Distributed Algorithms for Energy-Efficient Cluster ... - Semantic Scholar
Conf. on Wireless Networks (ICWN'05)
405
Distributed Algorithms for Energy-Efficient Cluster-head Election in Wireless Mobile Sensor Networks Chuan-Ming Liu and Chuan-Hsiu Lee Department of Computer Science and Information Engineering National Taipei University of Technology Taipei, Taiwan E-mail: {cmliu, s2598004}@ntut.edu.tw
Abstract— The cluster-based architecture is an effective way to achieve the objective of energy efficiency in wireless sensor networks. One of the critical issues in wireless sensor networks is data-gathering. In this paper, we consider the cluster-based protocol for data-gathering and explore how to elect the clusterheads with node mobility. Two efficient distributed algorithms for cluster-head election in terms of energy consumption are provided. The proposed algorithms will make each round have the same number of cluster-heads (except the final rounds) and guarantee that each round has at least one cluster-head elected. Two mobility models, Random Walk Mobility model and Random Direction Mobility model, are considered in this paper for node mobility. Last, we implement the algorithms and perform the experiments for evaluation. The experiment results show that our cluster-head election algorithms both outperform the clusterhead election strategy used in LEACH and can make the system live longer.
keywords: Wireless Sensor Networks, Mobility, DataGathering, Energy-Efficiency, Self-Organization. I. I NTRODUCTION A wireless sensor network consists of a large number of tiny, low-power, cheap sensors having sensing, data processing, and wireless communication components. The applications of wireless sensor networks are widely ranged, including battlefield surveillance, machine failure diagnosis, biological detection, inventory tracking, home security, smart spaces, environmental monitoring, and so on [1], [2]. A sensor network not only has the ability to sense the interest but also has the network features. It hence represents an improvement over the traditional sensor systems. In this work, we consider the wireless mobile sensor networks where sensors are capable of moving. The sensors in a wireless sensor network are deployed randomly inside the region of interest or close to it. A remote base station (BS) connected to the Internet is engaged to give commands to all the sensors and gather information from the sensors. In addition to sensing, the wireless sensors can process the acquired information and transmit messages to the BS as well as communicate to each others. A wireless sensor network scenario is depicted in Figure 1.
Base Station (Sink)
Internet
User Sensor Field
Sensor Nodes
Fig. 1. The architecture of a wireless sensor network in which the sensors are deployed randomly into the interest area (sensor field) and the BS (sink) connects to the Internet.
Because the sensors are randomly scattered in a sensor field, the wireless sensor network protocols or algorithms should have the capability of self-organizing. This means that the wireless sensors are more autonomous than the traditional sensors. Hence, it is more challenging to design the routing protocols in such a distributed environment. Besides, wireless sensors have many limitations, including modest processing power, little storage, short communication range, and limited power source. These limitations also make designing wireless sensor network protocols difficult. Since wireless sensors are low-powered, the constraint on the power consumption is an important topic when designing wireless sensor network protocols. Data-gathering (collecting and routing the sensed information) raises an important topic in wireless sensor networks due to the limited power of a sensor. To have an efficient datagathering protocol in terms of energy consumption is an ongoing research work. Many protocols have been proposed for data gathering or communication between wireless sensors [3], [4], [5], [6], [7], [8], [9], [10]. To our knowledge, most of these protocols work on static wireless sensor networks.
Conf. on Wireless Networks (ICWN'05)
Base Station
Cluster-head
Cluster
Fig. 2. A snapshot of the cluster-based mechanism for wireless sensor networks where the non-cluster-head sensors (unfilled circles) send the sensed information to the cluster-heads (filled circles) and the cluster-heads fuse the information received and transmit the fused message to the BS.
Nevertheless, in many applications, the sensors can move either by outside force or its mobility component. For example, the sensors attached to moving objects for tracking or scattered on the sea. Existing protocols for data-gathering can be classified by the way to organize the sensors into: cluster-based and noncluster-based protocols. In this paper, we consider the clusterbased data-gathering protocol with node mobility. Such a datagathering protocol consists of a series of rounds. In each round, there are two major phases: (1) cluster organizing and (2) message transmission phases. In Phase (1), the entire sensor network will be partitioned into different clusters. Each cluster consists of one cluster-head and a number of sensors. There are further two steps in Phase (1): cluster-head election step and cluster formation step. After all the clusters are organized, the cluster-heads aggregate data from the sensors in their clusters and then transmits information to the BS directly in Phase (2). Figure 2 captures the operations in a round and presents a cluster-based mechanism. We observe that the node mobility impacts on the phase of cluster organizing and the way to select the cluster-heads also effects the performance in terms of energy consumption. Many mobility models have been proposed [11] in ad hoc networks. Random Walk Mobility model [12] and Random Direction Mobility model [13] are widely used. In Random Walk Mobility model, a mobile node can randomly choose a speed and direction (within a given range, respectively) to move from its current location to a new location. After moving a constant period of time or a constant distance, each mobile node will calculate a new speed and direction to modify its current position. In the Random Direction Mobility model, each mobile node has a constant speed and can randomly choose a direction to travel. A mobile node can choose a new angular direction (between 0 and 180 degrees), if the boundary is reached. For a mobile sensor network, it is possible to set the move pattern for a sensor in advance and the above two
406 mobility models therefore can be applied. In this paper, we will consider these two mobility models. In this paper, we discuss the impact of the cluster-heads election on the energy consumption and propose two distributed algorithms for cluster-head election in wireless mobile sensor networks. In order to measure the performance of the proposed distributed algorithm, we consider the lifetime of a system in terms of the number of rounds. The lifetime of a system is the duration in which the system works and relates to (1) the energy consumption and (2) the quality of service. We will show that a better cluster-head election algorithm can make the system live longer. If the system can live longer, one does not need to reset a new system (re-scatter the sensors) so often. The organization for the rest of this paper is as follows. Section II gives the motivation for exploring the cluster-head election problem. We then describe further about the problem model considered in this paper in Section III. Two distributed cluster-head election algorithms are presented and compared in Section IV. Section V shows the experiment results. We then give conclusion remarks in Section VI. II. M OTIVATION Although many cluster-based routing protocols have been proposed [14], [8], [15], the cluster-head election problem are realized and discussed in recent years [16], [17]. A better cluster-head election algorithm may average the workload on each sensors and result in a longer system life time. If the difference in the number of cluster-heads among the rounds is large, the energy consumption will not be even among sensors and some sensors may use up their own energy quickly; therefore, deteriorate the system lifetime. If there exist some rounds in which there is no cluster-head elected, each sensor will communicate directly with the BS and hence consume a large amount of energy. In this paper, we provide two distributed cluster-head election algorithms for mobile sensors. The proposed algorithms will achieve the following objectives: (1) there is at least one cluster-head elected in each round and (2) the number of cluster-heads generated in each round is always the same (except the final rounds). The rule for deciding the cluster-heads in LEACH [14] provides a distributed way to elect cluster-heads and can be fully applied to a distributed environment like a mobile sensor network. However, when applying such a rule, the difference in the number of the cluster-heads between the rounds may be large and it happens that there is no cluster-heads at all in some rounds. We will show this more in Section V. The cluster-head election strategy in LEACH uses the following threshold to be the probability of being a cluster-head for each sensor: ( P if v ∈ G; 1−P (r mod P1 ) (1) Tv = 0 otherwise, where r is the number of rounds that have passed, P the desired percentage of cluster-heads, and G the set of sensors not being a cluster-head yet. From (1), a sensor is elected to
Conf. on Wireless Networks (ICWN'05) be a cluster-head according to the accumulative times of not being a cluster-head. The above threshold provides a good mechanism for each sensor to determine to be a cluster-head or not independently. In each round, the threshold is independent from the actual number of cluster-heads elected in the previous rounds. However, it is not necessary to exactly have the desired percentage of cluster-heads in every round. Hence, the above threshold is an ideal condition. For example, consider there are 100 sensors and suppose in the first round there are only 3 cluster-heads elected. Then there are 97 sensors left in the second round. By applying (1), the expected number of cluster-heads is greater than 5. Besides, we can further derive the probability of no cluster-head elected in a round as P )|G| . (2) Pf (r) = (1 − 1 − P (r mod P1 ) By observing (2), we can conclude that, using such a threshold, there can be no cluster-head elected in some rounds and the number of cluster-heads elected in a round is related to the total number of cluster-heads elected in the previous rounds. In summary, the mechanism (1) used in LEACH for cluster-head election fails to achieve the two objectives discussed above. III. S YSTEM M ODEL In this paper, we consider the wireless mobile sensor networks where: • The BS is fixed and located far away from the sensors. • All the sensors are homogeneous and power limited. • Each sensor is equipped with a Location Finding System [18]. • All sensor modes are time-synchronized [2]. We use the radio model in [14]. The radio dissipates Eelec = 50nJ/bit to run the transmitter or receiver circuitry and ǫamp = 100pJ/bit/m2 for the transmitter amplifier. The radios have power control and can expend the minimum required energy to reach the intended recipients. We also assume an r2 energy loss due to channel transmission [19]. The transmission cost and receiving cost for a k-bit message and a distance d using this radio model is 2 • Transmission: ET x (k, d) = Eelec × k + ǫamp × k × d , and • Receiving: ERx (k, d) = Eelec × k. Receiving data is also a high cost operation; therefore, the number of receptions and transmissions should be minimized to reduce the energy cost of a application. Recall that there are two steps in the phase of organizing clusters: one step is to elect the cluster-heads and the following step is to form the clusters. In order to compare the cluster-head election algorithms, we use the mechanism, CM (Clustering with Mobility), provided in [6], [7] for mobile sensor networks to form the clusters after cluster-heads have been elected. The basic idea of this mechanism is to predict the location of a sensor when organizing the clusters. Such a mechanism fits the Random Walk Mobility and Random Direction Mobility models discussed in this paper.
