Energy-Efficient Multi-Path Routing with Short Latency and Low Overhead for Wireless Sensor Networks

Liming He Telecommunications Engineering School

Beijing University of Posts and Telecommunications Beijing, China

Email: limhe08@yahoo.com.cn

Abstract

In wireless sensor networks (WSNs), multi-path

routing can find multiple paths from source node to sink node for achieving high reliability and high energy-efficiency. However, the existing multi-path routing protocols in the literature construct multiple paths with long latency and high overhead [9] [10] [11]. In this paper, a novel multi-path routing protocol for WSNs is proposed, which can discover multiple paths with short latency and low overhead. Additionally, its energy efficiency is higher than that of the existing ones. Performance analyses and simulation results show that our proposed protocol has much better performance than the existing ones in terms of latency, overhead and network lifetime. 1. Introduction

In recent years, a special type of ad hoc networks, wireless sensor network (WSN), has attracted much attention of researchers [1]. WSNs consist of a large number of tiny low-power devices capable of performing sensing and communication tasks. The applications of WSNs are quite numerous, for example, emergency applications, environment monitoring, information gathering in battlefields, operating indoors for intrusion detection and so on.

In addition to the common characteristics of traditional wireless networks like mobile ad hoc networks, WSNs have some specific ones, such as very limited energy and bandwidth resources, high density of node deployment, and cheap and unreliable sensor nodes which are prone to failures. Due to these inherent characteristics that distinguish WSNs from other networks, routing issue in WSNs is very challenging. Recently, intensive research efforts have been focused on designing routing protocols specifically for WSNs, such as [2-8].

Of these routing protocols in the literature, some can find multiple paths from source node to sink node. Sensing data will be delivered along these paths at the same time. Two main benefits can be achieved by multi-path routing: First, as the transmission load is balanced among multiple paths, the energy of nodes is burned more equitably. Thus, energy-efficiency is increased and the network lifetime is prolonged. Second, multiple paths provide redundancy for data transmission, which increases the reliability of delivery.

C. Intanagonwiwat et al. proposed a popular data aggregation paradigm for WSNs called Directed Diffusion [9], which can find multiple paths from source node to sink node for transmitting the data at lower rate. In [10], D. Ganesan et al proposed a novel braided multipath scheme, which results in several partially disjoint multipath schemes. In [11], B. Deb et al proposed a multipath protocol called ReInForM to support information awareness in sensor networks.

However, the existing multi-path routing protocols in the literature construct multiple paths with long latency and high overhead, which greatly decreases the applicable value of them. In this paper, a novel multi-path routing protocol for WSNs is proposed, which can discover multiple paths with short latency and low overhead. Additionally, its energy efficiency is higher than that of the existing ones. Our proposed protocol is composed of three phases: double routing trees construction, route discovery, data transmission.

The rest of this paper is organized as follows. Section 2 proposes the system model. Our proposed protocol is presented in Section 3. Performance evaluations are given in Section 4. Finally, the conclusion is made. 2. System model

In this paper, the following assumptions are made. The WSN is composed of a base station (BS) and a set

Eighth ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing

0-7695-2909-7/07 $25.00 © 2007 IEEEDOI 10.1109/SNPD.2007.529

161

Eighth ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing

0-7695-2909-7/07 $25.00 © 2007 IEEEDOI 10.1109/SNPD.2007.529

161

Eighth ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing

0-7695-2909-7/07 $25.00 © 2007 IEEEDOI 10.1109/SNPD.2007.529

161

Eighth ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing

0-7695-2909-7/07 $25.00 © 2007 IEEEDOI 10.1109/SNPD.2007.529

161

of sensor nodes which are scattered in an area A. The network is data centric. Via some kind of time synchronization scheme, such as [12], time synchronization has been well performed in the whole network i.e., all the sensor nodes have uniform clock.

Each sensor node is energy constrained and has the knowledge of its energy. Each node has a unique identifier (ID) and can directly communicate with its immediate neighbors. A link layer protocol ensures this communication. Every link between two nodes is bidirectional.

3. Our proposed routing protocol

First, we define: Source node: The node that has some sensing data to

send Sink node: The node that is interested in some

sensing data Path: The route from source node to sink node Query: The message that is sent by sink node and is

to find search messages. Search: The message that is sent by source node and

is to find query messages. Find: The message that is sent to find source node. The proposed multi-path routing protocol is

composed of three phases: double routing trees construction, route discovery, data transmission.

3.1. Double Routing Trees Construction

Double routing trees construction phase constructs

two trees: query tree which is rooted at sink node and search tree which is rooted at source node. The construction of two trees begins and ends at the same time. The period of construction is preassigned, which determines the size of two trees, i.e. the number of nodes in two trees. After two trees have been constructed, there are some nodes which belong to both of two trees, called shared nodes. The longer the construction period of two trees is, the more the shared nodes are.

