report

Introduction

When a sensor network is first activated, various tasks must be performed to establish the necessary infrastructure that will allow useful collaborative work to be performed. In particular, each node must discover which other nodes it can directly communicate with, and its radio power must be set appropriately to ensure adequate connectivity. Nodes near one another may wish to organize themselves into clusters, so that sensing redundancy can be avoided andscarce resources, such as radio frequency, may be reused across nonoverlapping clusters.

A wireless sensor network may consist of a large number of sensor nodes, and each node is equipped with sensors, microprocessors, memory, wireless transceiver, and battery. Once deployed, the sensor nodes form a network through short-range wireless communication. They collect environmental surveillance data and send them back to the data processing center, which is also called the sink node or base station. In many applications, wireless sensor networks are used to monitor some measures of interest, such as temperature, light intensity, air pressure, humidity etc.

The wireless sensors are mostly deployed in remote and hazardous locations, where manual monitoring is very difficult or almost impossible. Due to the low cost of wireless sensors, these can be deployed in large numbers. Apart from sensing, sensor nodes are equipped with data processing and communication capabilities. The sensing circuitry measures the parameters of interest (temperature, pressure, etc.) within its sensing range and transforms them into electrical signals. These electrical signals are processed and with the help of onboard radio they are transmitted to the remotely located sink node. Due to deployment of wireless sensors in unattended harsh environment, it is not possible to charge or replace their batteries. Therefore, energy efficient operation of wireless sensors to prolong the lifetime of overall wireless sensor Network is of utmost importance. Most of the energy consumption in wireless sensor node is attributed to transmitting/receiving, processing, and forwarding the data to neighboring nodes. The dense deployment and unattended nature of WSNs make it quite difficult to recharge node batteries. Therefore, energy efficiency is a major design goal in these networks.

Grouping sensor nodes into clusters has been widely used to achieve this objective. Clustering is especially important for sensor network applications where a large number of ad-hoc sensors are deployed for sensing purposes. If each and every sensor start to communicate and engage in data transmission in the network, a great network congestion and data collisions will be experienced. This will result to drain limited energy from the network. Node clustering will address these issues. Scalability of the network of those WSNs are useful to meet load balancing and efficient resource utilization constraints.

In cluster networks, sensors are partitioned into smaller clusters and cluster head (CH) for each cluster is elected. Sensor nodes in each cluster transmit their data to the respective CH and CH aggregates data and forward them to a central base station. Clustering through creating a hierarchical WSN facilitates efficient utilization of limited energy of sensor nodes and hence extends network lifetime. Although sensor nodes in clusters transmit messages over a short distance (within clusters), more energy is drained from CHs due to message transmission over long distances (CHs to the base Station) compared to other sensor nodes in the cluster. Periodic

re-election of CHs within clusters based on their residual energy is a possible solution to balance the power consumption of each cluster. In addition, clustering increases the efficiency of data transmission by reducing number of sensors attempting to transmit data in the WSN, aggregating data at CHs via intra-cluster communication and reducing total data packet loses.

General Sensor Network Architecture

The following figure shows the general sensor network architecture.

Fig. 1. General Sensor Network Architecture

Sensor Node: A sensor node is the core component of a WSN. Sensor nodes can take on multiple roles in a network, such as simple sensing; data storage; routing; and data processing.

• Clusters: Clusters are the organizational unit for WSNs. The dense nature of these networks require the need for them to be broken down into clusters to simplify tasks such a communication.

• Clusterheads: Clusterheads are the organization leader of a cluster. They often are required to organize activities in the cluster. These tasks include but are not limited to data-aggregation and organizating the communication schedule of a cluster.

• Base Station: The base station is at the upper level of the hierarchical WSN. It provides the communication link between the sensor network and the end-user.

• End User: The data in a sensor network can be used for a wide-range of applications. Therefore, a particular application may make use of the network data over the internet, using a PDA, or even a desktop computer. In a queried sensor network (where the required data is gathered from a query sent through the network). This query is generated by the end user.The clustering phenomenon , plays an important role in not just organization of the network, but can dramatically affect network performance.

