Abstract
Wireless sensor networks (WSN) are a recent class of distributed systems that consist of sensors. These sensors are essential in monitoring environmental or physical states, like temperature, sound, movement, vibration, or pollutants. As compared to most computers, which primarily rely on data supplied by users, sensor networks take into consideration the surrounding environment. This new concept has given rise to many new protocols that are specially designed for WSN, in which energy effects are considered. Consequently, much attention is given to routing protocols because they tend to be different depending on the network and application design. This paper analyses the various routing protocols for sensor networks and gives a categorization for the different methods pursued. Thus, the three main classes of routing protocols discussed in this research are data-centric, hierarchical, and location-based. Every routing protocol is evaluated under the necessary category. The research concludes with a general comparison and discussion of research issues.
Introduction
The growth of wireless sensor networks has motivated many researchers to take the advantage of sensor opportunities. WSN is now used in several industrial applications, such as process monitoring, machine health monitoring, and traffic control (Ramesh et al. 2006). In sensor networks, micro-sensors are normally developed with information processing and communication abilities. The sensing route determines the light that surrounds the sensor environment and transmits the ambient into electric signals. Transforming such signals enables the network system to know the characteristics of the objects or events occurring in the environment. The sensor then sends the collected information to the sink (command center) through a transmitter. The reduced cost and size of sensors due to technological innovations has increased the interest of researchers in expanding the use of unattended sensors. This has motivated serious research in recent years addressing the possibility of cooperation among sensors in processing data and the management of the sensing events and flow of information to the sink. A normal structural design for such shared distributed sensors is a network designed in an ad hoc manner, consisting of wireless links.
Networking idle sensor nodes are deemed to have a lot of impact on the reliability of most civil and military applications like safety, surveillance, and disaster administration. These systems work on data collected from different sensors to monitor activities in a specific area. Conversely, sensor nodes are restrained in energy distribution and bandwidth. Such limitations combined with a usual operation of many sensor nodes have created many issues in the design and organization of sensor networks. These problems enhance energy consideration at all the networking protocol layers. At the network layer, the main consideration is to find methods of setting up an energy-efficient route and reliable transmission of data from the sensor nodes to the command center so that network survival is sustained (Akkaya & Younis 2003).
Moreover, due to many features that differentiate sensor networks from modern communication and wireless ad hoc networks, routing in WSN is very challenging. Firstly, it is difficult to develop a global addressing scheme for the implementation of many sensor nodes. Thus, standard IP-based protocols cannot be used in sensor networks. Secondly, most applications of sensor networks require the flow of information from multiple sensor nodes to a specific sink, as opposed to classic communication networks. Thirdly, the data traffic produced has many redundancies because different sensors may give the same information in the environment. Such repetition needs to be dealt with by the routing protocols to enhance energy and bandwidth use. Fourthly, sensor nodes are restricted in regard to transmission power, energy, and processing ability, thus necessitates proper resource administration.
As a result of the mentioned differences, many new algorithms have been designed to address these challenges. These algorithms or rather routing mechanisms have taken into consideration the features of sensor nodes together with application and network requirements. These routing protocols can be categorized as data-centric, hierarchical or location-based. However, there are few protocols which are based on quality of service (QoS) knowledge. Data-centric protocols are based on query and rely on naming scheme used in data transmission. Hierarchical protocols try to cluster the nodes so that cluster heads can perform some estimates and lessening of data so as to save energy. Location-based protocols use the information from a location to route data to the realistic point in order to save energy (Raghavendra et al. 2004).
Therefore, this paper discusses the three categories of routing protocols for sensor networks. The aim is to compare and evaluate the routing protocols and discuss issues regarding the research of WSN routing protocols.
Data-Centric Protocols
Assigning global identifiers is difficult in many sensor networks due to the most number of nodes implemented. The absence of global identification together with arbitrary deployment of sensor nodes presents some difficulties in selecting a definite set of sensor nodes to be inquired. Thus, redundancy of data is paramount due to the transmission of data from every node. Since this affect energy consumption, it is possible to take into consideration the routing protocols that are able to determine a set of sensor nodes and efficiently use data aggregation in transmitting data. This has given rise to data-centric routing, which is dissimilar from conventional address-based routing in which routes are determined among addressable nodes in the network layer (Ramesh et al. 2006).
