we propose a hybrid clustering and routing protocol (HCR) that considers both the ..... gorithm for wireless sensor networks,â in Global Telecommunications.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
Hybrid Clustering and Routing Strategy with Low Overhead for Wireless Sensor Networks Zhezhuang Xu, Chengnian Long, Cailian Chen, Xinping Guan Department of Automation, School of Electronic, Information and Electrical Engineering Shanghai Jiao Tong University, Shanghai, P.R.China Email: {jmmouse,longcn,cailianchen,xpguan}@sjtu.edu.cn
Abstract—Dynamic Clustering is an efficient topology management approach for sensor networks. Observing the fact that the clustering has tight relations with inter-cluster routing problem, we propose a hybrid clustering and routing protocol (HCR) that considers both the cluster head selection and routing discovery problems. Random backoff and gradient routing strategies are used to achieve our design goals with low overhead. Considering the limited transmission range bounded by the hardware, the clustered network generated by HCR is ensured to be connected. Simulation results demonstrate that our proposed strategies ensure the connectivity of the network and improve the energy efficiency for data transmission.
Fig. 1.
Relations between cluster head selection and routing discovery
I. I NTRODUCTION Dynamic clustering is an efficient topology management scheme for sensor networks [1]. Generally, the dynamic clustering protocol includes three phases: cluster head selection, cluster formation and routing discovery. The cluster heads are dynamically selected according to the real-time status of the network, then the clusters are formed based on the location of the selected cluster heads and the cluster heads discover their routes to the sink by the routing algorithm. Cluster head selection is one of the most important issues in dynamic clustering [2]. There have been many contributions to the problems of cluster head selection for dynamic clustering. HEED [3] proposed an iteration-based algorithm to select well-distributed cluster heads. Observing the fact that HEED has high complexity on message exchanges, the random backoff strategy was adopted in [4], [5] to design the cluster head selection algorithm with low overhead. However in the previous works, cluster head selection and routing discovery are treated as two isolated problems, i.e. the cluster heads are selected without any consideration on the routing problem, and then the routing protocol is used to build up the routes to the sink among selected cluster heads. In this case, the efficiency of inter-cluster communication can hardly be achieved. Take an example shown in Fig.1, for the unreasonable cluster head selection, the edge e in Fig.1(a) for inter-cluster communication is larger than that in the Fig.1(b). Based on this observation, we propose a joint design between cluster head selection and routing discovery. However, the realization of this idea has several challenges: 1. The connectivity problem with limited transmission range: For the inter-cluster communication is built among cluster heads, as shown in Fig.1, the connectivity does not only depend on the node placement, but also on the cluster head
selection. Most dynamic clustering protocols assume that the nodes have enough transmission power to keep the network connected [2], [3], [6]. However, in practice, the node has maximum transmission range RM AX which is bounded by the hardware capability. If there exists any edge e that is longer than RM AX , the network will be disconnected. 2. The complexity of the protocol: In dynamic clustering, the clustered topology rebuilds with the change of the environment, which leads to frequent routing discovery among cluster heads. If the clustering and routing protocol are too complex, the growth of overhead may overwhelm the benefits of dynamic clustering. To overcome these challenges, we propose a hybrid clustering and routing protocol (HCR) in this paper. In the HCR protocol, clustering and routing are completed simultaneously with consideration of both the cluster head selection and routing discovery problems. Random backoff [4], [5] and gradient routing [7], [8] strategies are used to design the protocol with low overhead. Through the cost field establishment and the cost information exchanged in the cluster head selection, the HCR ensures both connectivity and energy efficiency of the network. The remainder of this paper is organized as follows. Section II provides the network model and states the clustering and routing problem that we address in this work. Section III describes the HCR protocol and the property of the HCR protocol is analyzed in Section IV. In Section V, simulations are carried out to evaluate the performance of HCR. Finally, Section VI concludes this paper and provides directions for our future work.