407 IV. A LGORITHMS FOR C LUSTER -H EAD E LECTION In this section, we provide two distributed algorithms. The first algorithm determines the cluster-heads by counting and the other distributed algorithm determines the cluster-heads by location. Cluster-Head Election by Counting This subsection describes the distributed algorithm of clusterhead election by counting, Algorithm ACE-C. Suppose the number of sensors in a mobile sensor network is N and we number the sensors from 0 to N −1. Each sensor hence can use the assigned number as an unique identifier (ID) in the sensor network. Assume that we expect there are C clusters in each round. These C cluster-heads can be decided as follows. For each sensor v, we use vid as the ID of v. Each sensor also keeps the number of cluster-heads generated so far to control the total number of cluster-heads in each round. When the cluster-head election step starts, each sensor sets the number of cluster-heads to be 0 initially. Sensor v first increases vid by 1 and then considers the remainder vr of vid divided by N: • If vr is 0, sensor v is a cluster-head and then broadcasts an advertisement message to all the other sensors. In the mean time, sensor v increases the number of cluster-heads generated by 1. • If vr is not 0, v is not a cluster-head and should be ready for receiving the advertisement message. After receiving the advertisement message, sensor v increases the number of cluster-heads generated by 1. The process for electing cluster-heads stops when the number of cluster-heads generated is equal to C. Figure 3 and Figure 4 show the algorithm and an example respectively. In Figure 4, there are 9 sensors numbered from 0 to 8 as the identifiers for them. Suppose the number of cluster-heads in each cluster is C = 3. By following the algorithm, sensors 6, 7, and 8 are the clusters-heads in the first round after 3 loops. The next three loops will decide the next three cluster-heads in the next round. The process continues. Figure 4 also shows a case that a sensor dies (uses up all of its energy) during the sensor network lifetime. Suppose sensor 6 dies in the third round. According to the algorithm, sensor 6 should be a cluster-head in the 4th round. However, since it has died, no message will be sent at loop 12. All the other sensors wait a period of time and receive nothing. They will assume sensor 6 has gone and then resume the loop again. At this time, sensor 5 will be a cluster-head at loop 13. Therefore, the cluster-heads in the 4th round are nodes 5, 7, and 8 instead. In general, Algorithm ACE-C using counting can generate exactly C cluster-heads unless there are less than C sensors left if the system needs C cluster-heads in each round. Cluster-Head Election with Location In this subsection, we provide a distributed algorithm of
Conf. on Wireless Networks (ICWN'05)
408
Algorithm of Cluster-head Election by Counting Input: C, the number of cluster-heads in a round N , the total number of sensors (1) number of cluster-heads, CH=0 (2) while CH < C do (2.1) vid = (vid + 1) mod N ; /* vid : sensor id */ (2.2) if (vid = 0) then (2.2.1) v is a cluster-head; (2.2.2) broadcast an advertisement message; (2.2.3) increases CH by 1 endif (2.3) wait a period of time to receive the advertisement message; (2.4) if (message is received) then (2.4.1) increases CH by 1 endif endwhile End Fig. 3. Algorithm ACE-C used in each sensor v for cluster-head election in each round.
node (id) 0 loop 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 0 9 1 10 2 11 3 12 4 13
1
2
3
4
5
6
7
8
2 3 4 5 6 7 8 0 1 2 3 4 5
3 4 5 6 7 8 0 1 2 3 4 5 6
4 5 6 7 8 0 1 2 3 4 5 6 7
5 6 7 8 0 1 2 3 4 5 6 7 8
6 7 8 0 1 2 3 4 5 6 7 8 0
7 8 0 1 2 3 4 5 6
8 0 1 2 3 4 5 6 7 8 0 1 4
0 1 2 3 4 5 6 7 8 0 1 2 3
u u u u
round 1
round 2
round 3
round 4
Fig. 4. An example of 9 sensors for Algorithm ACE-C; shaded shapes indicating the elected cluster-heads.
cluster-head election with location, Algorithm ACE-L especially for mobile sensors. The key idea is to use the node mobility to have the sensors to be a cluster-head in turns. For a sensor, to be a cluster-head or not depends on its location. Besides, the cluster-head generation is a result of the channel contention among sensors. Suppose that we expect there are C clusters in a mobile sensor network. The algorithm works as follows. At the very beginning, we give C fixed reference points. These C reference points will effect the location of each cluster-head and the priority for a node to contend the radio channel. In each round, when the cluster-head election step starts, every sensor is willing to be a cluster-head and computes the distances to all the reference points as well as contends the channel to broadcast the beacon of being a cluster-head. The sensor which successfully obtains the channel will be a clusterhead. Consider an arbitrary sensor v. We refer to the reference point which is closest to v as the main reference point of v. To consider with the node mobility and the probability to be
v11 d1
v3
v6
d3 v7
rp1 v4
d2
v12
v15
rp3
d4
v2
v8
v1 v5
v9 rp2
v14
v10
rp4
v13
Fig. 5.