In data-centric routing, the command center (sink) transmits queries to several regions and waits for data from the sensor nodes residing in the regions. Because data is requested through querying, data-centric protocol is characterized by attribute based naming and there is data aggregation during relying (Rousselot 2006). Therefore, this section discusses in details the data-centric protocols.
Flooding and gossiping
Flooding and gossiping are standard methods for transmitting data in WSN without the assistance of routing algorithms and topology management. In flooding, each sensor node broadcasts data packets to the neighboring sensors until the destination is reached or a maximum number of packet relay is reached. Whereas, gossiping is an integrated version of flooding, in which the sensor receiving a data packet sends the packet to its neighbor after random selection of the destination, the selected sensor also picks another neighbor randomly to relay the packet to and the process continues.
Despite its simplicity in implementation and no need of state maintenance, flooding experiences implosion, overlap, and resource blindness. Implosion is as a result of redundant messages relayed to the same node, overlap is when different nodes sensing the same section send similar data packets to the same sensor node, and resource blindness is as a result of uncontrolled energy consumption. Implosion is avoided by gossiping since a random node is selected for packet transmission instead of broadcasting. Though, this results to delays in data propagation within the sensor nodes (Lee 2004).
SPIN – Sensor Protocols for Information via Negotiation
The SPIN protocol names the data packets using meta-data so as to enhance its event-driven capability. At the beginning of transmission, meta-data are interchanged among sensor nodes through a data advertisement system, acting as the main idea of SPIN. After receiving new data from the neighboring node, each node advertises the data to the interested nodes. Thus, any node in need of the data, acquires the data through sending a request message (Ramesh et al. 2006). This feature of negotiation of meta-data or high-level descriptors solves the problem of implosion, overlapping of sensing regions, and resource blindness experienced in flooding. In this case, high energy efficiency is experienced. However, there is no classic format for meta-data and it is deemed to be application specific. Three types of messages are common in SPIN protocol; they enhance the exchange of data among sensor nodes. These include: ADV message that enable a sensor node to advertise a certain high-level descriptor, REQ message that enable a node to request a particular data, and DATA message which carries the exact data (Akkaya & Younis 2003).
One of the benefits of SPIN is that it localizes the topological changes because every node is required to identify only its single-hop neighbors. Another advantage is that SPIN uses energy efficiently as compared to flooding due to the fact that redundant data is reduced. Conversely, the data advertisement mechanism of SPIN does not assure the delivery of data. For example, if the nodes that are in need of the data are located further from the source nodes and the close neighbors of the source nodes are not in need of the data, it implies that the data may not reach its destination. Thus, SPIN is not a proper choice for applications which need consistent data delivery, such as intrusion detection.
Directed Diffusion
This routing protocol diffuses data through sensor nodes by integrating the naming scheme of the data in its process. This is in a view to eliminate unwanted operations of routing in the network layer so as to save energy. Direct Diffusion advocates for the implementation of attribute-value pairs for data and uses an on demand strategy to query the sensors based on the data pairs. For the querying to take place, an interest is generated by a list of attribute-value pairs like duration, interval, name of objects, and so on. The interest is then broadcasted by a sink through its neighbors. A node acquiring the interest can store it for later use. In addition, the nodes have the capability to perform data aggregation. The stored interests are subsequently used to evaluate the received data packets with the interest’s values. Numerous gradient fields are held in the interests. A gradient is a response link from a source neighbor which transmitted the interest. It is described by the duration, data rate, and time of expiration generated from the established interest’s fields. Therefore, by exploiting interest and gradients, paths are generated between sink and sources. Many paths can be created in order to choose one by reinforcement. The sink re-transmits the original interest message through the chosen path with a minimal interval therefore reinforcing the source node to relay data more often (Raghavendra et al. 2004).
In addition, path repairs are achievable in Directed Diffusion. When a certain path fails, a new or alternative path should be created to connect the source and the sink. Lee (2004) asserts that multiple paths should be set in advance so that one of the alternative paths is selected without incurring searching costs. However, there is extra overhead of maintaining many paths by using low data rate, since a lot of energy is required. Directed diffusion is different from SPIN because it is characterized by on demand data querying strategy. Thus, the advantages of Directed Diffusion include on demand route setup which results to energy efficiency and it does not require a node addressing mechanism since all communication is neighbor-to-neighbor. However, Directed Diffusion cannot be applied in continuous data delivery since it is query driven. Consequently, the naming schemes incorporated in Directed Diffusion depend on applications and they should be explicitly defined every time.