Pros and Cons of Clustering

The pros of Clustering are that it enables bandwidth reuse thus can improve the system capacity. Due to the fact that within a cluster, all the normal nodes send their data to the CHs so energy saving is achieved by absence of flooding, multiple routes, or routing loops. Due to the fact that clustering enables efficient resource allocation and thus help in better designing of power control and other advantage is due to the fact that any changes of nodes behavior within a cluster affect only that cluster but not the entire network, which will therefore be robust to these changes.

There are also several cons of existing clustering schemes in WSNs like in the selection of the cluster heads, some algorithm selects cluster heads only according to the ID number or residual energy of the sensor nodes. Since all the data in sensor network are sent to the base station, the traffic near the base station is higher. The sensor nodes in these areas will therefore run out energy earlier. The base station will then be isolated and as a result, the residual energy stored in the other sensor nodes will be wasted. Another disadvantage is the energy is wasted by flooding in route discovery and duplicated transmission of data by multiple routes from the source to the destination.

Design challenges in clustering algoriths

Wireless Sensor Networks present vast challenges in terms of implementation. Design goals targeted in traditional networking provide little more than a basis for the design in wireless sensor networks .Decomposition of a WSN into smaller clusters is considered to be a convenient and an efficient approach to prolong network lifetime through efficient energy utilization of WSN. Some important design considerations in designing clustering algorithms are discussed below:

• Limited Energy: Wireless sensor nodes have limited energy storage and once they are deployed, it is not practical to recharge or replace their batteries. With the capability of reducing the amount of data transmission, the clustering algorithms are more energy efficient compared to the direct routing algorithms. This can be achieved by balancing the energy consumption in sensor nodes by optimizing the cluster formation, periodically reelecting CHs based on their residual energy, and efficient intra-cluster and inter-cluster communication. But clustering algorithms should prevent high energy cluster reconstruction process.

• Network Lifetime: The energy limitation on nodes results in a limited network lifetime for nodes in a network. Clustering helps to prolong the network lifetime of WSNs through reducing in the number of nodes contending for channel access, data aggregation at CHs via intra-cluster communication and direct or multi-hop communication by CHs with a base station. Proper design should focus on increasing network lifetime.

• Limited Abilities: The small physical size and small amount of stored energy in a sensor node limits many of the abilities of nodes in terms of processing, memory, storage,and communication.

• Application Dependency: When designing protocols for WSNs, application robustness must be focused on, as protocols should be able to adapt to a variety of application requirements. Changes in the deployment environment can also be observed due to variations in the surrounding conditions.

• Secure Communication: The ability of a WSN to provide secure communication is ever more important when considering these networks for military applications. The self-organization of a network has a huge dependence on the application it is required for. An establishment of secure and energy efficient intra-cluster and inter-cluster communication is one of the important challenges in designing clustering algorithms since these tiny nodes are deployed unattended in most cases.

• Cluster formation and CH selection: Cluster formation and CHs selection are two of the important operations in clustering algorithms. Energy wastage in sensors in WSN due to direct transmission between sensors and a base station can be avoided by clustering the WSN. Clustering further enhances scalability of WSN in real world applications. Selecting optimum cluster size, election and reelection of CHs, and cluster maintenance are the main issues to be addressed in designing of clustering algorithms. The selection criteria to isolate clusters and to

choose the CHs should maximize energy utilization, as well as function for a variety of applications.

• Synchronization: Slotted transmission schemes such as TDMA allow nodes to regularly schedule sleep intervals to minimize energy used. Such schemes require synchronization mechanisms to setup and maintain the transmission schedule and the effectiveness of this mechanism must be considered.

• Data Aggregation: Data aggregation allows the differentiation between sensed data and useful data. In a densely populated network there are often multiple nodes sensing similar information. In network processing this process makes energy optimization possible and now it is fundamental in many sensor network schemes, as the power required for processing tasks is substantially less than communication tasks. As such, the amount of data transferred in-network should be minimized.