978-1-4244-6404-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
II. P ROBLEM S TATEMENT A. Network Model Considering a set of nodes and a sink deployed in a sensing area, we assume the sensor network has the following properties: 1) The mission of the sensor network is gathering sensor data from sensor nodes to sink, such that the destination of every sensing data is the sink. 2) The network is organized into clusters. The cluster member sends sensing data to its cluster head directly and the cluster head transmits the data to its nexthop cluster head for relaying data to the sink. The connectivity problem discussed in this paper is focused on the connectivity among cluster heads. 3) Each node has a maximum transmission range noted as RM AX . This motivates the need for keeping connectivity with the limitation. 4) The critical transmitting range (CTR) is less than RM AX . The CTR is the length of the longest edge of the Euclidean minimum spanning tree [9], [10] built on the nodes, e.g. the edge e in Fig.1(b). If this assumption could not be satisfied, the network is doomed to be disconnected even though every node is selected to be cluster head and uses its maximum power for data transmission. This assumption could be satisfied by reasonable node placement [11]. It worths to note that there are no assumptions on: • Physical location-awareness; • Distance estimation among nodes; • Specific node placement; B. Clustering and Routing Problem In this paper, the dynamic clustering is referred to the change of the status. Every node begins with the status of IDLE. When there is data needed to be transmitted, the node becomes candidate (CA). The clustering algorithm is run to make the decision about whether to be cluster head (CH) or cluster member (CM) that belongs to a cluster. Then it could transmit its data to the sink through the routes built by the routing algorithm. When the mission is completed, the node could return to IDLE status. Our goal is to design a hybrid clustering and routing protocol (HCR) with joint consideration of cluster head selection and routing discovery. Specifically, the design goals are given as follows: 1) The clustering and routing discovery are hybrid. At the end of the protocol, each node is either a CH or a CM, and has selected a node as its next-hop relay. 2) The network is ensured to be connected, i.e. every node could transmit its data to the sink using limited transmission range bounded by RM AX . 3) Cluster heads are well-distributed over the sensor field to achieve the energy efficiency. 4) The algorithm should be completely distributed and efficient in terms of message exchange.
III. P ROTOCOL D ESCRIPTION The basic idea of hybrid clustering and routing is to select the set of CHs that not only organize well-balanced clusters, but also form up the connected and efficient backbone for inter-cluster communication. Gradient routing [7] provides a simple and efficient solution for distributed routing among CHs. By the cost field establishment algorithm, every node obtains its cost denoted as k, to establish the cost field that presents the direction through which the sink can be reached. The cost information is exchanged distributively among the nodes for making the decision of cluster head selection and routing discovery. In order to control the topology, each node has two transmission ranges: the clustering range Rc and the inter-cluster transmission range Rt . How to determine the value of Rc and Rt will be discussed in Section IV and V. To clarify the idea of hybrid clustering and routing, the protocol is described in a simple scenario that all nodes begin with CA status and then choose to be CM or CH. Moreover, the metrics for clustering and routing are set as simple as possible in this paper. The scenarios and metrics used in related works could be readily adopted in our proposed framework with minor modifications. A. Cost Field Establishment In this paper, the hop count to the sink is used as the cost. The cost field is established by flooding the advertisement (ADV) message which contains the cost information kADV . The process of cost field establishment is divided into slots: At the beginning, the sink generates the ADV message including its cost k = 0, and then broadcast it in the range Rt at slot 0. The node received this ADV message reads the kADV and sets its cost k as kADV + 1, i.e. k = 0 + 1. Then it will broadcast the ADV message with its cost k = 1 in Rt at slot 1. The nodes with the same cost k will broadcast ADV messages in the same slot, so the length of the slot should be set long enough to ensure that these nodes could have enough time to broadcast their ADV messages. This process repeats until all nodes have received ADV message and set their own cost. A node may receive more than one ADV messages. In this case, the node just handles the f irst ADV message, and ignores all the following ADV messages. If a node broadcasts ADV message in slot k and does not receive any ADV message in the slot k + 1, the node will send a report (RPT) message to the sink. By receiving the RPT messages, the sink could obtain the maximum cost K in the network. The value of K depends on the node deployment area and the range Rt . At the end of cost field establishment, the sink broadcasts another ADV message with maximum cost information K over the network. The cost field establishment operates only once hence does not impact the complexity of the HCR protocol. B. Cluster Head Selection The hybrid clustering and routing algorithm is driven by the random backoff scheme [4]. The node decides whether to be CH or CM according to its random backoff time tb
