This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).
This paper presents an adaptive combinedmetricsbased clustering scheme for mobile wireless sensor networks, which manages the mobile sensors by utilizing the hierarchical network structure and allocates network resources efficiently A local criteria is used to help mobile sensors form a new cluster or join a current cluster. The messages transmitted during hierarchical clustering are applied to choose distributed gateways such that communication for adjacent clusters and distributed topology control can be achieved. In order to balance the load among clusters and govern the topology change, a cluster reformation scheme using localized criterions is implemented. The proposed scheme is simulated and analyzed to abstract the network behaviors in a number of settings. The experimental results show that the proposed algorithm provides efficient network topology management and achieves high scalability in mobile sensor networks.
Without a robust infrastructure, sensors in an adhoc network may be required to selforganize. Such sensor networks are selfconfiguring distributed systems and, for reliability, should also operate without centralized control. It has been shown that cluster architecture guarantees basic performance achievement in a mobile adhoc network [
In order to provide reliable communication in wireless adhoc networks, maintaining network connectivity is crucial. An implementation of the linked cluster architecture may consider the following tasks: cluster formation, cluster connectivity, and cluster reorganization. This paper presents an adaptive combinedmetricsbased clustering scheme for mobile sensor networks, which manages the mobile sensors by utilizing the hierarchical network structure and allocates network resources efficiently. In order to not to rely on a central controller, clustering is carried out by adaptive distributed control techniques via random waiting timers, which takes a number of metrics into account for cluster configuration, including neighboring node properties, residual energy level, and node mobility. To this end, the
In Phase I, clusterheads are selected and cluster members are assigned. The proposed algorithm generalizes the previous work [
Once the network topology is specified (as a hierarchical collection of clusters and distributed gateways), maintenance of the linked cluster architecture becomes an issue. In order to govern the hierarchical structure and the topology change efficiently, in Phase II a cluster reformation scheme using localized criterions is applied, especially on merger and division of clusters and reselection of clusterhead and gateway.
As we know, the clustering protocol is vital for a network to achieve scalability. However, a clusterbased network infrastructure requires extra cost for constructing and maintaining a cluster structure compared with a flatbased one. Therefore, the cost of clustering is a key issue to validate the effectiveness and scalability enhancement of a cluster structure. This paper explores the cost of the proposed scheme such as energy consumption for cluster formation and maintenance, communication complexity, and time complexity. By analyzing the proposed clustering scheme in different aspects qualitatively or quantitatively such as neighboring node properties, node density change, and simplified models for describing the process of cluster formation, its robustness and scalability can be clearly specified. The analytical results are compared to the behavior of the algorithm in a number of settings.
The rest of the paper is organized as follows: Section 2. reviews the current literature on clustering techniques for mobile adhoc networks. Section 3. describes the procedures of the SAMCA method. Section 4. presents the performance analysis of the SAMCA. In this section, the behavior of the proposed protocol is abstracted using neighboring sensor properties and simplified models, which serve to approximate its performance. Section 5. examines the energy usage of the algorithm. The result provided in [
Clusterbased selforganization is an important research topic for mobile adhoc networks since clustering allows to provide basic levels of system performance. A large variety of approaches for adhoc clustering have been proposed in literature [
The LowEnergy Adaptive Clustering Hierarchy (LEACH) of [
In [
The authors of [
Reference [
LCC [
In our previous works, the sensors are assumed to be fixed and the cluster size is not specified for cluster formation [
This section describes a dynamic distributed algorithm for organizing the mobile sensors using a timer and the characteristics of a network. The main assumptions are: (1) All sensors are homogeneous with the same transmission range; (2) Symmetric communication channel: all links between sensors are bidirectional; (3) Each sensor periodically broadcasts its local information to its neighbors. Note that there are no base stations to coordinate or supervise activities among sensors.