• Repair Mechanisms: Due to the nature of Wireless Sensor Networks, they are often prone to node mobility, node death and interference. All of these situations can result in ink failure. When looking at clustering schemes, it is important to look at the mechanisms in place for link recovery and reliable data communication.

• Quality of Service (QoS): Existing clustering algorithms for WSN mainly focus on providing energy efficient network utilization but pay less attention to QoS support in WSN. From an overall network standpoint, we can look at QoS requirements in WSNs. Many of these requirements are application dependant such as acceptable delay and packet loss tolerance. For example in applications such as habitat monitoring, there is no bound on acceptable delay, however in military tracking, even a small delay is unacceptable. QoS metrics must be taken into account in the design process.

Classification of clustering algoriths

Clustering in WSNs involves grouping nodes into clusters and electing a CH such that:

• The members of a cluster can communicate with their CH directly.• A CH can forward the aggregated data to the central base station through other CHs.

Thus, the collection of CHs in the network forms a connected dominating set. Research on clustering in WSNs has focused on developing centralized and distributed algorithms to compute connected dominating sets. Distributed approaches are more practical for large-scale deployment scenarios. Since obtaining an optimal dominating set is an NP-complete problem, the proposed algorithms are heuristic in nature.

We classify the clustering techniques based on two criteria:• The parameter(s) used for electing CHs• The execution nature of a clustering algorithm (probabilistic or iterative)

In self-organization schemes, CHs are nodes that consume more energy than cluster members when they involve in aggregating, processing and routing data. Clustering algorithms are heuristic in nature and NP hard. Distributed clustering algorithms are more feasible compared to centralized clustering algorithms since central control of large number of sensor nodes are not practical. Only distributed clustering algorithms therefore are considered in this analysis.

Existing clustering algorithms can be categorized into four groups vertically depending on cluster formation criteria and parameters used for CH election. Some algorithms make decisions on cluster formation and CHs selection based on pre-collected network information or heuristics with some specific assumptions on some desirable properties. Node identifiers, node weights based on the significance of the sensor node, number of neighboring nodes, probabilities assigned to nodes and residual energy of nodes are common parameters in selecting CHs. If overheads of collecting prior information about the network are significant or heuristics and assumptions are not much realistic, energy efficiency and higher network lifetime may not be achieved. Based on the cluster formation methodology and CH selection criteria, clustering algorithms are classified into:

1. Identity-based clustering algorithms,2. Neighborhood information based clustering algorithms, 3. Probabilistic clustering algorithms and4. Biologically inspired clustering algorithms

We will now provide an overview of the clustering algorithms that are most commonly considered when investing the self-organization of WSNs.

1. Identity-based clustering algorithms : Uniformly assigned unique identifiers are the key parameter for selecting CHs in Identity-based clustering algorithms [1][2]. For a sensor node to be the CH, it should have the highest identity among all one-hop sensor nodes [1] or the lowest identity among all nodes that are neither a cluster head nor are within one hop of already selected CH [2]. These algorithms may not favor the energy limited sensor networks since they drain the power of some nodes in the network. Generally these algorithms are coming under static clustering algorithms and do not change the CHs once selected. However, energy efficiency is not a primary objective of most of Identity-based clustering algorithms. A load balancing heuristic can be added to these algorithms and hence, longer, low variance CH durations can be achieved.The following are Identity-based clustering algorithms:

Linked Cluster Algorithm (LCA)]: In LCA, each node is assigned a unique ID number and has two ways of becoming a CH. The first way is if a node has the highest ID number in the set including all neighbor nodes and the node itself. The second way, assuming none of its neighbors are CHs, then it becomes a CH. LCA2 [2], an extension of LCA, was proposed to eliminate the election of an unnecessary number of CHs. Here the concept of a node being covered and noncovered is introduced. A node is covered if one of its neighbors is a CH. CHs are elected starting with the node having the lowest ID among non-covered neighbors. LCA2 generates smaller number of clusters compared to LCA.