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
Algorithm 1 Cost Field Establishment 1: if Sink then 2: Broadcast ADV message (kADV = 0) at slot 0; 3: end if 4: if Receive the f irst ADV message then 5: k ← kADV + 1; 6: Broadcast ADV message (kADV = k) at slot k; 7: end if
and received ADV messages. There are two kinds of ADV messages used for cluster head selection: the ADV message for clustering (AC), and the ADV message for routing discovery (AT). Both of them are broadcasted by CH and carry the following information of the sender CH: ID, cost (k), residual energy (Er ), and N EXT flag which notes if the sender has found its next hop relay. According to the received messages, every node maintains the following data: • JOIN list: record the cluster head candidates; • SERV list: record the requests for relay from cluster heads. At the beginning, the CA sets up a random backoff timer tb according to its cost k and residual energy Er , EM AX − Er × Tslot (1) tb = (K − k) × Tslot + EM AX where K is the maximum cost in the network and EM AX is the initial energy of the node. The process of cluster head selection breaks into K slots whose length is Tslot . The first set of cluster heads are selected from the nodes whose k = K, and then go through the cost field until reach the sink. After setting the tb , the CA listens to the channel for incoming messages. When it receives AC message, it checks the kAC included in the message. If its cost k ≥ kAC , the CA will record the sender information to the JOIN list. It ensures the cost of the cluster head is no more than the cost of its cluster members. When the CA receives AT message with kAT > k, it will check the NEXT flag in the AT message. If NEXT flag is clear, which means that the sender (CH) has not found a forwarding CH, the CA will become the CH candidate and record the sender information to the SERV list. If NEXT flag is set and the sender information is already in the SERV list, the CA will delete the sender from the SERV list. The Eqn.1 ensures the CA could not receive any AT message with kAT < k. When the random delay timer terminates, the node checks the JOIN and SERV list. If the JOIN list is not empty and the SERV list is empty, the CA sets its status as CM. Otherwise, the status is set as CH. C. Routing Discovery When the CA is selected to be CH, it broadcasts AC message in Rc and AT message in Rt . According to Eqn.1, the CH could not found its next-hop relay in the (K-k)th slot, hence the first AT message is broadcasted with N EXT = 0. Then the CH waits for its next-hop appears. When the CH
Algorithm 2 Cluster Head Selection 1: Set up random backoff Timer tb ; 2: repeat 3: if Receive AC message AND k ≥ kAC then 4: Record sender to JOIN list; 5: end if 6: if Receive AT message AND k < kAT then 7: if N EXT = 0 then 8: Record sender to SERV list; 9: else if Sender in SERV list then 10: Delete sender from SERV list; 11: end if 12: end if 13: until tb terminates 14: if SERV is empty AND JOIN is not empty then 15: status ← CM ; 16: else 17: status ← CH; 18: end if
receives the AT message that k > kAT , the CH selects the sender as its relay and broadcasts the second AT message with the updated NEXT flag to announce that it finds a relay. For the CH whose cost k is 1, its next-hop relay is destined to be the sink. For the node decides to be CM, it chooses its CH which has the lowest cost and highest residual energy from the JOIN list, and then starts to transmit sensing data to its CH. IV. P ROTOCOL A NALYSIS The HCR protocol organizes the network into clusters whose size is determined by the clustering range Rc . And the cluster head uses the transmission range Rt to forward data to its next-hop relay. According to the results obtained in [12], to minimize the energy consumption, every cluster head should use its maximum transmission range for inter-cluster communication. With joint consideration of connectivity, the value of Rc and Rt should be constrained by, Rc ≤ Rt = RM AX
(2)
Assume the Eqn.2 holds, the HCR protocol meets the design goals listed in Section II, that is discussed as follows. Definition 1. (Forwarding node) Suppose node A with cost k, the node has cost k-1 is the forwarding node of node A. Lemma 1. Each node has at least one forwarding node within the range Rt . Proof: In the cost field establishment, the ADV message is flooding over the network by rebroadcasted in Rt . Because the CTR is less than Rt (Assumption 4 in Section II), every node can receive the ADV message. The cost of the node k is set as the kADV +1 when it receives ADV message at the first time, hence the node has at least one forwarding node which is the sender of its first received ADV message.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
TABLE I S IMULATION PARAMETERS
Lemma 2. At the end of the HCR protocol, each node is either a cluster head or a cluster member (goal 1).
