Cross Layer Design And Wsn Optimization

Published Date: 02 Nov 2017

Introduction

1.1 Background

Recent advances in the area of wireless communications and embedded systems have enabled the development of small-sized, low-cost, low-power, multi-functional sensor nodes that can communicate over short distances wirelessly [1], [2]. These sensor nodes represent a significant improvement over traditional sensors, since these sensor nodes perform processing and communication functions in addition to the traditional sensing function. The processing and communication functions embedded in the sensor nodes essentially allow networking of these nodes, which in turn can facilitate sensing function to be carried out in remote/hostile areas. A network of sensor nodes (often referred to as a wireless sensor network) can be formed by densely deploying a large number of sensor nodes in a given sensing area, from where the sensed data from the various sensor nodes need to be transported to a monitoring station often located far away from the sensing area. The transport of data from a source node to the monitoring station can be carried out on a multihop basis, where other intermediate sensor nodes act as relay nodes. Thus, each sensor node, in addition to behaving as a source node, often needs to act as relay a node for data from other nodes in the network.

The wireless sensor nodes are powered by finite-energy. Being deployed in remote/hostile sensing areas, these batteries are not easily replaced or recharged. Thus, end of battery life in a node essentially means end of the node life, which in turn can result in a change network topology or in the end of network life itself. Thus, the network lifetime shows strong dependence on battery lifetime. Efficient use of battery energy is hence crucial to enhance the network lifetime. Enhancement or exploration of energy efficient techniques for increasing network lifetime has been an active area of research.

The proposed research work has been oriented towards lifetime maximization and overall network optimization by implementing the elephant swarm based cross layered architecture.

Enhancing the lifetime of wireless sensor networks had baffled researchers for quite some time now. The behavior of large elephant swarms motivated us to incorporate their behavior into wireless sensor networks. The complex elephant swarm behavior is incorporated using a cross layer approach. The elephant optimization discussed in this paper enables optimized routing techniques, adaptive radio link optimization and balanced scheduling to achieve a cumulative enhanced network performance. The proposed elephant swarm optimization is compared with the popular protocol and the Particle Swarm Optimization () Protocol.

Let us consider a topology of wireless sensor networks deployed over a specified geographical region. The sensor nodes are assumed to have homogenous energy properties and are battery operated which is the case most often than not. The sensor distribution over the geographical region is considered to be dense to achieve higher transmission data rates. Owing to dense deployments numerous links are established induce interference amongst the sensor nodes which needs to be minimized to achieve better network performance in terms of throughput. This paper introduces an Elephant Based Swarm Optimization model to enhance network life time. A cross layer approach is adopted to incorporate the elephant swarm optimization features.

Elephants are social animals [3] and exhibit advanced intelligence [4]. Elephants are often found to exist in a "fluid fission-fusion" social environment [5]. Elephants characterized by their good memory, their nature to coexist and survive within a "clan" [6] (a large swarm of more than 1000 elephants) socially formulated during testing times like migration and when the resources are scare. Elephants exhibit an unselfish behavior which enable them to grow and is the secret of their longevity. Keeping progress and survivability in mind the older elephants disassociate from the "clan". Elephants by nature are protective of their younger generation. Elephants communicate using varied advanced techniques which include acoustic communication, chemical communication, visual communication and tactile communication [7] [8]. Their memory empowers them with recognition, identification and problem solving scenarios [6]. All these features exhibited have influenced the authors to incorporate such behavior in wireless sensor networks to improve network performance.

In wireless communication system the energy and hence the nodal life time plays a vital role in assuring the quality of network and its sustainability in field. Therefore the enhancement of the system architecture for increasing the life time becomes a very important issue for research society. A number of researches have been done for optimizing the lifetime of nodes in WSNs. Now days a number of algorithms and protocols have been made based on evolutionary computing approach.

The elephant swarm model is complex and to realize such behaviors in wireless sensor networks the authors have proposed to adopt a cross layer approach to incorporate the elephant swarm model. Optimizations need to be adopted at the Routing Layer, Layer and the Radio Layer of the wireless sensor node. This paper introduces a cross layer approach to incorporate the elephant swarm optimization technique which is compared with the popular, protocol and the efficiency is proved in the experimental study discussed later in this manuscript.