LCA and LCA2 have very limited scope as clustering algorithms for WSN since they did not address the issue of limited energy of WSN. Both algorithms form 1-hop clusters requiring clock synchronization with time complexity of O(n). Load balancing of LCA/LCA2 focuses only on intra-cluster communication and is not favorable for real applications.

2. Neighborhood information based clustering algorithms :

In neighborhood information based clustering algorithms; sensors should have information about their neighbors and should be able to decide on number of neighbors within a pre-specified transmission range (cluster range). Based on connectivity-based heuristics considering number of neighbors, some algorithms elect sensors with maximum number of 1-hop neighbors as the CHs . Some other algorithms under this category use a combination of metrics in addition to node degree such as: transmission power; mobility; and the remaining energy of the nodes . Depending on specific application, any or all of these parameters will be utilized for CH selection. The power consumption at CHs can be reduced by using load balancing heuristic in these algorithms. This may further build a larger number of clusters within the network creating congestion in data routing to a base station. Re-clustering or CH reelection is not considered in these algorithms and mostly they are static clustering algorithms. The following are neighborhood information based clustering algorithms:

Highest-Connectivity Cluster Algorithm : In this scheme each node broadcasts the number of neighbors it has to the surrounding nodes. Here the connectivity of a node is considered. The node with the highest connectivity (highest degree) is elected CH, but in the case of a tie, the node with the lowest ID prevails. Node which has already selected a CH withdraws its intention to be a CH. The connectivity– based heuristic used in this scheme elects the sensor with maximum number of 1-hop neighbors as the CH. The creation of one-hop cluster and clock synchronization requirement limit the practical usage of the algorithm.

Max-Min D-Cluster Algorithm: This is a distributed algorithm for CH election, where no node is more than d hops away from the CH where d (>1) is a value selected for the heuristic in the algorithm. CHs are selected based on their node ID. Therefore a change in the network topology will not have much influence on the node clusters and CHs.

The time complexity of generating d-hop clusters is O(d) and this algorithm does not require clock synchronization as LCA/LCA2 or highest connectivity algorithm. This algorithm further provides a better load balancing compared to LCA/LCA2 algorithm and highest connectivity cluster algorithm.

Weighted Clustering Algorithm (WCA) : This algorithm is a non-periodic procedure for CH election. It is invoked on demand every time a reconfiguration of the network’s topology is unavoidable. A new election is invoked every time a sensor loses the connection with any CH, thus saving power. WCA is based on a combination of metrics that take into account several system parameters such as: the ideal node degree; transmission power; mobility; and the remaining energy of the nodes. Depending on the specific application, any or all of these parameters can be used as a metric to elect CHs. The election procedure is based upon a global parameter that is called combined weight.

Linked Cluster Algorithm (LCA)]: One of the first clustering algorithms developed and was meant for wired sensors, but has since been implemented for Wireless Sensor Networks. In LCA, each node is assigned a unique ID number and has two ways of becoming a CH. The first way is if a node has the highest ID number in the set including all neighbor nodes and the node itself. The second way, assuming none of its neighbors are CHs, then it becomes a CH. LCA2 , an extension of LCA, was proposed to eliminate the election of an unnecessary number of CHs. LCA2 generates smaller number of clusters compared to LCA. LCA and LCA2 have very limited scope as clustering algorithms for WSN since they did not address the issue of limited energy of WSN. Both algorithms form 1-hop clusters requiring clock synchronization with time complexity of O(n). Load balancing of LCA/LCA2 focuses only on intra-cluster communication and is not favorable for real applications. This algorithm attempted to provide better load balancing through reduced number of sensors in a cluster but the requirement of clock synchronization limits its applications.

Grid-clustering ROUting Protocol (GROUP) : In this algorithm one of the sinks (called the primary sink), dynamically and randomly builds the cluster grid, where CHs are arranged in a grid-like manner. Forwarding of data queries from the sink to source

node is propagated from the Grid Seed (GS) to its CHs, and so on. In the case of a location unaware data query the query is passed from the central most sink in the network to its nearest CH. That CH will then broadcast the message to neighboring CHs. If the data is location aware, then the requests are sent down the chain of CHs towards the specified region using unicast packets. For both data queries, data is transmitted upstream through the chain of CHs established during cluster formation. Energy conservation is achieved due to the lower transmission distance for upstream data.

