Maximal Independent Set with the latency bound 23R + 18. The authors in [9] propose an Improved First-t Scheduling (IFFS) algorithm and prove that their algorithm has the smaller latency bound 16R + 11 by reducing the number of connector nodes and nding the maximal non-conicting transmission schedule. In [10], Wan et al. propose the Enhanced Pipelined Aggregation Scheduling (E-PAS) algorithm with the aggrega(R) ))R + . The tion schedule latency is at most (1 + O( log 3 R main idea of E-PAS algorithm is similar to that of First-Fit Scheduling algorithm. However, instead of being xed, parent nodes are dynamically determined during the scheduling process based on the minimum cover set. All aforementioned schemes are centralized. There are also distributed approaches proposed in [11] and [12]. Although there are many studies on MLAS problem, they are not suitable for duty-cycled WSNs since the operation of nodes in such networks is dierent from that of conventional WSNs. To the best of our knowledge, paper [13] is the rst work on the MLAS problem in duty-cycled WSNs. The authors show that the MLAS problem in duty-cycled WSNs is NP-hard and propose an approximation Scheduling Algorithm (SA) to address this problem. The SA contains two phases: layered structure construction and working period scheduling. In the rst phase, a layered structure for aggregation is created using a maximal independent set. In the second phase, all nodes in the network are scheduled layer by layer using a sub-procedure minimal covering schedule. Although the SA can generate a collision-free schedule, the data aggregation delay is still high. In this paper, we investigate the MLAS problem in duty-cycled WSNs and propose a novel delay-ecient approach to reduce the data aggregation delay.

Table 1: List of frequently used symbols N (u ) Set of node us neighbor nodes A(u) Active time slot of u p (u ) Parent node of u in the aggregation tree u The number of children of u in the aggregation tree d(u, p(u)) Sleep latency between u and p(u) sch(u) Schedule for us transmission Wj Data aggregation schedule in round j W = Wj Data aggregation schedule S Set of nodes have already been scheduled to send data Rj Set of nodes have already been scheduled to receive data in round j R = Rj Set of nodes have already been scheduled to receive data

node u is active in one of T time slots, called A(u), and is sleeping in the remaining T 1 time slots in a working period. The duty cycle is dened as the ratio of the active time and the entire working period. Fig. ?? shows the working schedule of a sensor node u in duty-cycled WSNs with the active time slot is 2 and duty cycle is 20 % (T = 5). We assume that each node has only one data packet to send. Each node can send data in any time slot but only receive data in its active time slot. A node combines data from all its child nodes with its own data and then sends to its parent node. A sink node s V collects data from all other nodes. The aggregation process is complete when all data packets in the network are sent to the sink.

3.2

Problem formulation

3. 3.1

PRELIMINARIES Network model

We model the sensor network as a unit-disk graph G = (V, E ), where V is the set of nodes and E is the set of edges. An edge (u, v ) V if and only if node u is in the transmission range of node v . The whole lifetime of each node is divided into multiple working periods with the same length T time slots. Each

We formulate the MLAS problem in duty-cycled WSNs which is considered in [13] as follows. Given a sensor network G = (V, E ), we assign a working period M and a receiving node v for each node u V such that u can send data to v in working period M without collisions and the data aggregation delay is minimized. The collision occurs at a node if it receives two or more packets at the same time. We will discuss the collision in duty-cycled WSNs further in Subsection 4.3. In this paper, the schedule of node u is denoted as sch(u) = < M, v > which means node u is scheduled to send data to node v in v s active time slot A(v ) of working period M . We have to nd the collision-free schedule for all nodes in the network that minimize the data aggregation delay, i.e., minimize the required number of scheduled working periods for the sink to receive all data packets in the network.

4.

PROPOSED SCHEME

u 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 ...
T=5 Active time slot Sleep time slot

In this session, we show our proposed scheme, called Delay-Ecient Data Aggregation Scheduling in duty-cycled WSNs (DEDAS). First, the overview of DEDAS is presented. Then the detailed description of the proposed scheme is shown in Subsections 4.2 and 4.3. Finally, we provide some analysis of DEDAS in Subsection 4.4.