Lemma 3. At the end of the HCR protocol, each node has select one CH as its next-hop relay. Specifically, each CH has a forwarding CH or the sink as its next-hop relay (goal 1). Proof: For CM, the next-hop relay is its CH. Assume that a CM does not find a CH, i.e. its JOIN list is empty, it will choose to be CH. It is a contradiction. For the CH with k = 1, the next-hop relay is the sink. For the other CHs, the CH firstly broadcasts AT message with NEXT=0 in Rt . According to Lemma 1 and Eqn.(1), there is at least one forwarding CA that can receive the AT message. The CA records the sender information to the SERV list. For the SERV list is not empty, the CA with the lowest tb will become CH that will be selected as the relay for the sender CH.
Type
Parameter
Value
Application
Initial energy Maximum Tx range Data Packet Size Round
2J 80 m 125 Bytes 20 TDM frames
Radio Model
Eelec f s Ef usion Path-loss exponent (α)
50 nJ/bit 10 pJ/bit/m2 5nJ/bit/signal 2
1
ratio of disconnection
Proof: A similar theorem was proven in [3]. Therefore, the proof is omitted.
120
40
50
60
70
80
70
80
BSC HCR
100 80 60 40 20
30
40
50
60
clustering range (Rc)
(b) Length of the longest edge 140
BSC HCR
120 100 80 60 40 20 0 20
30
40
50
60
70
80
clustering range (Rc)
(c) Number of cluster heads
V. P ERFORMANCE E VALUATION
Fig. 2.
A. Simulation Parameters The performance of HCR protocol is evaluated in the following simulation scenario: 400 nodes are randomly deployed over an area of size 300 ∗ 300m2 , and the sink resides in the center of the area. The protocol is run in a typical scenario that is divided into rounds which includes set-up phase and data transmission phase [2]. The network topology is reorganized in set-up phase at the beginning of each round, and the data transmission is scheduled by TDMA. The simulation parameters are similar to those in [3], which are summarized in Table I. The performance of HCR compares with backoff strategy clustering (BSC) [4] which use similar backoff strategy to
30
140
number of cluster heads
Proof: Conducting HCR, the CH broadcasts one AC message and two AT messages, and the CM needs to send one join message to its selected next-hop relay. In the worst case, every node is selected to be CH, hence the number of exchanged messages in the network is upper-bounded by 3∗n, i.e. the message complexity of HCR is O(n).
0.2
(a) Ratio of disconnection
Proof: A node changes its status and selects its nexthop relay according to its tb which is determined by its local information and received messages broadcasted by the CHs within the range Rc and Rt . Observation 2. The HCR protocol has a worst-case message complexity of O(n) in the network (goal 4).
0.4
clustering range (Rc)
length of CTR
Observation 1. The HCR protocol is completely distributed (goal 4).
BSC HCR
0.6
0 20
Theorem 1. The HCR protocol generates a connected graph (goal 2). Proof: According to Lemma 3, the data from a CH with cost k could be forwarded to the sink in k hops. The CM could transmits the data to its CH in one hop. Therefore, the network is connected.
0.8
Topology Evaluation
select CHs. For the BSC does not include routing protocol, a simplified gradient routing protocol [7] is adopted in BSC. RM AX and Rt are both set to 80m in the HCR prtocol, and the BSC does not have limitation on the transmission range. B. Connectivity Fig.2 compares the topology generated by HCR and BSC with the clustering range Rc adjusted from 20 to 80. Each protocol is run in 50 different deployment which has 20 rounds for selecting different set of CHs. Fig.2(a) shows the ratio
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
700
lifetime (rounds)
VI. C ONCLUSIONS
BSC HCR
600 500 400 300 200 100 0 20
30
40
50
60
70
80
clustering range (Rc)
(a) Network lifetime 6
energy cost per received packet (nJ)
4.5
x 10
4
BSC HCR
3.5 3 2.5
ACKNOWLEDGEMENT
2 1.5 1 0.5 0 20
In this paper, a hybrid clustering and routing protocol has been proposed to improve the efficiency of inter-cluster communication. Using random backoff and gradient routing schemes, the clustering and routing are completed simultaneously with low overhead. Theoretical analysis and simulation results show that HCR ensures the connectivity of the network and improves the energy efficiency for data transmission. For our future works, we will try to formulate the connectivity problem in the scenario of dynamic clustering, and provide a guideline for cluster head selection. Moreover, in the proof of Theorem 1, we find that the data transmission completes in limited hops. It is an interesting work to study the real-time performance of the HCR protocol.