PEGASIS : It offers promising improvements with relation to network lifetime, however reliability may not be as promising. In PEGASIS, each node communicates with its nearest neighbor. This implementation may be more susceptible to failure due to gaps in the network.

Clustering Algorithm via Waiting Timer - CWAT: A decentralized algorithm for organizing clusters has been proposed for homogeneous sensors with the same transmission range. The performance of CWAT was evaluated using simplified simulations. It is observed that the generalization of the proposed algorithm is needed to see its performance with respect to load balancing, CH reelection and energy usage across the network.

3 .Probabilistic clustering algorithms:

In Probabilistic clustering algorithm, a prior probability assigned to each sensor node is used to determine CHs. The probabilities assigned to individual node in the cluster facilitate individual node to decide on their election as a CH in the cluster while considering few other primary parameters. In addition to the probability assigned to each node, residual energy at nodes or node degree is taken as the primary parameter to elect CH. Clustering algorithms in this category shows faster convergence in addition to energy efficient network utilization, efficient load balancing and low message overheads. The following are Probabilistic clustering algorithms:

Low-Energy Adaptive Clustering Hierarchy (LEACH) : This algorithm was one of the first major improvements on conventional clustering. It provides a balancing of energy usage by random rotation of CHs. The algorithm is also organized in such a manner that data fusion can be used to reduce the amount of data transmission. The decision of whether a node elevates itself to CH is made dynamically at each interval, to minimize overhead in CH establishment. This decision is a function of the percentage of optimal CHs in a network (determined a priori on application), in combination with how often and the last time a given node was the CH. This scheduling scheme allows for energy minimization as nodes can turn off their radio during all but their scheduled time-slot.

Therefore LEACH provides an uniform load balancing in one-hop sensor networks. Localized coordination scheme used in LEACH provides better scalability for cluster formation and better load balancing enhances the network lifetime.

Two-Level LEACH (TL-LEACH) [14] : This algorithm is an extension to LEACH and utilizes two levels of CHs (primary and secondary). The primary CH in each cluster communicates with the secondaries and the corresponding secondaries communicate with the nodes in their subcluster. Data-fusion can also be performed as in LEACH. In addition, communication with a cluster is still scheduled using TDMA time-slots. The organization of a round will consist of first selecting the primary and secondary CHs using the same mechanism as LEACH, with a priori probability of being elevated to a primary CH less than that of a secondary node. The two-level structure of TL-LEACH reduces the amount of nodes that need to transmit to the base station, effectively reducing the total energy usage.

Energy Efficient Clustering Scheme (EECS) : EECS is similar to LEACH with some enhancement in cluster formation and cluster head selection process. According to residual energy of sensor nodes, a constant number of CHs are elected using localized competition process without iteration. In EECS, clusters are formed by dynamic sizing of clusters based on cluster distance from the base station. The result is an algorithm that addresses the problem that clusters at a greater distance from the base station require more energy for transmission than those that are closer. This provides much lower message overheads and uniform distribution of CHs compared to LEACH.

Hybrid Energy Efficient Distributed Clustering (HEED) : HEED is a multi-hop clustering algorithm for Wireless Sensor Networks. CHs are chosen based on two important parameters: residual energy and intra-cluster communication cost. Residual energy of each node is used to probabilistically choose the initial set of CHs, as commonly done in other clustering schemes. In HEED, Intra-Cluster Communication Cost reflects the node degree or node’s proximity to the neighbor and is used by the nodes in deciding to join the cluster. Low cluster power levels promote an increase in spatial reuse while high cluster power levels are required for inter-cluster communication as they span two or more cluster areas. HEED provides a uniform CH distribution across the network and better load balancing. However, knowledge of the entire network is needed to determine intra-cluster communication cost and configuration of those parameters might be difficult in the practical world.