4.1
Figure 1: Working schedule of a sensor in dutycycled WSNs.

Overview

The proposed scheme consists of two phases: 1) data aggregation tree construction and 2) data aggregation schedul-

ing. In the rst phase, a shortest path tree (SPT) is constructed by a Dijkstra algorithm with the sleep latency between nodes as the distance metric. Then a Balancing SPT procedure is carried out to reduce the number of high-degree nodes in the network. In the second phase, based on the aggregation tree constructed in the previous phase, a collisionfree schedule is generated. Table 1 shows the symbols that we use in DEDAS.

u u z v z y v
Aggregation tree link Data link

v 0 1 2 3 4 0 1 2 3 4 z 0 1 2 3 4 0 1 2 3 4 y u 0 1 2 3 4 0 1 2 3 4 y 0 1 2 3 4 0 1 2 3 4
T=5 Active time slot Sleep time slot

Figure 2: An example of Balancing SPT procedure. its previous parent node is decreased when a node selects a sibling node as its new parent node, nodes cannot transmit sequentially in the same working period. Therefore, the aggregation delay is unchanged or increases. Case I : if node v is a sibling node of z , z is the parent node of y , and A(y ) < A(v ) < A(z ), then node v selects z as its new parent node and y selects v as its new parent node. In this case, nodes y and v can transmit concurrently in the same working period. Case II : if node v is a leaf sibling node of z and A(v ) < A(z ), then node v selects z as its new parent node. Because v and z can transmit simultaneously in the same working period, we need only one working period if A(p(z )) > A(z ) or two working periods if A(p(z )) A(z ) for the parent node p(z ) of z to receive all data from v and z . Without this procedure, the required number of working periods for p(z ) to receive all data from v and z is two in both cases. Case III : if node v is a sibling node of z , p(z) z > 1, and A(v ) < A(z ), then node v selects z as its new parent node. In this case, since p(z) z > 1, the degree of p(z ) is decreased by 1 when v selects z as its parent node. Additionally, v and z can transmit concurrently in the same working period because A(v ) < A(z ). Hence, the required number of working periods for p(z ) to receive all data from v and z is decreased by 1. In this paper, we propose two schemes: Delay-Ecient Data Aggregation Scheduling without Balancing SPT procedure (DEDAS) and Delay-Ecient Data Aggregation Scheduling with Balancing SPT procedure (DEDAS-B). The only dierence between DEDAS and DEDAS-B is in DEDAS-B, we utilize the Balancing SPT procedure after building the SPT to construct the data aggregation tree.

4.2

Data Aggregation Tree Construction

In this phase, a SPT rooted at the sink is built with the sleep latency between nodes as the distance metric. The sleep latency between node u and v is calculated as follows: d(u, v ) = A(v ) A(u) A(v ) + T A(u) if A(v ) > A(u) if A(v ) A(u) (1)

Theorem 1. With a dened data aggregation tree, the aggregation delay in duty-cycled WSNs cannot be smaller than Tdelaymin = max(i ), where i is the number of children (degree) of node i (i = 1, 2, 3, . . . , N ) in the data aggregation tree. Proof. We prove Theorem 1 by contradiction. Let k be the node that has k = max(i ), i = 1, 2, 3, . . . , N . We assume that Tdelay < max(i ) = k . This means at least two child nodes of k can transmit in the same working period. However, this assumption contradicts the truth that at least k working periods are required for node k to receive all data from its child nodes without collision. Therefore, Tdelay < max(i ) = k must be incorrect and Tdelay max(i ) = k , i = 1, 2, 3, . . . , N . Based on Theorem 1, reducing the number of high-degree nodes is important to minimize the data aggregation delay. The SPT can work well in low-density networks. However, when the network density increases, many nodes in the network have the same parent node in the SPT. Because the data from child nodes of the same parent node cannot be transmitted concurrently, the aggregation delay increases in high-density networks. Therefore, we propose a Balancing SPT procedure to reduce the number of high-degree nodes. The main idea of the Balancing SPT procedure is that a node will select a sibling node (i.e., the node has the same parent node in the aggregation tree) as its new parent node such that a pipelined schedule is generated. By this method, the aggregation delay decreases because the number of highdegree nodes decreases and nodes can transmit consecutively in the same working period. Fig. 2 shows an example of the Balancing SPT procedure. The left side of the gure shows the topology and the right side shows the working schedule of nodes. Node v is the sibling of z . If node v selects z as its new parent node instead of u, the number of child nodes of u is decreased by 1. In addition, because A(v ) = 0, A(z ) = 1, and A(u) = 3, when node z is the parent node of v , we create a pipelined schedule, i.e., node v and z can transmit in the same working period. Consequently, we can save one working period for node u to receive all data from z , v , and y , compared with the case where v selects u as its parent node. Following this principle, we propose three cases in which the pipelined schedule is generated, thus the aggregation delay is reduced. In other cases, although the degree of