In order to transit network operation from transitional phase to steady phase smoothly, the system warmup time may be required in a mobile network. The study in [
After passing the warmup time, each sensor sets a random waiting timer for selecting clusterheads. Similar to the setting of warmup time, the initial value of waiting time of sensor
If the random waiting timer expires (i.e.,
Given the implementation of the SAMCA, there are three different kinds of sensors: (1) the clusterheads, (2) sensors with an assigned cluster ID, and (3) sensors without an assigned cluster ID, which will join any nearby cluster and become 2hop sensors. Thus, the topology of the adhoc network is now represented by a hierarchical collection of clusters. By adjusting randomized waiting timers, the sensors can coordinate themselves into sensible clusters, which can then be used as a basis for further communication and data processing. The procedures of cluster construction is outlined in the SAMCA of
Observe that the process of clusterhead selection induces nonoverlapping clusters. Accordingly, to interconnect two adjacent nonoverlapping clusters, one cluster member from each cluster must become a gateway This subsection presents a method of choosing distributed gateways for adjacent nonoverlapping clusters. Local information, such as neighboring density and mobility of the 1hop neighbors, is applied to select gateways and further achieve communication between clusters. The result of gateway selection is that each cluster
Denote the sensor which can communicate with more than one cluster as a border sensor. Note that only border sensors are eligible for gateway selection. Let
Based on the cluster formation, sensors can obtain local information and know the number of neighboring sensors in adjacent clusters. When initializing the gateway selection process, clusterheads broadcast messages within the clusters. In order to achieve intercluster communication, a pair of border sensors with a shorter distance and similar speed may have a larger chance to be gateways. Thus, the border sensors exchange information (e.g., the distances between the neighboring sensors and their speed information) to compute the weighting values
With the node mobility
The proposed cluster reformation strategy considers several cases (e.g., a sensor joining/leaving a cluster, cluster merging/splitting, and clusterhead/gateway reselection) and maintains clusters in a fully distributed way. In order to keep the hierarchical structure efficiently, load for each cluster should be equivalent. Thus, the cluster size is a key parameter to achieve balanced load among clusters. However, due to node movement, the cluster size changes as time proceeds. In the proposed approach, referring to the neighboring and its cluster sizes, each clusterhead triggers the cluster merge/split process if necessary. Let the upper bound
Moreover, in sensor network applications, it is possible that the energy level in the clusterhead is below the threshold or the clusterhead may malfunction during the network operation. As a result, reselecting a new clusterhead may be required. For maintaining the network connectivity, gateway reselection may be necessary as well in many situations such as discovering new clusters, the link down events of the distributed gateways, merging and splitting clusters, the energy issues and so on. Therefore, this selfadaptive organization is essential in mobile wireless sensor networking systems.
When a sensor node wants to join a cluster, it may check all the links with its neighbors and measure the distances by the received signal strength. Then it chooses the nearest neighbor and joins its cluster when the cluster size constraint is satisfied. Otherwise, the sensor forms a cluster of its own, which may be merged by nearby clusters later. When a sensor leaves a cluster, this link down event can be detected by not receiving periodical broadcasting messages and the neighboring sensors can update the knowledge of its neighborhood. Other possible schemes for handling new admissions and releases of sensors in a cluster can be found in [
This subsection describes how the two neighboring clusters
In order to balance the load among clusters, the cluster with a large cluster size should be divided into smaller clusters. There are many ways to solve this problem. One possible approach is to consider the split cluster as a small network and to apply the Phase I of the SAMCA algorithm (as detailed in Section 3.1.). Thus, if the cluster size
Notice that since the neighbor sensor properties depends on the type of sensor mobility occurring in the network, the upper bound
This subsection presents a method of choosing a new clusterhead for an existing cluster. The proposed distributed technique operates much like the SAMCA in utilizing a random timer. Once the energy in the current clusterhead is below the threshold of the energy level
For real applications, it is possible that the clusterhead may malfunction before broadcasting the reselection message. One solution is that if a certain amount of time has passed with no messages from the clusterhead, then the cluster members begin their timers and apply the algorithm to reinitialize the network into new clusters and to help balance the energy burden.
Like the clusterhead reselection, the contenders of the reselection process should satisfy the energy level constrain, which leads to the modified weighting value by multiplying a factor
To simply the performance analysis, the mobile sensors are supposed to be uniformly distributed within the restricted region and the sensor velocity distribution is the same across the region (i.e., the random walk mobility model). By analyzing the proposed clustering scheme in different aspects qualitatively or quantitatively such as neighboring node properties, node density change, and simplified models for describing the process of cluster formation, its robustness and scalability can be clearly specified.