Time Controlled Clustering Algorithm - TCCA : Similar to LEACH, the operation of TCCA is divided into rounds enabling better load distribution among sensor nodes. Each round consists of a cluster setup phase targeting at cluster formation and CH selection, and a steady state phase focusing cyclic collection, aggregation and transfer of data at CH to a base station. Node’s residual energy and a desired CH probability are considered in the eligibility criteria for CH selection. Once the CH is selected it advertises its selection as the CH to the neighboring nodes by sending an advertisement message (ADV) which includes its node id, initial Time-To-Live (TTL), its residual energy and a timestamp. TTL selected based on residual energy of the node is used to limit the size of the cluster to be formed.

4. Biologically inspired clustering algorithms:

Recently proposed biologically inspired clustering algorithms utilize swarm intelligence techniques which model the collective behavior of social insects like ants. These algorithms are not yet matured and improvements are to be sought. In these clustering algorithms, colonial closure model which has been derived based on ant colonies are used. Biologically inspired clustering algorithms show that they can dynamically control the CH selection while achieving uniform distribution of CHs and optimal number of clusters.Following is biologically inspired clustering algorithm:

ANTCLUST based clustering : In this, Swarm Intelligence based clustering algorithm has been proposed. Swarm intelligence is a technique used to model the collective behavior of social insects like ants and shows the properties of robustness, distributed problem solving capabilities, de-centralized performance. ANTCLUST is a model of an ant colonial closure to solve clustering problems. In the ANTCLUST-based clustering method, sensor nodes with more residual energy become cluster heads independently. Then, randomly chosen nodes meet with each other and clusters are created, merged, and discarded through local meetings. Each sensor node with less residual energy chooses a cluster based on the residual energy of the cluster head, its distance to the cluster head, and an estimation of the cluster size. Eventually energy efficient clusters are formed that result in an extension of the lifetime of a sensor network.

Conclusion:

Wireless sensor networks (WSNs) have attracted significant attention over the past few years. A growing list of civil and military applications can employ WSNs for increased effectiveness; especially in hostile and remote areas. Examples include disaster management, border protection, combat field surveillance. In these applications a large number of sensors are expected, requiringcareful architecture and management of the network.

Grouping nodes into clusters has been the most popular approach for support scalability in WSNs.Clustering is an important technique that

▫ Prolongs network lifetime▫ Reduces channel contention▫ Reduces collisions

Significant attention has been paid to clustering algorithms.

References:

[1] Ossama Younis, Marwan Krunz, and Srinivas Ramasubramanian, “Node Clustering in Wireless Sensor Networks: Recent Developments and Deployment Challenges”, IEEE Network, 2006.[2] P. Kumarawadu*, D. J. Dechene+, M. Luccini+, and A. Sauer, “Algorithms for Node Clustering in Wireless Sensor Networks: A Survey”, IEEE,2008[3] D.J. Baker and A. Epheremides, “The Architectural Organization of a Mobile Radio Network via a Distributed Algorithm,” IEEE Transactions on Communications, 1981, vol. Com-29,no.11. [4] M. Chatterjee, S. K. Das and D. Turgut, “WCA: A Weighted Clustering Algorithm for Mobile Ad Hoc Networks,” Clustering Computing, 2002, vol. 5, pp. 193–204[5] A. Amis, R. Prakash, T. Vuong and D. Huynh, “Max-Min D-Cluster Formation in Wireless Ad Hoc Networks,” IEEE INFOCOM, March 2000.[6] M. Ye, C. Li, G. Chen and J. Wu, “An Energy Efficient Clustering Scheme in Wireless Sensor Networks,” Ad Hoc & Sensor Wireless Networks,2006, vol.1, pp.1–21.[7] Mao YE, Chengfa LI, Guihai CHEN and Jie WU, “An Energy Efficient Clustering Scheme in Wireless Sensor Networks”, Ad Hoc & Sensor Wireless Networks, 2005, Vol. 3, pp. 99-119