4.3

Data Aggregation Scheduling

After the data aggregation tree is built in the rst phase, each node in the network is scheduled to send data to its parent node such that its transmission schedule does not collide with other transmission schedules. The pseudo code of the Delay-Ecient Data Aggregation Scheduling (DEDAS) algorithm is shown in Algorithm 1. DEDAS runs in a number of rounds until V is empty, i.e., all nodes in the network except the sink s are scheduled. In each round, the following process is repeated. We dene two sets: F1j contains leaf nodes whose active time slot is smaller than that of their parent node and F2j contains leaf nodes whose active time slot is greater than or equal to that of parent node. Then F1j and F2j are sorted in non-decreasing order of the sleep latency. The set of leaf nodes Fj is the union of sorted sets F1j and F2j (lines 4-7).

When Fj is not empty, a leaf node u in the head of Fj is selected. The available working period cw(p(u)) for u to transmit data to p(u) is calculated by the Scheduling function (line 10). This available working period should be checked whether or not the transmission of u to p(u) in cw(p(u)) causes collisions with the transmission of other nodes in set S , using the Collision-free validation procedure (line 11). Here, S is the set of nodes that have already been scheduled to send data. After checking the collision-free criteria, if the transmission schedule of u does not lead to a collision, u is scheduled to transmit to p(u) in working period cw(p(u)) and the sets S , R, Rj , Fj , and G are updated. Otherwise, u is removed from Fj and will be considered for scheduling in the next round (lines 12-18). Algorithm 1 DEDAS(G, s, Data Aggregation Tree ) Input: G = (V, E ); sink node s V ; Data Aggregation Tree Output: Data aggregation schedule W 1: j 0; cw(i) 0, i V ; V V \ {s}; E E ; G = (V , E ); S ; R ; W 2: while V = do 3: j j + 1; Rj ; Wj 4: F1j Leaf node i V that has A(i) < A(p(i)) 5: F2j Leaf node i V that has A(i) A(p(i)) 6: Sort F1j and F2j in non-decreasing order of d(i, p(i)) 7: Fj = F1j F2j 8: while Fj = do 9: u The rst node in Fj 10: cw(p(u)) Scheduling function(u, p(u)) 11: if Collision-free validation(u, S, R, Rj , cw(p(u))) then 12: sch(u) cw(p(u)), p(u) 13: Wj Wj {sch(u)} 14: S S { u }; R R { p (u )} 15: Rj Rj {p(u)}; Fj Fj \ {u} 16: V V \ {u}; E E \ {link(u, p(u))} 17: else 18: Fj F j \ { u} 19: end if 20: end while 21: W W Wj 22: end while 23: return W

Algorithm 2 describes the Scheduling function. The updated value of cw(p(u)) depends on the active time slots and the current scheduled working periods of u and p(u). In case A(u) < A(p(u)), node u can transmit its recently received data to a parent node in the same receiving working period if cw(u) > cw(p(u)) (line 5). Otherwise, it should wait for the next working period of p(u) for its transmission (line 3). If A(u) A(p(u)), node u should also wait for the next working period of u (or p(u)) to transmit its received data (line 8). Collision is a challenge in duty-cycled WSNs. Assume that two or more nodes are sending in the same time slot of a working period. If there exists a node in their overlapped transmission area whose active time slot is equal to this time slot, this node cannot successfully receive any data. This situation is called collision. Collision occurs due to signal interference, which can be divided into two types: primary and secondary interference [14]. Fig. 3 shows two types of interference causing collision in duty-cycled WSNs. In Fig. 3(a), primary interference occurs when node u and x are scheduled to send data to v in the same working period 2. Hence, v will not receive any data from u and x. Secondary interference occurs when a node which is within the transmission range of other nodes is scheduled to send (receive) a packet in the same working period with its unintended receivers (senders). For example, in Fig. 3(b), there are two ongoing scheduled transmissions from u to v and from x to y in the same working period 2. Because A(v ) = A(y ) = 1 and v is in the transmission range of its unintended sender x, v will not receive successfully data from its intended sender u.
Data from node u
v u x

v u x

Data from node x Active time slot Sleep time slot

u 0 1 2 3 4 0 1 2 3 4 ... x 0 1 2 3 4 0 1 2 3 4 ...
Collision v 0 1 2 3 4 0 1 2 3 4 ...

u 0 1 2 3 4 0 1 2 3 4 ... x 0 1 2 3 4 0 1 2 3 4 ...
Collision

v 0 1 2 3 4 0 1 2 3 4 ... y 0 1 2 3 4 0 1 2 3 4 ...