An example of cluster-head election by location.
a cluster-head of a sensor, we set up the delay time of node v according to the distance from v’s current location to the main reference point. The delay time is basically the priority to contend a channel for node v. According to the delay time, each sensor sends the beacon of being a cluster-head. If node v receives a beacon of begin a cluster-head from other sensor u during the delay time or the channel contention and node u has the same main reference point as v, sensor v stops the competition and will be not be a cluster-head in this round. Figure 5 shows an example of electing C = 4 clusterheads among 15 sensors using our algorithm at some arbitrary time instance. Suppose the 15 sensors are v1 , v2 , · · · , and v15 and the four reference points are rp1 , rp2 , rp3 , and rp4 , respectively. Consider sensor v3 . The distances from v3 to all the reference points rp1 , rp2 , rp3 , and rp4 are d1 , d2 , d3 , and d4 , respectively and d1 is the smallest distance. Reference point rp1 is the main reference point of sensor v3 and v3 has the smallest delay time among all the sensors which have rp1 as the main reference point. Therefore, sensor v3 is a cluster-head by referring to rp1 . Similarly, sensors v5 , v6 , and v10 are the cluster-heads by referring to rp2 , rp3 , and rp4 , respectively. The algorithm of cluster-head election by location, Algorithm ACE-L, is shown in Figure 6. In Step (1), sensor v calculates the distances to all the reference points and uses the reference point having the shortest distance as the main reference point. Step (2) then examines whether there is an elected cluster-head which uses the same main reference point as v. If there is one, the algorithm stops. Otherwise, the algorithm moves to Step (3). In Step (3), node v sets up the delay time according to some rule. In our case, we use the distance to the main reference point to set up the delay time. After setting the delay time, node v again examines whether there is an elected cluster-head which uses the same main reference point as Step (2). If the delay time passes, node v will transmit its beacon to be a cluster-head. The correctness comes from the property of transmission among sensors in a wireless environment. Under a fixed
Conf. on Wireless Networks (ICWN'05)
Fig. 6.
Algorithm ACE-L used in each sensor v for cluster-head election.
frequency, only one sensor can use that frequency to transmit data at an arbitrary time instance. Therefore, at each time slot, there is only one sensor can transmit the cluster-head beacon. Suppose there are C reference points, rp1 , · · · , rpC . By the property mentioned above, if a sensor uses reference point rpi as the main reference point for some i ≤ C and obtains the channel, all the other sensors use reference point rpi as the main reference point will not send any beacon and the cluster-head is thus elected. For the other reference points, the cluster-heads can be elected similarly. Since Algorithm ACEL using reference points, it is possible that Algorithm ACE-L generates less than C cluster-heads when all the sensors are close to some reference points and far away from the other reference points (i.e. the distribution of the position of sensor is skewed). Such a scenario may occur especially when there are only few sensors left in the system and will effect the performance of Algorithm ACE-L. We will see this more in Section V. Comparison between Cluster-head Election Algorithms We now compare Algorithm ACE-C and Algorithm ACEL. First of all, since Algorithm ACE-C only uses the ID to decide the cluster-heads without considering the location, the elected cluster-heads may be close to each other. As a result, the cluster sizes (i.e. the number of sensors in a cluster) are different dramatically among all the generated clusters. It turns out that some sensors may consume a large amount of energy and the system lifetime becomes short. Algorithm ACE-L uses the locations to elect the cluster-heads; hence, can avoid such a situation. Consider that there are 100 sensors in a sensor network and the expected number of clusters in each round is 5. The expected cluster size thus is 20. We apply Algorithm ACE-C and ACE-L to generate the cluster-heads respectively and both algorithms use CM to form the clusters. Figure 7 shows the distribution of the standard deviation of the cluster size generated by Algorithm ACE-C and ACE-L
18 : ACE−C : ACE−L
16
14
12
percentage
Algorithm of Cluster-head Election by Location Input: C reference points (rps) (1) compute the distances to all the rps and set distance to be the distance to the main rp; (2) if receive a cluster-head beacon (chb) from other sensor u and sensors u and v have the same main rp then (2.1) cluster-head ← true; (2.2) exit (3) else (3.1) delay time← distance divided by 10 (3.2) while(delay time decreases one) do (3.2.1) if receive a chb from other sensor u and sensors u and v have the same main rp then (3.2.1.1) cluster-head ← true; (3.2.1.2) exit endif endwhile (3.3) transmit its cluster-head beacon endif End
409
10
8
6
4
2
0
0
5
10 15 standard deviation of cluster size
20
25
Fig. 7. The distribution of the standard deviation of the cluster size generated by Algorithms ACE-C and ACE-L for 200,000 rounds.
for 200,000 rounds. The results indicate that the cluster size produced by Algorithm ACE-L is more even than the one generated by Algorithm ACE-C. Indeed, in our experiments, Algorithm ACE-L generates more percentage of clusters of size between 18 to 22 than Algorithm ACE-C does. Therefore, Algorithm ACE-L leads to a better performance in terms of energy consumption and the experiment results demonstrate this trend. Secondly, since Algorithm ACE-C does not consider the location, it fits for both mobile and static sensor networks. In contrast with Algorithm ACE-C, Algorithm ACE-L is designed especially for a mobile environment. When applying Algorithm ACE-L to a static environment, some sensors will use up all the energy quickly because the static reference points will make a sensor being a cluster-head in a consecutive rounds. Hence, Algorithm ACE-L is not fit for a static sensor network. To adopt Algorithm ACE-L to a static environment, one can use dynamic reference points instead. However, every sensor needs to know the moving pattern of the reference points. Last, because using the reference points, the mobility models will effect Algorithm ACE-L. In this paper, we consider that the sensors are randomly scattered and each sensor moves randomly under the two mobility models considered. The impact of the mobility on the cluster-head election using Algorithm ACE-L becomes less. For a mobility model where the sensors are distributed non-uniformly, using Algorithm ACE-L to elect cluster-heads may not be a good selection. On the other hand, since Algorithm ACE-C does not consider the location, the mobility models will not effect Algorithm ACE-C at all. V. E XPERIMENTS In this section, we present the simulation for the two algorithms provided in this paper and compare both of them
Conf. on Wireless Networks (ICWN'05) System Lifetime(200m x 200m;1J,T=0.875), Random Direction Mobiltiy Model 100 : CM : ACE−C : ACE−L(5 rp) : ACE−L(4 rp)
90
80 accumulated number of dead nodes
with LEACH. All the cluster-head election algorithms use CM mechanism in [6], to form the clusters. The experimental result shows that these two algorithms, ACE-C and ACEL, make the whole system live longer than LEACH does. We first discuss the cluster size generated by the clusterhead election algorithms. Then we show the system lifetimes for different cluster-head election algorithms and make a comparison between these algorithms. In the experiments, we use two mobility models: Random Walk Mobility model and Random Direction Mobility model. The sensor network area is 200m × 200m. There are 100 mobile sensors in the area. The speed of each sensor is from 0 to 1 m/sec and the initial energy in a sensor is 1.0 J. We have each sensor send a 2000-bits data packet to the base station in each round. The period of one round is 5 seconds. Initially, all the sensors are assumed to be scattered randomly in the area. For the cluster-head election, the initial probability P of a sensor to be a cluster-head is 0.05 when we consider LEACH protocol. The number of cluster-heads in each round is therefore 100 × 0.05 = 5.
410
70
60
50
40
30
20
10
0
300
400
500
600
700 round
800
900
1000
1100
1200
Fig. 8. The system lifetime of the wireless mobile sensor network which has 100 sensors in the range of 200m× 200m and uses the Random Direction Mobility model; each sensor having initial energy 1.0J and random speed between 0.0 and 1.0 m/sec. System Lifetime(200m x 200m;1J,T=0.875), Random Walk Mobiltiy Model 100
System Lifetime We now consider the energy efficiency by measuring the system lifetime in terms of the total number of rounds which a wireless mobile sensor network system experiences. We first consider the round at which the first sensor dies. Recall that, if the number of cluster-heads in each round is even, the energy consumption is even among all the sensors and hence the system lifetime is longer. In particular, even number of cluster-heads in each round makes the first sensor died in a latter round. Table V shows this trends since our algorithms can make even number of cluster-heads among the rounds. In Table V, the first dead node comes out earlier when using Algorithm ACE-C compared with Algorithm ACE-L since Algorithm ACE-C may generate the cluster-heads close to each other.