Rumor Routing
Rumor routing is a variant of Directed Diffusion and is mostly projected for cases in which geographic routing strategy are not appropriate. In real sense, Directed Diffusion disseminates the query in the whole network when there is lack of geographic mechanism to diffuse responsibilities. However, in some instances the data requested from the nodes is minimal and therefore the flooding concept is not necessary. A different approach may be employed, in which the events are flooded when the number of events is small and there exists a significant number of queries. Rumor routing act in event and query flooding. The notion is to route the query to nodes that are experiencing a certain event as opposed to flooding the whole network to acquire information on the current events (Ramesh et al. 2006).
The rumor routing protocol incorporates old packets, called agents to flood events within the network. Once a node determines an event, it appends it to its local table and develops an agent. Agents relay in the network in order to transmit information regarding local events to remote nodes. After a query for an event is generated by a node, the specific nodes that are familiar with the route can react to the query by using its event table. Therefore, the cost of communicating to the entire network is generally reduced. Only one path is used for transmission of data between source and destination as compared to Directed Diffusion in which data is sent through many paths at low rates. According to Akkaya and Younis (2003), rumor routing has the capability of ensuring that energy is saved and can as well deal with failure of nodes. Though, rumor routing protocol functions well when there is minimal number of events. If the number of events is large, the cost of sustaining agents and event table in every node may be difficult to pay off if there is less interest on the events from the sink.
CADR – Constrained Anisotropic Diffusion Routing
This is a data-centric protocol that aims to be a general type of Directed Diffusion. In this mechanism, two methods are proposed: “information-driven sensor querying (IDSQ) and CADR” (Akkaya & Younis 2003 p.333). The notion is to query the sensor nodes and route data packets so as to exploit the information gain, whilst reducing the bandwidth. This is attained by triggering only the sensors that are near a certain event and dynamically changing the data routes. The main distinction from Directed Diffusion is the concern of information gain and transmission cost. In CADR protocol, every node analyses an information/cost aim and directs data derived from the local information/cost gradient and end-user needs. In IDSQ, the querying node is able to establish which node is capable to give the essential information whilst conserving the energy cost. Moreover, IDSQ is deemed to be a complementary optimization technique since it does not explicitly define how the query and information are directed among sensors and the sink. Because of its ability to diffuse queries using a set of information mechanism, the CADR is more efficient than Directed Diffusion in regard to energy conservation (Lee 2004).
Hierarchical Protocols
Most communication networks take into account the scalability of the network, thus wireless sensor networks are not left behind. A network which is single-tiered can make the gateway to experience data overload due to the increased sensors density. This problem may cause delays in communication and poor tracking of events. Consequently, single-tier architecture does not give an option of increasing the number of sensors in the network because of inability to cope with increased load. In order to enable sensor networks to experience the concept of scalability, networking clustering has been followed in some routing protocols (Zhao & Guibas 2004).
Hierarchical routing aims at controlling the use of energy by sensor nodes through the task of incorporating the nodes in communicating severally within a certain cluster and by enabling data aggregation and fusion in a view to minimize the number of messages relayed to the sink. The formation of a cluster is normally derived from the energy reserve of nodes and the nearness of the sensor to the cluster head (Akkaya & Younis 3003). Therefore, this section analyses the hierarchical routing protocols.
LEACH – Low-Energy Adaptive Clustering Hierarchy
LEACH is one of the major hierarchical routing protocols used in WSN. The LEACH algorithm creates clusters of sensor nodes in regard to the strength of signals acquired and use local cluster heads like routers to the sink. In this case, energy is saved due to the fact that transmissions are performed by the cluster heads fairly than all sensor nodes. The best possible number of cluster heads is projected to be five percent of the number of all nodes. The cluster performs data processing including data fusion and aggregation. They also alter dynamically over time so as to control the energy loss of nodes. This resolution is carried out by a sensor node selecting a random value between 0 and 1. The node is made a cluster head for the current transmission if the value is less than the following equation (Akkaya & Younis 3003 p. 335).
T (n) = { p / 1 – p * (r mod 1/p) if n € G, { 0 otherwise, p is the expected percentage of cluster heads, r is the present round, and G is the number of nodes which have not acted as cluster heads for the previous 1/p rounds.