In each node of the network, there is a local register called register 1, which has the structure shown in Fig. 1a. The original values of the Query field, Search field and State field are 0. If a node enters the query tree, the Query field in its local register is set to 1. If it enters the search tree, the Search field is set to 1.

The process of query tree construction is as follows: First, one node of the network is interested in some

sensing data and becomes sink node. It enters query tree to become the root of the tree, and assign itself level 0. Then, sink node broadcasts query messages, whose structure is shown in Fig. 1b. In a query

message, the level and ID of the node from which the message is sent are recorded. The Lifetime field is the end point of time of construction period which is preassigned. Another preassigned parameter is Threshold field which represents the threshold of residual energy in the nodes. Every neighbor of sink node receives a query message sent from sink node. It records the time of receiving the message in the Time Stamp field of its local register 1, compares the time and the Lifetime field of the message. If the former is earlier than the latter, it records the ID of sink node, assigns itself a level which is one greater than the Level of the message and sets the Query field to 1. Thus, the node enters query tree and becomes a child of sink node. Then, the node compares its residual energy value and the Threshold field of the message. If the former is larger than the latter, it adds 1 to the nm field of the message. Contrarily, it adds 1 to the nl field of the message. After that, the node checks the Search field of its local register 1. If the Search field is 0, it acts as follows: Ⅰ. Compares the nm and nl. If nm is larger than nl, it sets the State field to 1; Ⅱ. broadcasts the query messages with new Level, Father ID and nm or nl to its neighbors. If the Search field is 1, i.e. the node has entered search tree and becomes a shared node, it compares the nm and nl. If nm is less than nl, it multiplies the State field by 0. The above process continues. What should be noticed is that a node which has entered the query tree can receive a query message again. When this happens, the node neglects any such messages. If a node receives a query message later than the Lifetime of the message, it neglects this message, too. If no query messages are transmitted in the whole network, the process of query tree construction is complete. The pseudo codes of the process after a node receives a query message are shown in Table 1.

The process of search tree construction is similar to that of query tree except that query is substituted by search. Source node broadcasts the search messages at the same moment when sink node broadcasts the query messages. The Lifetime and Threshold fields of search messages are the same to those of query messages. If no query and search messages are transmitted in the whole network, the process of double routing trees construction is complete. The common nodes of two trees, so called shared nodes, are important for our proposed protocol and will be used in the route discovery phase.

Fig. 2 shows an example of double routing trees construction. For simplicity, only part of all the nodes in the network is drawn in Fig. 2. Node 0 is sink node, and Node 18 is source node. Green arrows represent query messages, and blue arrows represent search messages. Here, an example of neglected messages is

162162162162

given: The messages from Node 3 arrive at Nodes 0, 2 and 6, but they all have entered query tree. Thus, these messages are discarded, as the red arrows in Fig. 2 show. What should especially be noticed is that Node 6 is the neighbor of Nodes 2 and 3. Both Nodes 2 and 3 send messages to Node 6. But the message from Node 2 arrives at Node 6 first, so Node 6 becomes the child of Node 2.

Fig. 3 shows double routing trees constructed at last. The query tree and search tree both have four levels. The green nodes are of level 1. The blue nodes are of level 2. The red nodes are of level 3, and are shared nodes.

TABLE 1. THE PSEUDO CODES OF THE PROCESS AFTER A NODE RECEIVES A QUERY MESSAGE.

IF (Time Stamp ≤ Lifetime)

THEN

{

Records the ID of the node from which the message is sent.

Assigns itself a level which is one greater than the Level of the message.

Sets the Query field to 1.

IF (its residual energy value ≥ Threshold)

THEN nm = nm + 1

ELSE nl = nl + 1

IF (Search = 0)

THEN IF (nm > nl)

THEN

{

Sets the State field to 1.

Broadcasts the query messages with new Level, Father ID and nm or nl to its neighbors.

}

ELSE IF (nm < nl )

THEN Multiplies the State field with 0.

}

ELSE Neglects this message.

(a)

(b)

(c) (d)

Figure 1. (a) The structure of local register 1(b) The structure of query, search messages (c) The structure of find messages

(d) The structure of local register 2.

Figure 2. The process of double routing trees construction

Figure 3. The double routing trees constructed at last

163163163163

3.2. Route Discovery In this phase, multiple paths from source node to

sink node will be discovered. In double routing trees constructed in the first phase, for each shared node, one and only one path can be found from source node to this shared node then to sink node. One shared node can determine one path. The State field of local register 1 in one shared node represents the energy distribution in the path determined by this shared node. If the State field is 1, the path has more nodes with higher energy level, so is suitable for data transmission.