90
: CM : ACE−C : ACE−L(5 rp) : ACE−L(4 rp)
80 accumulated number of dead nodes
Cluster Size As mentioned in Section II, by using the cluster-head election strategy of LEACH protocol, the difference in the number of cluster-heads between the rounds may be large and it happens that there is no cluster-heads at all in some rounds. The experiment results show that the range of the number of cluster-heads in a round can be from 0 to 10 when applying the rule in LEACH and the percentage of the rounds in which there is no cluster-heads elected is about 12% to 15%. Such a cluster-head election will deteriorate the performance in terms of the energy consumption. Table V shows the distribution of the number of cluster-heads generated in each round. The results also show that if one expects C cluster-heads in each round, our distributed algorithms will elect C cluster-heads exactly unless there are fewer than C sensors left in the system. Recall that algorithm ACE-L may generate less than C clusterheads when there are only few sensors left in the system. Therefore Algorithm ACE-C will generate more rounds having C cluster-heads than Algorithm ACE-L does.
70
60
50
40
30
20
10
0
300
400
500
600
700 round
800
900
1000
1100
1200
Fig. 9. The system lifetime of the wireless mobile sensor network which has 100 sensors in the range of 200m× 200m and uses the Random Walk Mobility model; each sensor has initial energy 1.0J and random speed between 0.0 and 1.0 m/sec.
Figure 8 and 9 show the system lifetimes when each sensor has initial energy 1.0 J in areas of 200m× 200m on Random Direction Mobility model and Random Walk Mobility model, respectively. From the plots, both of our algorithms make the system live longer than LEACH does; therefore, result in a better performance than LEACH in terms of energy consumption. This indicates that the two mechanisms we provide for electing cluster-heads indeed improve the performance because these two mechanisms can avoid the case that all the sensors send their data directly to the BS. In general, Algorithm ACE-L is better than Algorithm ACE-C since the position of cluster-heads in each round is more distributed in interested area. In some rounds, there could be five clusterheads in dense region for Algorithm ACE-C. In such a case, the cluster-heads will consume the energy dramatically as
Conf. on Wireless Networks (ICWN'05)
411
TABLE I T HE PERCENTAGE OF DIFFERENT NUMBER OF CLUSTER - HEADS ELECTED IN A
# of cluster-heads LEACH with CM ACE-C ACE-L
0 12% 0% 0%
1 3% 0% 0.6%
2 6% 0.2% 0.4%
3 15% 0% 0.4%
4 14% 0.1% 0.7%
5 15% 99.7% 97.9%
ROUND DURING A SYSTEM LIFETIME .
6 12% 0% 0%
7 10% 0% 0%
8 9% 0% 0%
over 9 4% 0% 0%
TABLE II T HE ROUND OF FIRST DEAD NODE FOR EACH ALGORITHMS . LEACH with CM ACE-C ACE-L
Random Walk Mobility Model 355 546 605
mentioned in Section IV. Furthermore, we also set the number of reference points to be 4 which are located at the centers of the four quadrants of the rectangle sensing region. For 5 reference points, we add one reference at the center of the sensing region. In our experiments, the result shows that using 4 reference points outperforms using 5 reference points in terms of the energy consumption. Such a result relates to the distribution of the reference nodes. Besides, the performance under the Random Direction Mobility model is better than the one under the Random Walk Mobility model since the CM mechanism predicts the sensor location more precisely in the Random Direction Mobility model. VI. C ONCLUSIONS In this paper, we first discuss the rule used in LEACH for cluster-head election and then provide two distributed algorithms which can elect C cluster-heads exactly if one expects C cluster-heads. The algorithm using counting can be applied to both static and mobile environments but may result in the case that all the cluster-heads are close to each other. Algorithm ACE-L using reference points is specially for the mobile sensors and performs better than Algorithm ACE-C. Combining our algorithms for cluster-head election and the CM mechanism for organizing the clusters, we can get two data-gathering protocols which lead to a longer system lifetime compared with LEACH; therefore, are more efficient in terms of energy consumption. Two mobility models are considered in the experiments: Random Walk Mobility model and Random Direction Mobility model. We are now currently working on analyzing what the best number and distribution of reference points are when employing Algorithm ACE-L to elect the cluster-heads. Besides, we also consider to use other mobility models in the experiment in order to analyze the impact of the mobility model on the cluster-head election. R EFERENCES [1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “Wireless sensor networks: a survey,” Computer Networks, vol. 38, pp. 393–422, 2002. [2] A. Cerpa, J.Elson, D. Estrin, L. Girad, M. Hamilton, and J. Zhao, “Habitat monitoring: Application driver for wireless communication technology,” in Proceedings of ACM SIGCOMM Workshop on Data Communications in Latin America and the Caribbean, 2001, pp. 3–5.
Random Direction Mobility Model 451 670 767
[3] Z. Cheng, M. Perillo, B. Tavli, W. Heinzelman, S. Tilak, and N. AbuGhazaleh, “Protocols for local data delivery in wireless microsensor networks,” in In Proceedings of the 45th IEEE Midwest Symposium on Circuits and Systems (MWSCAS ’02), 2002, pp. 623–626. [4] E. J. Duarte-Melo and M. Liu, “Data-gathering wireless sensor networks: organization and capacity,” Computer Networks, vol. 43, pp. 519–537, 2003. [5] K. Kalpakis, K. Dasgupta, and P. Namjoshi, “Maximum lifetime data gathering and aggregation in wireless sensor networks,” in Proceedings of the 2002 IEEE International Conference on Networking (ICN’02), 2002, pp. 685–696. [6] C.-M. Liu and C.-H. Lee, “Power efficient communication protocols for data gathering on mobile sensor networks,” in Proceedings of the 60th IEEE Vehicular Technology Conference, Fall, 2004. [7] C.-M. Liu, C.-H. Lee, and L.-C. Wang, “Power-efficient communication algorithms for wireless mobile sensor networks,” in Proceedings of the 1st ACM international workshop on Performance evaluation of wireless ad hoc, sensor, and ubiquitous networks, 2004, pp. 121–122. [8] S. Lindesy, C. Raghavendra, and K. M. Sivalingam, “Data gathering algorithms in sensor networks using energy metrics,” IEEE Transactions on Parallel and Distributed Systems, vol. 13, no. 9, pp. 924–935, 2002. [9] D. Petrovic, R. C. Shah, K. Ramchandran, and J. Rabaey, “Data funneling: Routing with aggregation and compression for wireless sensor networks,” in Proceedings of the First IEEE International Workshop on Sensor Network Protocols and Application, 2003, pp. 156–162. [10] M. Younis, M. Youssef, and K. Arisha, “Energy-aware management for cluster-based sensor networks,” Computer Networks, vol. 43, no. 5, pp. 649–668, 2003. [11] T. Camp, J. Boleng, and V. Davies, “A survey of mobility models for ad hoc network research,” Wireless Communication and Mobile Computing, vol. 2, no. 5, pp. 483–502, 2002. [12] D. Davies, “Evaluating mobility models within an ad hoc network,” Master’s thesis, Colorado School of Mines, 2000. [13] E. Royer, P. Melliar-Smith, and L. Moser, “An analysis of the optimum node density for ad hoc mobile networks,” in Proceedings of the IEEE International Conference on Communications (ICC01), 2001. [14] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energyefficient communication protocol for wireless microsensor networks,” in Proceedings of the 33rd Hawaii International Conference on System Sciences, 2000, pp. 1–10. [15] D. Tian and N. D. Georganas, “A node scheduling scheme for energy conservation in large wireless sensor networks,” Wireless Communications and Mobile Computing, vol. 3, pp. 271–290, 2003. [16] O. Younis and S. Fahmy, “Heed: A hybrid, energy-efficient, distributed clustering approach for ad-hoc sensor networks,” IEEE Transactions on Mobile Computing, vol. 3, no. 4, pp. 366–379, 2004. [17] L. Zhao, X. Hong, and Q. Liang, “Energy-efficient self-organization for wireless sensor networks: A fully distributed approach,” in Proceedings of IEEE Global Telecommunications Conference, 2004, pp. 2728 – 2732. [18] N. Bulusu, J. Heidemann, and D. Estrin, “Gps-less low cost outdoor localization for very small devices,” IEEE Personal Communication Magazine, vol. 7, no. 5, pp. 28–34, 2000. [19] T. S. Rappaport, Wireless Communications. Prentice-Hall, 1996.