30
40
50
60
70
80
clustering range (Rc)
(b) Energy consumption per received packet at sink Fig. 3.
Energy Efficiency Evaluation
of disconnection in the network, where the disconnection is calculated by the percentage of rounds in which the topology exists any edge that is longer than RM AX . In BSC, the connectivity can hardly be achieved when the Rc grows larger than 40. On the other hand, the topology generated by HCR is ensured to be connected, and the connectivity has no relation with Rc . The reason could be found in Fig.2(b). The length of the longest edge (CTR) in BSC increases with the growth of Rc , while that in HCR is tightly bounded by Rt . It is because that there are more cluster heads generated in HCR to ensure connectivity, as shown in Fig.2(c).
C. Energy Efficiency Fig.3(a) compares the network lifetime with HCR to BSC, where network lifetime is the time until the first node dies. In BSC, the network lifetime increases with the growth of Rc , because data aggregation reduces the amount of data for inter-cluster communication. The Rc has less influence on the performance of HCR, such that the network lifetime is shorter than that in BSC when the Rc is longer than 40. It is because the cluster head selection is determined by both Rc and Rt . The HCR sacrifices the network lifetime to ensure the connectivity of the network. Fig.3(b) shows the comparison of energy consumption per packet received at the sink. The energy consumption increases when the Rc grows up. With joint consideration of the routing problem, the HCR has better routes for inter-cluster communication, hence the energy consumption in HCR is much more efficient than that in BSC.
The authors wish to thank the anonymous reviewers for their comments and suggestions to improve the presentation of this paper. We would like to thank the financial support from National Basic Research Project of China (Project Number, No.2010CB731800); NSF of China under the grant No.60974123, 60604012; and Science and Technology Commission of Shanghai Municipality under the grant No. 08511501600. R EFERENCES [1] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “A survey on sensor networks,” IEEE Commun. Mag., vol. 40, no. 8, pp. 102–105, 2002. [2] W. B. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan, “An application-specific protocol architecture for wireless microsensor networks,” IEEE Trans. Wireless Commun., vol. 1, no. 4, pp. 660–670, 2002. [3] O. Younis and S. Fahmy, “Heed: A hybrid, energy-efficient, distributed clustering approach for ad hoc sensor networks,” IEEE Trans. Mobile Comput., vol. 3, no. 4, pp. 366–379, 2004. [4] Y. Cao and C. He, “A distributed clustering algorithm with an adaptive backoff strategy for wireless sensor networks,” IEICE Transactions on Communications, vol. E89-B, no. 2, pp. 609–613, 2006. [5] S. Fang, S. Berber, and A. Swain, “An overhead free clustering algorithm for wireless sensor networks,” in Global Telecommunications Conference, 2007. GLOBECOM ’07. IEEE, Nov. 2007, pp. 1144–1148. [6] S. Soro and W. B. Heinzelman, “Cluster head election techniques for coverage preservation in wireless sensor networks,” Ad Hoc Networks, vol. 7, no. 5, pp. 955–972, 2009. [7] F. Ye, G. Zhong, S. Lu, and L. Zhang, “Gradient broadcast: A robust data delivery protocol for large scale sensor networks,” Wireless Networks, vol. 11, no. 3, pp. 285–298, 2005. [8] P. Huang, H. Chen, G. Xing, and Y. Tan, “Sgf: A state-free gradientbased forwarding protocol for wireless sensor networks,” ACM Transactions on Sensor Networks, vol. 5, no. 2, 2009. [9] M. Penrose, “The longest edge of the random minimal spanning tree,” Annals of Applied Probability, vol. 7, no. 2, pp. 340–361, 1997. [10] P. Gupta and P. Kumar, “Critical power for asymptotic connectivity,” in Decision and Control, 1998. Proceedings of the 37th IEEE Conference on, vol. 1, 1998, pp. 1106–1110 vol.1. [11] M. Younis and K. Akkaya, “Strategies and techniques for node placement in wireless sensor networks: A survey,” Ad Hoc Networks, vol. 6, no. 4, pp. 621–655, 2008. [12] S. Olariu and I. Stojmenovic, “Design guidelines for maximizing lifetime and avoiding energy holes in sensor networks with uniform distribution and uniform reporting,” in INFOCOM 2006. 25th IEEE International Conference on Computer Communications. Proceedings, April 2006, pp. 1–12.