In each node of the network, there is the other local register called register 2, whose structure is shown in Fig. 1d. The process of route discovery is as follows:

First, each shared node checks the State field of its local register 1. If the State field is 1, this shared node sends a find message, which has only one field: Sender ID, as shown in Fig. 1c. The find message is delivered hop by hop along the reverse direction of search message transmission. When the message passes by a node, the node records the ID of the message sender in its local register 2, updates the message with its ID, then sends the message to next hop. Eventually, the find message arrives at source node. Thus, the half path from source node to the shared node is discovered. As the other half path from the shared node to sink node is a branch of query tree and has been found, the whole path from source node to sink node is discovered. Since multiple shared nodes can send find messages, multiple paths can be discovered. 3.3. Data Transmission

When source node has collected sensing data, it

sends the data along multiple paths discovered in the previous phase to sink node. 4. Evaluating our proposed protocol

In the process of double routing trees construction, query and search messages are transmitted by broadcasting from sink node and source node to exterior. We assume that the delivery speed of query and search messages is the same, and that the delivery speed of the messages in diverse directions is the same. Thus, query and search messages form two circles whose centers are sink node and source node respectively. As time goes, the two circles are larger and larger. When they intersect, the shared nodes emerge. The Lifetime field of query and search messages determines the size of the two circles,

Figure 4. The concept of our protocol and Directed Diffusion

thus determines the number of shared nodes. The longer the Lifetime is, the larger the two circles are, and the more the shared nodes are.

Fig. 4 shows the two circles whose radius is R. The distance between two centers of them, i.e. the distance between sink node and source node, is D. The two circles intersect and form a central angle of Φ. It is assumed that the density of nodes in the network is ρ, i.e. the number of nodes in unit area is ρ.

We will evaluate our proposed protocol by four parameters: the latency, the number of paths, the overhead and the network lifetime. 4.1. The latency of our proposed protocol

The latency of our proposed protocol is composed of the time of double routing trees construction and the time of route discovery. First, we analyze the former, and then we evaluate the whole latency through simulation.

The Lifetime field of query and search messages represents the time of double routing trees construction, which is the time during which the messages propagate from sink node and source node to the circumference and is defined as T. T can be divided into two parts: the propagation time in the space, TP, and the time during which the message settles at nodes, TS. So, we have: P ST T T= + (1)

First, we calculate TS. In one circle of Fig. 4, we focus on a sector whose central angle is α. The number of nodes in this sector is:

2

0 0

12

Rn dr r d R

αρ θ ρα= ⋅ ⋅ =∫ ∫ (2)

During the period of T, one message passes by h nodes, i.e. the number of nodes in one radius of the circle is h. Then, we have:

164164164164

0

n hd hα

θ α= =∫ (3)

From (2) and (3), we get:

212

h Rρ= (4)

At each node, the time taken by the message is divided into the receive time, tR, and the send time, tS. So ( 1) ( 1)S R ST h t h t= − + − (5)

Now, we calculate TP. The propagation speed of query and search messages in the space is defined as V. Thus, we have:

PRTV

= (6)

Then

2 21 1[ ( 1) ( 1) ]2 2R S

RT R t R tV

ρ ρ= + − + − (7)

As ρ, tR and tS are all fixed, R is determined by T, thus by the Lifetime field of query and search messages.

Through simulation, we compare the latency of our proposed protocol with that of Directed Diffusion [9] and the protocol in [10], as shown in Fig. 5. It can be seen that the latency of our protocol is shorter than that of both Directed Diffusion and the protocol in [10]. Especially, the improvement for our protocol to the protocol in [10] is very large.

0 50 100 150 200 250 300

20

40

60

80

100

120

140

160

180

200

220

the

late

ncy

(mill

isec

ond)

the number of all the nodes

Our Proposed Protocol Directed Diffusion The Protocol in [10]

Figure 5. The latency versus the number of all the nodes

4.2. The number of paths The central angle of Φ is:

2 arccos( )2DR

φ = (8)

The length of the chord corresponding to the central angle of Φ is derived as:

2 24L R D= − (9)

Thus, the number of shared nodes, i.e. the maximum number of paths, is:

2 24S R Dρ= − (10)

It is assumed that the probability for that the State field in one shared node equals 1 is p. Thus, the number of paths discovered in route discovery phase will be:

2 2 24 4( ) 1RQ p R D p DD

ρ ρ= − = − (11)

4.3. The overhead of our proposed protocol

We evaluate the overhead of our proposed protocol by the total number of messages transmitted in the network. Averagely, every node in the two circles of Fig. 4 only transmits one message. Thus, the total number of messages equals the number of nodes in the area covered by the two circles, which is defined as NT and is derived as:

2 21 12 [( ) sin ]2 2 2TN R R DR φρ π φ= − + (12)

From (8), (9) and (12), we get: 2 2 2 212 {[ arccos( ) ] 4 }

2 4TDN R R D R DR

ρ π= − + − (13)

If messages are only transmitted from sink node by broadcasting like in Directed Diffusion [9], a single circle whose radius is D will be constructed, as shown in Fig. 4. We assume that every node still transmits only one message. Then, the total number of messages under this condition is: 2

SN Dρπ= (14)

We compare the total number of messages in our proposed protocol and that in Directed Diffusion [9], as shown in Fig. 6. It can be seen that the number of messages increases as the D/R increases in both two protocols. But the increasing speed of our protocol is much smaller than that of Directed Diffusion [9].