405
Distributed Algorithms for Energy-Efficient Cluster-head Election in Wireless Mobile Sensor Networks Chuan-Ming Liu and Chuan-Hsiu Lee Department of Computer Science and Information Engineering National Taipei University of Technology Taipei, Taiwan E-mail: {cmliu, s2598004}@ntut.edu.tw
Abstract— The cluster-based architecture is an effective way to achieve the objective of energy efficiency in wireless sensor networks. One of the critical issues in wireless sensor networks is data-gathering. In this paper, we consider the cluster-based protocol for data-gathering and explore how to elect the clusterheads with node mobility. Two efficient distributed algorithms for cluster-head election in terms of energy consumption are provided. The proposed algorithms will make each round have the same number of cluster-heads (except the final rounds) and guarantee that each round has at least one cluster-head elected. Two mobility models, Random Walk Mobility model and Random Direction Mobility model, are considered in this paper for node mobility. Last, we implement the algorithms and perform the experiments for evaluation. The experiment results show that our cluster-head election algorithms both outperform the clusterhead election strategy used in LEACH and can make the system live longer.
keywords: Wireless Sensor Networks, Mobility, DataGathering, Energy-Efficiency, Self-Organization. I. I NTRODUCTION A wireless sensor network consists of a large number of tiny, low-power, cheap sensors having sensing, data processing, and wireless communication components. The applications of wireless sensor networks are widely ranged, including battlefield surveillance, machine failure diagnosis, biological detection, inventory tracking, home security, smart spaces, environmental monitoring, and so on [1], [2]. A sensor network not only has the ability to sense the interest but also has the network features. It hence represents an improvement over the traditional sensor systems. In this work, we consider the wireless mobile sensor networks where sensors are capable of moving. The sensors in a wireless sensor network are deployed randomly inside the region of interest or close to it. A remote base station (BS) connected to the Internet is engaged to give commands to all the sensors and gather information from the sensors. In addition to sensing, the wireless sensors can process the acquired information and transmit messages to the BS as well as communicate to each others. A wireless sensor network scenario is depicted in Figure 1.
Base Station (Sink)
Internet
User Sensor Field
Sensor Nodes
Fig. 1. The architecture of a wireless sensor network in which the sensors are deployed randomly into the interest area (sensor field) and the BS (sink) connects to the Internet.
Because the sensors are randomly scattered in a sensor field, the wireless sensor network protocols or algorithms should have the capability of self-organizing. This means that the wireless sensors are more autonomous than the traditional sensors. Hence, it is more challenging to design the routing protocols in such a distributed environment. Besides, wireless sensors have many limitations, including modest processing power, little storage, short communication range, and limited power source. These limitations also make designing wireless sensor network protocols difficult. Since wireless sensors are low-powered, the constraint on the power consumption is an important topic when designing wireless sensor network protocols. Data-gathering (collecting and routing the sensed information) raises an important topic in wireless sensor networks due to the limited power of a sensor. To have an efficient datagathering protocol in terms of energy consumption is an ongoing research work. Many protocols have been proposed for data gathering or communication between wireless sensors [3], [4], [5], [6], [7], [8], [9], [10]. To our knowledge, most of these protocols work on static wireless sensor networks.
Conf. on Wireless Networks (ICWN'05)
Base Station
Cluster-head
Cluster
Fig. 2. A snapshot of the cluster-based mechanism for wireless sensor networks where the non-cluster-head sensors (unfilled circles) send the sensed information to the cluster-heads (filled circles) and the cluster-heads fuse the information received and transmit the fused message to the BS.
Nevertheless, in many applications, the sensors can move either by outside force or its mobility component. For example, the sensors attached to moving objects for tracking or scattered on the sea. Existing protocols for data-gathering can be classified by the way to organize the sensors into: cluster-based and noncluster-based protocols. In this paper, we consider the clusterbased data-gathering protocol with node mobility. Such a datagathering protocol consists of a series of rounds. In each round, there are two major phases: (1) cluster organizing and (2) message transmission phases. In Phase (1), the entire sensor network will be partitioned into different clusters. Each cluster consists of one cluster-head and a number of sensors. There are further two steps in Phase (1): cluster-head election step and cluster formation step. After all the clusters are organized, the cluster-heads aggregate data from the sensors in their clusters and then transmits information to the BS directly in Phase (2). Figure 2 captures the operations in a round and presents a cluster-based mechanism. We observe that the node mobility impacts on the phase of cluster organizing and the way to select the cluster-heads also effects the performance in terms of energy consumption. Many mobility models have been proposed [11] in ad hoc networks. Random Walk Mobility model [12] and Random Direction Mobility model [13] are widely used. In Random Walk Mobility model, a mobile node can randomly choose a speed and direction (within a given range, respectively) to move from its current location to a new location. After moving a constant period of time or a constant distance, each mobile node will calculate a new speed and direction to modify its current position. In the Random Direction Mobility model, each mobile node has a constant speed and can randomly choose a direction to travel. A mobile node can choose a new angular direction (between 0 and 180 degrees), if the boundary is reached. For a mobile sensor network, it is possible to set the move pattern for a sensor in advance and the above two
406 mobility models therefore can be applied. In this paper, we will consider these two mobility models. In this paper, we discuss the impact of the cluster-heads election on the energy consumption and propose two distributed algorithms for cluster-head election in wireless mobile sensor networks. In order to measure the performance of the proposed distributed algorithm, we consider the lifetime of a system in terms of the number of rounds. The lifetime of a system is the duration in which the system works and relates to (1) the energy consumption and (2) the quality of service. We will show that a better cluster-head election algorithm can make the system live longer. If the system can live longer, one does not need to reset a new system (re-scatter the sensors) so often. The organization for the rest of this paper is as follows. Section II gives the motivation for exploring the cluster-head election problem. We then describe further about the problem model considered in this paper in Section III. Two distributed cluster-head election algorithms are presented and compared in Section IV. Section V shows the experiment results. We then give conclusion remarks in Section VI. II. M OTIVATION Although many cluster-based routing protocols have been proposed [14], [8], [15], the cluster-head election problem are realized and discussed in recent years [16], [17]. A better cluster-head election algorithm may average the workload on each sensors and result in a longer system life time. If the difference in the number of cluster-heads among the rounds is large, the energy consumption will not be even among sensors and some sensors may use up their own energy quickly; therefore, deteriorate the system lifetime. If there exist some rounds in which there is no cluster-head elected, each sensor will communicate directly with the BS and hence consume a large amount of energy. In this paper, we provide two distributed cluster-head election algorithms for mobile sensors. The proposed algorithms will achieve the following objectives: (1) there is at least one cluster-head elected in each round and (2) the number of cluster-heads generated in each round is always the same (except the final rounds). The rule for deciding the cluster-heads in LEACH [14] provides a distributed way to elect cluster-heads and can be fully applied to a distributed environment like a mobile sensor network. However, when applying such a rule, the difference in the number of the cluster-heads between the rounds may be large and it happens that there is no cluster-heads at all in some rounds. We will show this more in Section V. The cluster-head election strategy in LEACH uses the following threshold to be the probability of being a cluster-head for each sensor: ( P if v ∈ G; 1−P (r mod P1 ) (1) Tv = 0 otherwise, where r is the number of rounds that have passed, P the desired percentage of cluster-heads, and G the set of sensors not being a cluster-head yet. From (1), a sensor is elected to
Conf. on Wireless Networks (ICWN'05) be a cluster-head according to the accumulative times of not being a cluster-head. The above threshold provides a good mechanism for each sensor to determine to be a cluster-head or not independently. In each round, the threshold is independent from the actual number of cluster-heads elected in the previous rounds. However, it is not necessary to exactly have the desired percentage of cluster-heads in every round. Hence, the above threshold is an ideal condition. For example, consider there are 100 sensors and suppose in the first round there are only 3 cluster-heads elected. Then there are 97 sensors left in the second round. By applying (1), the expected number of cluster-heads is greater than 5. Besides, we can further derive the probability of no cluster-head elected in a round as P )|G| . (2) Pf (r) = (1 − 1 − P (r mod P1 ) By observing (2), we can conclude that, using such a threshold, there can be no cluster-head elected in some rounds and the number of cluster-heads elected in a round is related to the total number of cluster-heads elected in the previous rounds. In summary, the mechanism (1) used in LEACH for cluster-head election fails to achieve the two objectives discussed above. III. S YSTEM M ODEL In this paper, we consider the wireless mobile sensor networks where: • The BS is fixed and located far away from the sensors. • All the sensors are homogeneous and power limited. • Each sensor is equipped with a Location Finding System [18]. • All sensor modes are time-synchronized [2]. We use the radio model in [14]. The radio dissipates Eelec = 50nJ/bit to run the transmitter or receiver circuitry and ǫamp = 100pJ/bit/m2 for the transmitter amplifier. The radios have power control and can expend the minimum required energy to reach the intended recipients. We also assume an r2 energy loss due to channel transmission [19]. The transmission cost and receiving cost for a k-bit message and a distance d using this radio model is 2 • Transmission: ET x (k, d) = Eelec × k + ǫamp × k × d , and • Receiving: ERx (k, d) = Eelec × k. Receiving data is also a high cost operation; therefore, the number of receptions and transmissions should be minimized to reduce the energy cost of a application. Recall that there are two steps in the phase of organizing clusters: one step is to elect the cluster-heads and the following step is to form the clusters. In order to compare the cluster-head election algorithms, we use the mechanism, CM (Clustering with Mobility), provided in [6], [7] for mobile sensor networks to form the clusters after cluster-heads have been elected. The basic idea of this mechanism is to predict the location of a sensor when organizing the clusters. Such a mechanism fits the Random Walk Mobility and Random Direction Mobility models discussed in this paper.