(a) Primary Interference

(b) Secondary Interference

Figure 3: Two types of interference in duty-cycled WSNs. The Collision-free validation procedure is shown in Algorithm 3. The returned value of a Boolean variable b after carrying out Algorithm 3 indicates whether the collision exists. In order to check the collision-free schedule, initially, b is assigned to TRUE. First, we check whether the collision exists due to primary interference (lines 2-4). If the parent node p(u) of u belongs to Rj , the collision will occur when u is scheduled to transmit data to p(u) in this working period. Therefore, a FALSE value is returned (line 3). Next, the collisions due to secondary interference are checked and b is assigned to FALSE if any of the following cases occurs: There is any neighboring node s S of p(u) whose parent node p(s) has the same schedule (i.e., working period) and active time slot with p(u) (lines 5-9). There is any us neighboring node r R that has the same schedule and active time slot with p(u) (lines 1014).

Algorithm 2 Scheduling function(u, p(u)) Input: u; p(u) Output: cw(p(u)): The available working period for u to transmit data to p(u) 1: if A(u) < A(p(u)) then 2: if cw(u) cw(p(u)) then 3: cw(p(u)) cw(p(u)) + 1 4: else 5: cw(p(u)) cw(u) 6: end if 7: else 8: cw(p(u)) max{cw(p(u)), cw(u)} + 1 9: end if 10: return cw(p(u))

Algorithm 3 Collision-free validation(u, S, R, Rj , cw(p(u))) Input: u; S ; R; Rj ; cw(p(u)) Output: A Boolean variable b b = T RU E : No collision b = F ALSE : Collision exists 1: b TRUE 2: if p(u) Rj then 3: b FALSE 4: else 5: for each node s S do 6: if (p(u) N (s)) and (A(p(u)) = A(p(s))) and (cw(p(u)) = cw(p(s))) then 7: b FALSE 8: end if 9: end for 10: for each node r R do 11: if (r N (u)) and (A(p(u)) = A(r)) and (cw(p(u)) = cw(r)) then 12: b FALSE 13: end if 14: end for 15: end if 16: return b

DEDAS and DEDAS-B in terms of data aggregation delay. Because IFFS and E-PAS do not consider the duty cycle, we make a modication to adapt them to duty-cycled environment as shown in [13]: given the output schedule (v , M ) of a node u which means u sends its data to v in time slot M , we make u send data to v in A(v ) of M -th working period. All nodes are randomly deployed in a region of size 200m 200m and have the same transmission range. The sink is located in the corner. Each node randomly selects its active time slot within the working period. For each set of parameter congurations, the reported results are the average of 15 runs on random topologies.

5.2
5.2.1

Simulation Results
Variable Node Density

4.4

Analysis

In this part, rst, we prove that our proposed scheme DEDAS generates a collision-free schedule. Second, we show that DEDAS is a better approach to solve the MLAS problem in duty-cycled WSNs than other existing schemes. In DEDAS, the working period for transmitting data of each node is allocated in an incremental manner. In each round j , a leaf node u is scheduled to send data to p(u) in cw(p(u)) if and only if it satises the collision-free criteria. Otherwise, it cannot be scheduled in this round. In the next round, the value of cw(p(u)) is increased by 1 and the collision-free criteria are checked again. This procedure is repeated for all leaf nodes in each round. Thus, the set Sj that is added to S in each round always contains nodes such that their transmissions do not interfere with other transmissions of already scheduled nodes in S . Since S = Sj , the collision-free schedule is achieved.
j

In this experiment, we x the nodes transmission range r = 30m and the duty cycle at 10% (T = 10), and vary the node density = 0.0075 0.03 nodes/m2 . The node N density is calculated by = 2 , where N is the number of L nodes and L is the dimension of the region where sensors are deployed. Fig. 4 shows the data aggregation delay of DEDAS, DEDAS-B, IFFS, E-PAS, and SA when the node density is varied. The aggregation delay of DEDAS-B is about 2 to 4 times smaller than that of SA, DEDAS, E-PAS, and IFFS. For example, when the node density is 0.025 nodes/m2 , the aggregation delay of DEDAS-B, SA, DEDAS, E-PAS, and IFFS is 21, 47, 54, 84, and 82 working periods, respectively. DEDAS-B outperforms other schemes because it allows more nodes to transmit successively in the same working period by generating the pipelined schedule.