165165165165

When D/R is about 1.33, two lines intersect. We can adjust the Lifetime field of the messages to make the D/R be larger than 1.33. Then, NT is always smaller than NS, i.e. the overhead of our protocol is less than that of Directed Diffusion.

4.4. Network Lifetime

Energy efficiency of our protocol is evaluated by

the network lifetime. Through simulation, we compare the network lifetime using our proposed protocol with that using Directed Diffusion [9] and the protocol in [10], as shown in Fig. 7. In Fig. 7, we can see that the network lifetime using our protocol is much longer than that using other protocols.

5. Conclusion

In this paper, a novel multi-path routing protocol for WSNs is proposed to discover multiple paths with short latency, low overhead and high energy efficiency.

1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 23

4

5

6

7

8

9

10

11

12

13

D/R

the

tota

l num

ber

of m

essa

ges

(R=

1,=

1) our proposed protocolDirected Diffusion

Figure 6. The total number of messages versus D/R

0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 3200

20406080

100120140160180200220240260280300320340360380400

netw

ork

lifet

ime

(min

utes

)

the number of all the nodes

Our proposed protocol Directed Diffusion The protocol in [10]

Figure 7. The network lifetime versus the number of all the nodes

Our proposed protocol is composed of three phases: double routing trees construction, route discovery and data transmission. Mathematical analyses and simulation results show the much better performance of our proposed protocol than that of the existing ones. 6. References [1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam and E.

Cayirci, “A Survey on Sensor Networks,” IEEE Communication Magazine, vol. 40, no. 8, pp. 102-114, August 2002.

[2] R. C. Shah and J. Rabaey, “Energy Aware Routing for Low Energy Ad hoc Sensor Networks,” Proc. of IEEE WCNC 2002, vol. 1, March 2002.

[3] Y. Yu, D. Estrin and R. Govindan, “Geographical and Energy-Aware Routing: A Recursive Data Dissemination Protocol for Wireless Sensor Networks,” UCLA CompSci. Dept. tech. rep., UCLA-CSD TR-010023, May 2001.

[4] D. Braginsky and D. Estrin, “Rumor Routing Algortithm for Sensor Networks,” Proc. of 1st Wksp. Sensor Networks and Apps., Atlanta, GA, October 2002.

[5] J. Liu, Feng Zhao and D. Petrovic, “Information-directed Routing in Ad hoc Sensor Networks,” IEEE JSAC, vol. 23, no. 4, pp. 851-861, April 2005.

[6] Yu He and C. S. Raghavendra, “XVR: X Visiting-pattern Routing for Sensor Networks,” Proc. of INFOCOM 2005, vol. 3, pp. 1758-1769, March 2005.

[7] Y. T. Hou, Yi Shi, J. H. Reed and K. Sohraby, “Flow Routing for Variable Bit Rate Source Nodes in Energy-constrained Wireless Sensor Networks,” Proc. of ICC 2005, vol. 5, pp. 3057-3062, May 2005.

[8] Q. Fang, J. Gao, L. J. Guibas, V. de Silva and L. Zhang, “GLIDER: Gradient Landmark-Based Distributed Routing for Sensor Networks,” Proc. of INFOCOM 2005, vol. 1, pp. 339-350, March 2005.

[9] C. Intanagonwiwat, R Govindan and D. Estrin, “Directed Diffusion: a Scalable and Robust Communication Paradigm for Sensor Networks,” Proc. of ACM MobiCom 2000, pp. 56-67, Boston, MA, 2000.

[10] D. Ganesan, R. Govindan, S. Shenker and D. Estrin, “Highly-Resilient, Energy-Efficient Multipath Routing in Wireless Sensor Networks,” Mobile Computing and Communication Review (MC2R), vol. 1, no. 2.

[11] B. Deb, S. Bhatnagar, B. Nath, “ReInForm: Reliable Information Forwarding using Multiple paths in sensor networks,” Proc. of 28th Annual IEEE Conf. on Local Computer Networks (LCN), October 2003.

[12] Liming He and Geng-Sheng (G.S.) Kuo, “A Novel Time Synchronizaiton Scheme in Wireless Sensor Networks,” Proc. of IEEE VTC 2006 Spring, May 7-10, 2006.

166166166166