The sensor distribution and neighboring sensor properties in mobile adhoc networks are crucial metrics for describing the network performance such as connectivity and topology management. This subsection derives the statistic properties of the critical neighbor number (CNN) in a typical network topology and proposes a sensible way to select the key parameters for the SAMCA algorithm such as the
According to the random walk mobility model and Theorem 1 [
Observe that the above analysis is suitable for any transmitting range. However, overly small transmission ranges may result in isolated clusters whereas overly large transmission ranges may result in a single cluster. Therefore, in order to optimize energy consumption and encourage linking between clusters, it is sensible to consider the critical transmission power which will result in a reasonably connected network. This
Given the random walk mobility model, Theorem 2 may suggest a good initial value for the search of optimized range assignment strategies to provide a high probability of connectivity The performance of the SAMCA with different selections of
This subsection analyzes the change of node density caused by the sensor movement and examines how mobility affects a clusterbased infrastructure. Since a high density distribution change means an unstable network infrastructure, the rate of the density change may be considered as an indicator of topology change. The following analysis investigates the characteristics of the link available time distribution between the clusterhead and a cluster member. Furthermore, the Lindeberg theorem [
The link available time distribution of a onehop connectivity between two sensors is described to investigate the property of the node density change rate in a cluster. At a given time
This subsection reviews the probability for analyzing the performance of the proposed scheme. Readers may refer [
Suppose for each
Denote
For our case, the Lindeberg condition [
By Theorem 3, the distribution of the number of sensors moving out of the cluster after time Δ
To proceed the investigation, let
Observe that the results presented in previous subsections show that
Based on Theorem 1, after some time
Since the connectivity among sensors and the mobility of sensors play important roles in the SAMCA, it is reasonable to investigate the performance from the perspective of these parameters. Moreover, in order to understand the transition of network operation, the effect of system warmup time on cluster formation is examined. Therefore, we abstract the behavior of the proposed algorithm in the transitional phase and steady phase of the sensor spatial distribution using two simplified models which approximate the desired global behavior and serve to analyze its performance.
The first simplified model, the Mobility and Density Model (MDM), is applied to describe the behavior of the algorithm in the transitional phase without experiencing the warmup operation. As detailed in
If the sensor is not already chosen as a clusterhead and its neighboring sensors are not already in other clusters, then the sensor with the largest
In the steady phase of network operation, since the sensor spatial distribution converges uniformly, we may describe the network topology using the Averaged Model (AM) and the statistic properties of neighboring sensor density. The AM approximates the number of neighboring sensors that will eventually join the new cluster by the expectation of the number of neighboring sensors of each sensor in the network. Thus, a simple formula for predicting the number of clusterheads is
Given the estimated number of clusters
This section considers the energy consumption of the SAMCA assuming homogenous sensors. The total power requirements include the power required to transmit messages
In the initialization phase, each sensor broadcasts a
The energy consumption for determining gateways is evaluated based on the description in Section 3.1. There are three possible determinations of a gateway in a cluster: (a) a 1hop cluster member, (b) a 1hop cluster member with a 2hop member, and (3) a 2hop cluster member. In order to simplify the presentation, the main notations are introduced as follows: let
When clusterheads broadcast messages to trigger the gateway selection procedure, the number of transmission
When applying the procedure for choosing gateways, the border sensors compute the weighting values and then broadcast messages to update the connectivity information and further activate the linked cluster architecture. Accordingly, the number of transmission
Thus, based on the energy needed to transmit and receive, the total energy consumption for gateway selection can be assessed by
For clusterhead selection process, each sensor initiates 2 rounds of local flooding to its 1hop neighboring sensors, one for broadcasting sensor ID and the other for broadcasting cluster ID, to select clusterheads and form 2hop clusters. Hence, the time complexity is
When merging two nearby clusters according to the procedures in Section , the energy consumption for determining the candidate cluster and initializing the merge process yields
Note that
For the split procedure, the energy consumption analysis in Section 5.1. can be applied since the split cluster may be regarded as a small network and the SAMCA algorithm may be used to spread the energy usage over the splitted cluster.
For cluster merging procedure,
The scenarios are generated with input parameters such as network size, speed, transmission range, and random waiting time for clusterhead contention in order to evaluate different aspects of performance such as cluster stability, overhead consumption, network behaviors and so on.