LEACH experiences a high factor (4-8) in reduction of energy dissipation as compared to direct communication and the least transmission routing protocol. The nodes in the sensor networks expire randomly and lively clustering enhances the system operation. Another advantage of the LEACH algorithm is its distribution capability, which needs no global network knowledge. Yet, LEACH employs a single-hop transmission in which each node is able to relay directly to the cluster head and the sink. Thus, LEACH is not appropriate to networks implemented in large regions. Consequently, dynamic clustering results to extra overhead, which may reduce the gain in energy consumption (Ramesh et al. 2006).
PEGASIS and Hierarchical – PEGASIS: Power-Efficient GAthering in Sensor Information Systems
PEGASIS is a hierarchical clustering protocol that improves the capability of the LEACH protocol. PEGASIS creates chains rather than multiple clusters. Each node relays and obtains from a neighbor and only a single node is chosen from the chain to act as a transmitter to the sink. Collected data packets move from one node to another, they are aggregated and finally transmitted to the sink (base station). According to Zhao and Guibas (2004), the chain is constructed through the incorporation of a greedy algorithm. PEGASIS is more efficient than the LEACH because it eliminates the overhead that is as a result of dynamic clusters in LEACH algorithm and by the method of transmission and reception reduction by employing data aggregation. Though, PEGASIS creates excessive delay for remote node on the chain. Furthermore, the single chain selected may fail.
Hierarchical-PEGASIS is an improvement to PEGASIS. Its main objective is to minimize the delay acquired for packets in the process of communicating to the sink and suggests a solution to the data collection problem by taking into account the energy and delay metric. Concurrent transmissions of data are aimed at, so as to reduce delays. To prevent collisions and signal obstruction among the sensor nodes, two techniques have been researched. The first technique involves signal coding, i.e. CDMA. The other technique enables distant nodes to transmit simultaneously (Rousselot 2006). Even though the PEGASIS algorithms eliminate the clustering overhead experienced in the LEACH algorithm, they also need dynamic topology modification because energy of the sensor is not kept.
TEEN and APTEEN
Threshold sensitive Energy Efficient sensor Network protocol (TEEN) is specifically developed to deal with abrupt alterations in the sensed parameters like sound. TEEN aims at employing a hierarchical algorithm and the data-centric protocol so as to increase responsiveness. The WSN structure is designed according to the hierarchical ordering in which closer nodes create clusters and the mechanism proceeds to the second level until the sink is found. The APTEEN (Adaptive TEEN) protocol extends the features of the TEEN algorithm. Its objective is to detect periodic data collections and to respond to time-critical events. APTEEN acts on three distinct query types: historical, one-time, and persistent. However, the main disadvantages of these mechanisms are the overhead and difficulty of creating clusters in several levels, and maintaining the attribute-based query identification (Etefia 2004).
Location-Based Protocols
Most routing protocols used in the WSN need site information for sensor nodes. This information is required mainly to estimate the distance between two sensor nodes in order to aggregate the energy being consumed. Location information can be used in a more reliable method because sensor networks do not have a classical addressing scheme and they are distantly located in a certain area. Thus, if the location information of a particular sensor is known, a query can be dispersed in that region only, so as to reduce the number of transmissions (Raghavendra et al. 2004). Therefore, this section looks at some of the location-based protocols.
MECN and SMECN
Minimum energy communication network (MECN) protocol creates and sustains a minimum energy networks for WSN by exploiting minimum power GPS. MECN presumes a master-location as the information sink, which is normally applied in WSN. This protocol determines a relay region for each sensor node. The identified relay region contains nodes in the neighboring region in which transmission that takes place within the nodes in the area is more energy efficient as compared to direct transmission. Determining the efficient paths is performed by the means of a local search for each node, taking into consideration its relay region (Akkaya & Younis 2003). Furthermore, MECN can be able to reconfigure itself and therefore it can be able to withstand node failure or implementation of new nodes.
The Small MECN (SMECN) extends the functioning of MECN. In MECN, it is presumed that each sensor node can be able to relay to every other node, which is generally not possible in most cases. The SMECN considers other factors that affect the transmission of data packets. Though, the network is seen as fully connected like the MECN. The relay region created by SMECN is generally smaller than the one developed by MECN. Consequently, the number of propagations for relays will decrease. This implies that SMECN uses less energy than MECN and the cost of sustaining the network is low. Though, constructing a smaller network with few nodes presents more overhead in the protocol (Akkaya & Younis 2003).