5.2.2

Variable Transmission Range

Previous works do not consider the duty cycle [9], [10] or only consider the active time slot of the receiver but do not consider the sleep latency between sender and receiver when scheduling a node [13]. However, this sleep latency aects data aggregation delay and the eect becomes greater in low duty cycles [6]. Because DEDAS considers the sleep latency between nodes when building the aggregation tree in the rst phase as well as scheduling a node in the second phase, the aggregation delay is decreased. In addition, in our proposed scheme, we generate a pipelined schedule and allow more nodes to transmit in the same working period by Algorithm 2 and Algorithm 3. Therefore, our proposed scheme can signicantly reduce the data aggregation delay.

Fig. 5 illustrates the number of working periods of ve algorithms when the node density is set to 0.015 nodes/m2 , the duty cycle is 10% (T = 10), and the transmission range is varied from 15 to 65m. When the transmission range increases, since the number of neighbors increases, the probability of a collision rises, leading to an increase of aggregation delay. In low-density networks, i.e., low transmission ranges, DEDAS outperforms SA and its performance is close to that of DEDAS-B. In high-density networks, high-degree nodes cause high aggregation delay due to primary interference. As a result, the eciency of DEDAS decreases steeply when the node density increases. DEDAS-B can overcome this drawback and signicantly reduce the data aggregation delay thanks to the Balancing SPT procedure. For instance, with the transmission range of 60m, the aggregation delay of DEDAS-B is only 30 working periods compared to that of SA, DEDAS, E-PAS, and IFFS is 57, 93, 136, and 127 working periods, respectively.

5.2.3

Variable Duty Cycle

5. 5.1

PERFORMANCE EVALUATION Simulation Environment

In this section, we evaluate the performance of our proposed scheme using MATLAB. We implement SA [13], IFFS [9], and E-PAS [10] to compare their performances to

In this experiment, we investigate the eect of a duty cycle on data aggregation delay. We x the node density at 0.025 nodes/m2 and transmission range at 30m. The duty cycles are 50%, 33.33%, 25%, 20%, 12.5%, 10%, 6.67%, 5%, 3.33%, 2%, 1.25%, and 1%, corresponding to T = 2, 3, 4, 5, 8, 10, 15, 20, 30, 50, 80, and 100. Fig. 6 shows that the required number of working periods of DEDAS-B, DEDAS, and SA decreases when the duty cy-

1 0 0 9 0

N u m b e r o f w o r k in g p e r io d s

N u m b e r o f w o r k in g p e r io d s

8 0 7 0 6 0 5 0 4 0 3 0 2 0 1 0 0

N u m b e r o f w o r k in g p e r io d s

S A IF F E -P D E D E

S A S D A S D A S -B

1 4 0 1 3 0 1 2 0 1 1 0 1 0 0 9 0 8 0 7 0 6 0 5 0 4 0 3 0 2 0 1 0 0 1 5 2 0

S A IF F E -P D E D E

9 0

S A S D A S D A S -B

8 0 7 0 6 0 5 0 4 0 3 0 2 0 1 0 S A IF F E -P D E D E S A S D A S D A S -B

0 .0 1 0

0 .0 1 5

0 .0 2 0

0 .0 2 5

0 .0 3 0

2 5

N o d e d e n s ity ( n o d e s /m 2 )

3 0 3 5 4 0 4 5 5 0 T r a n s m is s io n r a n g e ( m )

5 5

6 0

6 5

0 0

1 0

2 0

3 0

4 0

5 0

6 0

7 0

8 0

9 0

1 0 0

T ( tim e s lo ts )