407 IV. A LGORITHMS FOR C LUSTER -H EAD E LECTION In this section, we provide two distributed algorithms. The first algorithm determines the cluster-heads by counting and the other distributed algorithm determines the cluster-heads by location. Cluster-Head Election by Counting This subsection describes the distributed algorithm of clusterhead election by counting, Algorithm ACE-C. Suppose the number of sensors in a mobile sensor network is N and we number the sensors from 0 to N −1. Each sensor hence can use the assigned number as an unique identifier (ID) in the sensor network. Assume that we expect there are C clusters in each round. These C cluster-heads can be decided as follows. For each sensor v, we use vid as the ID of v. Each sensor also keeps the number of cluster-heads generated so far to control the total number of cluster-heads in each round. When the cluster-head election step starts, each sensor sets the number of cluster-heads to be 0 initially. Sensor v first increases vid by 1 and then considers the remainder vr of vid divided by N: • If vr is 0, sensor v is a cluster-head and then broadcasts an advertisement message to all the other sensors. In the mean time, sensor v increases the number of cluster-heads generated by 1. • If vr is not 0, v is not a cluster-head and should be ready for receiving the advertisement message. After receiving the advertisement message, sensor v increases the number of cluster-heads generated by 1. The process for electing cluster-heads stops when the number of cluster-heads generated is equal to C. Figure 3 and Figure 4 show the algorithm and an example respectively. In Figure 4, there are 9 sensors numbered from 0 to 8 as the identifiers for them. Suppose the number of cluster-heads in each cluster is C = 3. By following the algorithm, sensors 6, 7, and 8 are the clusters-heads in the first round after 3 loops. The next three loops will decide the next three cluster-heads in the next round. The process continues. Figure 4 also shows a case that a sensor dies (uses up all of its energy) during the sensor network lifetime. Suppose sensor 6 dies in the third round. According to the algorithm, sensor 6 should be a cluster-head in the 4th round. However, since it has died, no message will be sent at loop 12. All the other sensors wait a period of time and receive nothing. They will assume sensor 6 has gone and then resume the loop again. At this time, sensor 5 will be a cluster-head at loop 13. Therefore, the cluster-heads in the 4th round are nodes 5, 7, and 8 instead. In general, Algorithm ACE-C using counting can generate exactly C cluster-heads unless there are less than C sensors left if the system needs C cluster-heads in each round. Cluster-Head Election with Location In this subsection, we provide a distributed algorithm of
Conf. on Wireless Networks (ICWN'05)
408
Algorithm of Cluster-head Election by Counting Input: C, the number of cluster-heads in a round N , the total number of sensors (1) number of cluster-heads, CH=0 (2) while CH < C do (2.1) vid = (vid + 1) mod N ; /* vid : sensor id */ (2.2) if (vid = 0) then (2.2.1) v is a cluster-head; (2.2.2) broadcast an advertisement message; (2.2.3) increases CH by 1 endif (2.3) wait a period of time to receive the advertisement message; (2.4) if (message is received) then (2.4.1) increases CH by 1 endif endwhile End Fig. 3. Algorithm ACE-C used in each sensor v for cluster-head election in each round.
node (id) 0 loop 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 0 9 1 10 2 11 3 12 4 13
1
2
3
4
5
6
7
8
2 3 4 5 6 7 8 0 1 2 3 4 5
3 4 5 6 7 8 0 1 2 3 4 5 6
4 5 6 7 8 0 1 2 3 4 5 6 7
5 6 7 8 0 1 2 3 4 5 6 7 8
6 7 8 0 1 2 3 4 5 6 7 8 0
7 8 0 1 2 3 4 5 6
8 0 1 2 3 4 5 6 7 8 0 1 4
0 1 2 3 4 5 6 7 8 0 1 2 3
u u u u
round 1
round 2
round 3
round 4
Fig. 4. An example of 9 sensors for Algorithm ACE-C; shaded shapes indicating the elected cluster-heads.
cluster-head election with location, Algorithm ACE-L especially for mobile sensors. The key idea is to use the node mobility to have the sensors to be a cluster-head in turns. For a sensor, to be a cluster-head or not depends on its location. Besides, the cluster-head generation is a result of the channel contention among sensors. Suppose that we expect there are C clusters in a mobile sensor network. The algorithm works as follows. At the very beginning, we give C fixed reference points. These C reference points will effect the location of each cluster-head and the priority for a node to contend the radio channel. In each round, when the cluster-head election step starts, every sensor is willing to be a cluster-head and computes the distances to all the reference points as well as contends the channel to broadcast the beacon of being a cluster-head. The sensor which successfully obtains the channel will be a clusterhead. Consider an arbitrary sensor v. We refer to the reference point which is closest to v as the main reference point of v. To consider with the node mobility and the probability to be
v11 d1
v3
v6
d3 v7
rp1 v4
d2
v12
v15
rp3
d4
v2
v8
v1 v5
v9 rp2
v14
v10
rp4
v13
Fig. 5.
An example of cluster-head election by location.
a cluster-head of a sensor, we set up the delay time of node v according to the distance from v’s current location to the main reference point. The delay time is basically the priority to contend a channel for node v. According to the delay time, each sensor sends the beacon of being a cluster-head. If node v receives a beacon of begin a cluster-head from other sensor u during the delay time or the channel contention and node u has the same main reference point as v, sensor v stops the competition and will be not be a cluster-head in this round. Figure 5 shows an example of electing C = 4 clusterheads among 15 sensors using our algorithm at some arbitrary time instance. Suppose the 15 sensors are v1 , v2 , · · · , and v15 and the four reference points are rp1 , rp2 , rp3 , and rp4 , respectively. Consider sensor v3 . The distances from v3 to all the reference points rp1 , rp2 , rp3 , and rp4 are d1 , d2 , d3 , and d4 , respectively and d1 is the smallest distance. Reference point rp1 is the main reference point of sensor v3 and v3 has the smallest delay time among all the sensors which have rp1 as the main reference point. Therefore, sensor v3 is a cluster-head by referring to rp1 . Similarly, sensors v5 , v6 , and v10 are the cluster-heads by referring to rp2 , rp3 , and rp4 , respectively. The algorithm of cluster-head election by location, Algorithm ACE-L, is shown in Figure 6. In Step (1), sensor v calculates the distances to all the reference points and uses the reference point having the shortest distance as the main reference point. Step (2) then examines whether there is an elected cluster-head which uses the same main reference point as v. If there is one, the algorithm stops. Otherwise, the algorithm moves to Step (3). In Step (3), node v sets up the delay time according to some rule. In our case, we use the distance to the main reference point to set up the delay time. After setting the delay time, node v again examines whether there is an elected cluster-head which uses the same main reference point as Step (2). If the delay time passes, node v will transmit its beacon to be a cluster-head. The correctness comes from the property of transmission among sensors in a wireless environment. Under a fixed
Conf. on Wireless Networks (ICWN'05)
Fig. 6.