GAF – Geographic Adaptive Fidelity
This is a location-based protocol that is mainly used in mobile informal networks, but can be used in WSN also. GAF turns off the inactive nodes within the network without interfering with the network functionality. It then creates a virtual grid for the region in use. Every node relates itself with a point in the virtual grid by the use of GPS-indicated site. Nodes residing in the same position are regarded as having similar cost of packet transmission. This similarity is dealt with by ensuring that some nodes are inactive so as to save energy. Therefore, this protocol is deemed to enhance the lifetime of the network when the number of nodes increases.
In addition, nodes in the network change their states from active to sleeping in order to maintain a balanced network operation. According to Zhao and Guibas (2004), GAF defines three states for the nodes: discovery, for establishing the neighboring nodes in the grid, sleep, when the node is inactive, and active, when the node is involved in transmission. Despite its categorization as a location-based protocol, GAF can be regarded as a hierarchical algorithm. Thus, for every grid area, a representative node is considered a leader to relay data to other sensor nodes (Ramesh et al. 2006).
GEAR – Geographic and Energy-Aware Routing
GEAR takes into consideration the energy aware and environmental aware neighbor choice heuristics to transmit a packet to a destined region. The notion is to minimize the number of interests in Directed Diffusion through taking into account a specific area instead of sending the interests to the entire network. Thus, energy is conserved since GEAR enhances the capability of Directed Diffusion. Furthermore, GEAR ensures that every node is able to maintain two types of costs. Firstly, the estimated cost of reaching the sink through the neighboring nodes. This is an integration of outstanding energy and distance to destination. Secondly, the learning cost takes into consideration the transmission around holes (no closer neighbors) in the network. In essence, GEAR consists of two stages: transmitting packets towards the target area and relaying packets within the site (Etefia 2004).
Conclusion and Research Issues
Many researchers have continued to take interest in routing protocols in the recent years. This has been able to generate new challenges as compared to the conventional guided networks. This paper has compared and evaluated the recent research of routing protocols in wireless sensor networks. The protocols were categorized into three main groups, that is, data-centric, hierarchical, and location based. Data-centric protocols are algorithms that name the data and query the sensor nodes on some parameters of packet data. Many researchers follow this mechanism since clusters and specialized nodes are not required as compared to hierarchical and protocol-based networks. Though, the naming schemes used in data-centric protocols may not be suitable for compound queries and they normally rely on applications. A more reliable scheme is the most burning issue that requires research in this category of routing.
Alternatively, the hierarchical protocols categorize sensor nodes to reliably route the sensed data to the destination. The cluster heads are in other times selected as dedicated nodes that are less limited to energy. A cluster-head analyses the data and transmits it to the destination on behalf of the other nodes that reside within the group. The main research issue in hierarchical protocol is how to create clusters that optimize the energy consumption and modern communication metrics. Consequently, location-based protocols use the location information of a sensor node to create a communication region. The major research issue that needs consideration in this category is the proper use of location information so as to help in energy efficient transmission. Though, the performance of these protocols is better in terms of energy conservation, there is need to create protocols that address issues like quality of service (QoS). Another major issue in routing protocols is the accounting of node mobility. Most of the present protocols presume that the sensor nodes are stationary, thus there is need to consider mobility in complex networks. Furthermore, WSN requires further research so as to create routing protocols that are able to combine guided network with wired networks.
References
Akkaya, K & Younis, M 2003, A survey on routing protocols for wireless sensor networks. Web.
Etefia, B 2004, Routing Protocols for Wireless Sensor Networks, Web.
Lee, W 2004, Overview of Sensor Network Routing Protocols, Web.
Raghavendra et al. 2004, Wireless Sensor Networks, Springer, New York.
Ramesh et al. 2006, ‘Data-Aggregation Techniques In Sensor networks: A Survey’, IEE Communications Surveys, vol. 8, no. 4, 2006.
Rousselot, J 2006, Dynamic Networks: Routing Protocols for Wireless Sensor Networks, Web.
Zhao, F & Guibas L.J 2004, Wireless Sensor Networks, An Information Processing Approach, Morgan Kaufmann Publishers, San Francisco.