Figure 4: Aggregation delay with Figure 5: Aggregation delay with Figure 6: Aggregation delay with variable node density. variable transmission range. variable duty cycle. cle decreases, i.e., the number of time slots in a working period T increases. The reason is that the increase of T causes nodes to have many opportunities to transmit concurrently without collision. DEDAS and DEDAS-B consider not only the active time slot of the receiver but also the sleep latency between the sender and receiver. Therefore, they achieve a smaller aggregation delay compared with other schemes in every duty cycle. [5] O. D. Incel, A. Ghosh, and B. Krishnamachari. Scheduling Algorithms for Tree-based Data Collection in Wireless Sensor Networks. In book chapter: Theoretical Aspects of Distributed Computing in Sensor Networks. Springer, 2011. [6] Y. Gu and T. He. Dynamic Switching-based Data Forwarding for Low-Duty-Cycle Wireless Sensor Networks. IEEE Transactions on Mobile Computing, 10(12): 1741 1754, December 2011. [7] X. Chen, X. Hu, and J. Zhu. Minimum Data Aggregation Time Problem in Wireless Sensor Networks. In Proceedings of International Conference on Mobile Adhoc Sensor Networks (MSN), pages 133142, December 2005. [8] S. C. -H. Huang, P. -J. Wan, C. T. Vu, Y. S. Li, and F. Yao. Nearly Constant Approximation for Data Aggregation Scheduling in Wireless Sensor Networks. In Proceedings of IEEE INFOCOM 07, pages 366372, May 2007. [9] M. Ren, L. Guo, and J. Li. A New Scheduling Algorithm for Reducing Data Aggregation Latency in Wireless Sensor Networks. International Journal of Communication, Network and System Sciences, 3(8): 679688, August 2010. [10] P. -J. Wan, S. C. -H. Huang, L. Wang, Z. Wan, and X. Jia. Minimum-latency Aggregation Scheduling in Multihop Wireless Networks. In Proceedings of MobiHoc 09, pages 185194, May 2009. [11] B. Yu, J. -Z. Li, and Y. Li. Distributed Data Aggregation Scheduling in Wireless Sensor Networks. In Proceedings of IEEE INFOCOM 09, pages 21592167, April 2009. [12] X. Xu, S. Wang, X. Mao, S. Tang, and X. Y. Li. A Delay Ecient Algorithm for Data Aggregation in Multi-hop Wireless Sensor Networks. IEEE Transactions on Parallel and Distributed Systems, 22(1): 163175, January 2011. [13] B. Yu and J. -Z. Li. Minimum-Time Aggregation Scheduling in Duty-cycled Wireless Sensor Networks. Journal of Computer Science and Technology, 26(6): 962 970, November 2011. [14] S. Ramanathan and E. L. Lloyd. Scheduling algorithms for multihop radio networks. IEEE/ACM Transactions on Networking, 1(2): 166167, April 1993.

6.

CONCLUSIONS

In this paper, we propose a novel delay-ecient scheme, DEDAS, to address the MLAS problem in duty-cycled WSNs. The objective is to generate the collision-free schedule that minimizes the aggregation delay. Through analysis and simulation, we show that our proposed scheme can reduce signicantly the aggregation delay compared with other schemes by taking into account the sleep latency between nodes as well as generating pipelined schedules. In the future, we will study a distributed approach to solve the MLAS problem in duty-cycled WSNs. Besides, we will also investigate other tree structures for data aggregation.

7.

ACKNOWLEDGMENTS

This research was supported in part by MKE(NIPA, KEIT) and MEST(NRF), Korean government, under IRTRC NIPA-2012-(H0301-12-3001), IT R&D Program [10041244, Smart TV 2.0 Software Platform], and PRCP(2011-0018397), respectively.

8.

REFERENCES

[1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. A survey on sensor networks. IEEE Communications Magazine, 40(8): 102114, August 2002. [2] R. Szewczyk, A. Mainwaring, J. Anderson, and D. Culler. An Analysis of a Large Scale Habitat Monitoring Application. In Proceedings of Sensys 04, pages 214226, November 2004. [3] N. Xu, S. Rangwala, K. Chintalapudi, and D. Ganesan. A Wireless Sensor Network for Structural Monitoring. In Proceedings of Sensys 04, pages 1324, November 2004. [4] S. Madden, M. J. Franklin, J. M. Hellerstein, and W. Hong. A Wireless Sensor Network for Structural Monitoring. In Proceedings of 5th Symposium on Operating Systems Design and Implementation (OSDI), November 2002.

You might also like