Algorithm ACE-L used in each sensor v for cluster-head election.
frequency, only one sensor can use that frequency to transmit data at an arbitrary time instance. Therefore, at each time slot, there is only one sensor can transmit the cluster-head beacon. Suppose there are C reference points, rp1 , · · · , rpC . By the property mentioned above, if a sensor uses reference point rpi as the main reference point for some i ≤ C and obtains the channel, all the other sensors use reference point rpi as the main reference point will not send any beacon and the cluster-head is thus elected. For the other reference points, the cluster-heads can be elected similarly. Since Algorithm ACEL using reference points, it is possible that Algorithm ACE-L generates less than C cluster-heads when all the sensors are close to some reference points and far away from the other reference points (i.e. the distribution of the position of sensor is skewed). Such a scenario may occur especially when there are only few sensors left in the system and will effect the performance of Algorithm ACE-L. We will see this more in Section V. Comparison between Cluster-head Election Algorithms We now compare Algorithm ACE-C and Algorithm ACEL. First of all, since Algorithm ACE-C only uses the ID to decide the cluster-heads without considering the location, the elected cluster-heads may be close to each other. As a result, the cluster sizes (i.e. the number of sensors in a cluster) are different dramatically among all the generated clusters. It turns out that some sensors may consume a large amount of energy and the system lifetime becomes short. Algorithm ACE-L uses the locations to elect the cluster-heads; hence, can avoid such a situation. Consider that there are 100 sensors in a sensor network and the expected number of clusters in each round is 5. The expected cluster size thus is 20. We apply Algorithm ACE-C and ACE-L to generate the cluster-heads respectively and both algorithms use CM to form the clusters. Figure 7 shows the distribution of the standard deviation of the cluster size generated by Algorithm ACE-C and ACE-L
18 : ACE−C : ACE−L
16
14
12
percentage
Algorithm of Cluster-head Election by Location Input: C reference points (rps) (1) compute the distances to all the rps and set distance to be the distance to the main rp; (2) if receive a cluster-head beacon (chb) from other sensor u and sensors u and v have the same main rp then (2.1) cluster-head ← true; (2.2) exit (3) else (3.1) delay time← distance divided by 10 (3.2) while(delay time decreases one) do (3.2.1) if receive a chb from other sensor u and sensors u and v have the same main rp then (3.2.1.1) cluster-head ← true; (3.2.1.2) exit endif endwhile (3.3) transmit its cluster-head beacon endif End
409
10
8
6
4
2
0
0
5
10 15 standard deviation of cluster size
20
25
Fig. 7. The distribution of the standard deviation of the cluster size generated by Algorithms ACE-C and ACE-L for 200,000 rounds.
for 200,000 rounds. The results indicate that the cluster size produced by Algorithm ACE-L is more even than the one generated by Algorithm ACE-C. Indeed, in our experiments, Algorithm ACE-L generates more percentage of clusters of size between 18 to 22 than Algorithm ACE-C does. Therefore, Algorithm ACE-L leads to a better performance in terms of energy consumption and the experiment results demonstrate this trend. Secondly, since Algorithm ACE-C does not consider the location, it fits for both mobile and static sensor networks. In contrast with Algorithm ACE-C, Algorithm ACE-L is designed especially for a mobile environment. When applying Algorithm ACE-L to a static environment, some sensors will use up all the energy quickly because the static reference points will make a sensor being a cluster-head in a consecutive rounds. Hence, Algorithm ACE-L is not fit for a static sensor network. To adopt Algorithm ACE-L to a static environment, one can use dynamic reference points instead. However, every sensor needs to know the moving pattern of the reference points. Last, because using the reference points, the mobility models will effect Algorithm ACE-L. In this paper, we consider that the sensors are randomly scattered and each sensor moves randomly under the two mobility models considered. The impact of the mobility on the cluster-head election using Algorithm ACE-L becomes less. For a mobility model where the sensors are distributed non-uniformly, using Algorithm ACE-L to elect cluster-heads may not be a good selection. On the other hand, since Algorithm ACE-C does not consider the location, the mobility models will not effect Algorithm ACE-C at all. V. E XPERIMENTS In this section, we present the simulation for the two algorithms provided in this paper and compare both of them
Conf. on Wireless Networks (ICWN'05) System Lifetime(200m x 200m;1J,T=0.875), Random Direction Mobiltiy Model 100 : CM : ACE−C : ACE−L(5 rp) : ACE−L(4 rp)
90
80 accumulated number of dead nodes
with LEACH. All the cluster-head election algorithms use CM mechanism in [6], to form the clusters. The experimental result shows that these two algorithms, ACE-C and ACEL, make the whole system live longer than LEACH does. We first discuss the cluster size generated by the clusterhead election algorithms. Then we show the system lifetimes for different cluster-head election algorithms and make a comparison between these algorithms. In the experiments, we use two mobility models: Random Walk Mobility model and Random Direction Mobility model. The sensor network area is 200m × 200m. There are 100 mobile sensors in the area. The speed of each sensor is from 0 to 1 m/sec and the initial energy in a sensor is 1.0 J. We have each sensor send a 2000-bits data packet to the base station in each round. The period of one round is 5 seconds. Initially, all the sensors are assumed to be scattered randomly in the area. For the cluster-head election, the initial probability P of a sensor to be a cluster-head is 0.05 when we consider LEACH protocol. The number of cluster-heads in each round is therefore 100 × 0.05 = 5.
410
70
60
50
40
30
20
10
0
300
400
500
600
700 round
800
900
1000
1100
1200
Fig. 8. The system lifetime of the wireless mobile sensor network which has 100 sensors in the range of 200m× 200m and uses the Random Direction Mobility model; each sensor having initial energy 1.0J and random speed between 0.0 and 1.0 m/sec. System Lifetime(200m x 200m;1J,T=0.875), Random Walk Mobiltiy Model 100
System Lifetime We now consider the energy efficiency by measuring the system lifetime in terms of the total number of rounds which a wireless mobile sensor network system experiences. We first consider the round at which the first sensor dies. Recall that, if the number of cluster-heads in each round is even, the energy consumption is even among all the sensors and hence the system lifetime is longer. In particular, even number of cluster-heads in each round makes the first sensor died in a latter round. Table V shows this trends since our algorithms can make even number of cluster-heads among the rounds. In Table V, the first dead node comes out earlier when using Algorithm ACE-C compared with Algorithm ACE-L since Algorithm ACE-C may generate the cluster-heads close to each other.
90
: CM : ACE−C : ACE−L(5 rp) : ACE−L(4 rp)
80 accumulated number of dead nodes
Cluster Size As mentioned in Section II, by using the cluster-head election strategy of LEACH protocol, the difference in the number of cluster-heads between the rounds may be large and it happens that there is no cluster-heads at all in some rounds. The experiment results show that the range of the number of cluster-heads in a round can be from 0 to 10 when applying the rule in LEACH and the percentage of the rounds in which there is no cluster-heads elected is about 12% to 15%. Such a cluster-head election will deteriorate the performance in terms of the energy consumption. Table V shows the distribution of the number of cluster-heads generated in each round. The results also show that if one expects C cluster-heads in each round, our distributed algorithms will elect C cluster-heads exactly unless there are fewer than C sensors left in the system. Recall that algorithm ACE-L may generate less than C clusterheads when there are only few sensors left in the system. Therefore Algorithm ACE-C will generate more rounds having C cluster-heads than Algorithm ACE-L does.