Consider a sensor moving within the bounded region
The sensor mobility is obtained by sampling from a uniform distribution with maximum value
The network topology management is investigated from three perspectives: (1) with varying the
From the first perspective, as shown in
From the second perspective,
Considering the third perspective,
With the random walk mobility model, the longterm sensor spatial distribution
Accordingly,
With parameters
Given the network density, the node mobility, and the cluster formation parameters,
This set of experiments considers the energy consumption of cluster formation (Phase I) in the SAMCA scheme. Assume that the communication channel is errorfree and the mean node mobility is
Referring to the analysis in Section 5.1. and the experimental result, the proposed selforganization scheme (
The effect of mobility on the link down rate is studied with varying transmission ranges. The various properties investigated in Section 4. characterize the behavior of the links of mobile sensors, which can be used as a basis for analyzing the performance bounds of the proposed SAMCA protocol. Thus, the link dynamics may be used as an indicator of topology change and the distribution of link available time can be used to examine the link stability in the network. Given the network density (
The network behaviors in the transitional and steady phases are explored when applying the SAMCA approach and the simplified models in a random network of 100 sensors. In each method, the results of 100 typical runs are merged. Given in
The graph suggests that the MDM method and the AM method well approximate the SAMCA performance. This is reasonable because the MDM retains global connectivity and mobility information for cluster construction in the transitional phase. Due to the uniform convergence of the sensor spatial distribution, the approximation of the AM method may be a way to predict the performance of the SAMCA. Moreover, with an appropriate transmission range (
The final set of experiments compares the robustness of the SAMCA algorithm with other clusterbased selforganization schemes. Simulation study is conducted to show that the performance of the ESAC [
This paper introduces an adaptive distributed clustering scheme for mobile wireless sensor networks. The proposed algorithm performs cluster formation and linkage using random waiting timers and local information. It investigates the key features of cluster construction and maintenance such as transmission range and constraints of cluster size so that the load and overhead of clusterheads in each cluster can be balanced. As shown in simulation experiments, the proposed scheme achieves cluster stability and adaptability in mobile sensor networking systems. Moreover, the experimental results also demonstrate that the simplified approximate models and analysis can well describe the network behaviors, which suggests that the approximation may be a sensible way to assess the performance of the proposed algorithm. On the basis of the clusterbased network topology, this selfconfiguring technique can be applied to provide efficient topology management in mobile wireless sensor networks.
The update strategy for the random waiting time of sensor
Virtual cluster construction flowchart for the SAMCA algorithm.
The initial cluster formation of random networks of 100 (left) and 200 (right) sensors with
Example of cluster merge process of a random network; the network topology before cluster merging (left) and the network topology after cluster merging (right). The red ∇s represent clusterheads and the black ∇s (right) represent role changes from a clusterhead to a cluster member in the merged cluster.
Example of cluster split process of a random network; the network topology before cluster splitting (left) and the network topology after cluster splitting (right). The yellow ∇s represent the cluster members in the split cluster (left).
Example of distributed gateway reselection during the cluster merge procedure. The connection between a pair of distributed gateways is illustrated by a dashed line: (a) before gateway reselection and (b) after gateway reselection.
The typical runs of random networks of 200 sensors with a given mean node speed (2.5 m/sec), different
The relationship between the average number of clusters and the
Average number of clusters for considering different cluster constraints [
The impact of node mobility on cluster stability given the network density and the cluster formation parameters,
The impact of cluster parameter selection on cluster formation over a time period of 200 seconds.
Average number of clusterhead (left) and gateway (right) changes with varying the node mobility and the cluster constraint parameters; the number of sensors
The comparison of the initial cluster formation (filled symbol) and the stable one (hollow symbol) for different settings of cluster parameters (left); the number of cluster merging and splitting in a network given the network density
The frequency of cluster merging and splitting in a network given the network density
The comparison of a typical network topology applying the proposed scheme (left) and that using the MOBIC [
The energy consumption for cluster formation: the clusterhead selection and distributed gateway selection.
The average
The comparison of the number of clusterheads (left) and distributed gateways (right) when applying the SAMCA approach and the simplified models.
The average number of clusterhead changes using the SAMCA and the ESAC heuristic.
The Mobility and Density Model.
Assign a probability to sensor Let Assign
Select a clusterhead
Update the probability distribution
Normalize the updated probability distribution.
Store the normalized probability distribution.