70
60
50
40
30
20
10
0
300
400
500
600
700 round
800
900
1000
1100
1200
Fig. 9. The system lifetime of the wireless mobile sensor network which has 100 sensors in the range of 200m× 200m and uses the Random Walk Mobility model; each sensor has initial energy 1.0J and random speed between 0.0 and 1.0 m/sec.
Figure 8 and 9 show the system lifetimes when each sensor has initial energy 1.0 J in areas of 200m× 200m on Random Direction Mobility model and Random Walk Mobility model, respectively. From the plots, both of our algorithms make the system live longer than LEACH does; therefore, result in a better performance than LEACH in terms of energy consumption. This indicates that the two mechanisms we provide for electing cluster-heads indeed improve the performance because these two mechanisms can avoid the case that all the sensors send their data directly to the BS. In general, Algorithm ACE-L is better than Algorithm ACE-C since the position of cluster-heads in each round is more distributed in interested area. In some rounds, there could be five clusterheads in dense region for Algorithm ACE-C. In such a case, the cluster-heads will consume the energy dramatically as
Conf. on Wireless Networks (ICWN'05)
411
TABLE I T HE PERCENTAGE OF DIFFERENT NUMBER OF CLUSTER - HEADS ELECTED IN A
# of cluster-heads LEACH with CM ACE-C ACE-L
0 12% 0% 0%
1 3% 0% 0.6%
2 6% 0.2% 0.4%
3 15% 0% 0.4%
4 14% 0.1% 0.7%
5 15% 99.7% 97.9%
ROUND DURING A SYSTEM LIFETIME .
6 12% 0% 0%
7 10% 0% 0%
8 9% 0% 0%
over 9 4% 0% 0%
TABLE II T HE ROUND OF FIRST DEAD NODE FOR EACH ALGORITHMS . LEACH with CM ACE-C ACE-L
Random Walk Mobility Model 355 546 605
mentioned in Section IV. Furthermore, we also set the number of reference points to be 4 which are located at the centers of the four quadrants of the rectangle sensing region. For 5 reference points, we add one reference at the center of the sensing region. In our experiments, the result shows that using 4 reference points outperforms using 5 reference points in terms of the energy consumption. Such a result relates to the distribution of the reference nodes. Besides, the performance under the Random Direction Mobility model is better than the one under the Random Walk Mobility model since the CM mechanism predicts the sensor location more precisely in the Random Direction Mobility model. VI. C ONCLUSIONS In this paper, we first discuss the rule used in LEACH for cluster-head election and then provide two distributed algorithms which can elect C cluster-heads exactly if one expects C cluster-heads. The algorithm using counting can be applied to both static and mobile environments but may result in the case that all the cluster-heads are close to each other. Algorithm ACE-L using reference points is specially for the mobile sensors and performs better than Algorithm ACE-C. Combining our algorithms for cluster-head election and the CM mechanism for organizing the clusters, we can get two data-gathering protocols which lead to a longer system lifetime compared with LEACH; therefore, are more efficient in terms of energy consumption. Two mobility models are considered in the experiments: Random Walk Mobility model and Random Direction Mobility model. We are now currently working on analyzing what the best number and distribution of reference points are when employing Algorithm ACE-L to elect the cluster-heads. Besides, we also consider to use other mobility models in the experiment in order to analyze the impact of the mobility model on the cluster-head election. R EFERENCES [1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “Wireless sensor networks: a survey,” Computer Networks, vol. 38, pp. 393–422, 2002. [2] A. Cerpa, J.Elson, D. Estrin, L. Girad, M. Hamilton, and J. Zhao, “Habitat monitoring: Application driver for wireless communication technology,” in Proceedings of ACM SIGCOMM Workshop on Data Communications in Latin America and the Caribbean, 2001, pp. 3–5.
Random Direction Mobility Model 451 670 767
[3] Z. Cheng, M. Perillo, B. Tavli, W. Heinzelman, S. Tilak, and N. AbuGhazaleh, “Protocols for local data delivery in wireless microsensor networks,” in In Proceedings of the 45th IEEE Midwest Symposium on Circuits and Systems (MWSCAS ’02), 2002, pp. 623–626. [4] E. J. Duarte-Melo and M. Liu, “Data-gathering wireless sensor networks: organization and capacity,” Computer Networks, vol. 43, pp. 519–537, 2003. [5] K. Kalpakis, K. Dasgupta, and P. Namjoshi, “Maximum lifetime data gathering and aggregation in wireless sensor networks,” in Proceedings of the 2002 IEEE International Conference on Networking (ICN’02), 2002, pp. 685–696. [6] C.-M. Liu and C.-H. Lee, “Power efficient communication protocols for data gathering on mobile sensor networks,” in Proceedings of the 60th IEEE Vehicular Technology Conference, Fall, 2004. [7] C.-M. Liu, C.-H. Lee, and L.-C. Wang, “Power-efficient communication algorithms for wireless mobile sensor networks,” in Proceedings of the 1st ACM international workshop on Performance evaluation of wireless ad hoc, sensor, and ubiquitous networks, 2004, pp. 121–122. [8] S. Lindesy, C. Raghavendra, and K. M. Sivalingam, “Data gathering algorithms in sensor networks using energy metrics,” IEEE Transactions on Parallel and Distributed Systems, vol. 13, no. 9, pp. 924–935, 2002. [9] D. Petrovic, R. C. Shah, K. Ramchandran, and J. Rabaey, “Data funneling: Routing with aggregation and compression for wireless sensor networks,” in Proceedings of the First IEEE International Workshop on Sensor Network Protocols and Application, 2003, pp. 156–162. [10] M. Younis, M. Youssef, and K. Arisha, “Energy-aware management for cluster-based sensor networks,” Computer Networks, vol. 43, no. 5, pp. 649–668, 2003. [11] T. Camp, J. Boleng, and V. Davies, “A survey of mobility models for ad hoc network research,” Wireless Communication and Mobile Computing, vol. 2, no. 5, pp. 483–502, 2002. [12] D. Davies, “Evaluating mobility models within an ad hoc network,” Master’s thesis, Colorado School of Mines, 2000. [13] E. Royer, P. Melliar-Smith, and L. Moser, “An analysis of the optimum node density for ad hoc mobile networks,” in Proceedings of the IEEE International Conference on Communications (ICC01), 2001. [14] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energyefficient communication protocol for wireless microsensor networks,” in Proceedings of the 33rd Hawaii International Conference on System Sciences, 2000, pp. 1–10. [15] D. Tian and N. D. Georganas, “A node scheduling scheme for energy conservation in large wireless sensor networks,” Wireless Communications and Mobile Computing, vol. 3, pp. 271–290, 2003. [16] O. Younis and S. Fahmy, “Heed: A hybrid, energy-efficient, distributed clustering approach for ad-hoc sensor networks,” IEEE Transactions on Mobile Computing, vol. 3, no. 4, pp. 366–379, 2004. [17] L. Zhao, X. Hong, and Q. Liang, “Energy-efficient self-organization for wireless sensor networks: A fully distributed approach,” in Proceedings of IEEE Global Telecommunications Conference, 2004, pp. 2728 – 2732. [18] N. Bulusu, J. Heidemann, and D. Estrin, “Gps-less low cost outdoor localization for very small devices,” IEEE Personal Communication Magazine, vol. 7, no. 5, pp. 28–34, 2000. [19] T. S. Rappaport, Wireless Communications. Prentice-Hall, 1996.