1.1.1 Wireless Sensor Network:

Wireless sensor networks (WSNs) usually contain thousands or millions of sensors, which are randomly and widely deployed. Wireless sensor networks (WSNs) generally consist of a large number of low-cost and low-power sensor nodes that are small in size and able to be employed in a wide range of applications such as in the military, environmental sensing and habitat monitoring [9]. Sensors are powered by battery, which cannot rechargeable after deployment. But sensor networks are designed to last. Sensor nodes collaborate to be able to cope with the environment: they operate completely wirelessly, and are able to spontaneously create a network, assemble the network themselves, dynamically adapt to device failure and degradation, manage movement of sensor nodes, and react to changes in task and network requirements as shown in Fig. 1. Thus, energy efficiency is an important issue in sensor networks. Since routing consumes a lot of energy, an efficient routing scheme in sensor networks is also important.

Given that the sensor nodes are usually irreplaceable, the network protocol for WSNs should be designed so that important performance parameters are optimized, such as extended network lifetime, energy consumption, and data throughput. Clustering is one of the design methods that have been proven to result in significant improvements in wireless sensor networks by managing the network energy consumption efficiently [10]-[11]. The application of the clustering-based approach has the advantage of minimizing the number of nodes that take part in long distance communication with the base station through the utilization of cluster head nodes and, consequently, reduce the energy consumption of the network.

WSN is a network of spatially dispersed sensor nodes prepared with sensing, compute, authority, and communication module to monitor a convinced occurrence such as environmental data or object pathway [12]. In present situation, the amount of sensor node may be twenty to thirty but in prospect it might consist of nth authority additional sensor as well as actuator systems [13].

The situation of the sensor nodes might not be pre-determined and could need sensor nodes to be equipped through self systematize protocols [14]. Usually, sensor nodes scrutinize and intelligence the occurrence with a sensing module, procedure the data with a compute module, and send the data to a necessary purpose over a radio crossing point with a communication module.

Figure 1.1: Illustrative sensor network architecture

Illustrative sensor network architecture is shown in Figure 1.1 in which sensor nodes are distributed over a particular area of interest to collect data, process them, and send them to a sink node for further processing.

1.1.2 Cross Layer Design and WSN Optimization

It is argued that layered architectures are well suited for wired communication but they do not perform well in wireless networks [15]. In the following sections, an overview of traditional layered architectures is presented and argued why they are not suited for wireless communication networks.

1.1.2.1 Layered Approach

The Open Systems Interconnection (OSI) reference model [16] divides the network architecture into seven well defined logical layers, each layer responsible for some specific task [16]. The real world implementations of the layered approach including TCP/IP protocol show the importance of layered architectures. The corresponding architectures are shown in Figure 1-2. Layer-wise functionalities discussed in [17] and the needs to divert from traditional architectures are described as follows:

Physical layer is used to transmit raw bits over wired or wireless channel [17]. It is composed of different hardware modules, for example, a radio in WSNs. Radio is a gateway of sensor node to the external world, and is the main source of energy utilization. There are several factors which effect the power consumption on the physical layer including modulation scheme, data rate, transmit power, and different modes of operation. In traditional systems, like wireless local area networks, power is not a major issue but it is one of the basic limiting factor in the wide spread applicability of WSN applications; therefore, the physical layer needs to be re-considered in the WSN context.

Link layer is composed of medium access and logical link control functionalities [17]. In context of WSNs, at link layer, there are different sources of energy wastage comprising collision, overhearing, control packet overhead, and idle listening [18]. Such sources of energy wastage do not pose problems in wired networks because of unlimited power supply. These issues require reconsidering the already existing layered protocol architectures.

Main functionalities of network layer include routing of information, topology control, best route determination, and network layer addressing [17].

Routing in low power WSNs has different characteristics as compared to traditional routing and wireless ad-hoc networks [19]. These characteristics discussed in [19] include: firstly, global addressing and hence classical IP-based routing is not possible because of sheer number of sensor nodes. Even if the number of nodes is not very high, the nodes normally have to know their positions and utilizing position information for routing decisions reduce control packet overhead. Secondly, in most cases, data are sent from different regions towards a sink node while in traditional systems, for example, in wireless ad-hoc networks, the source destination pair may change constantly. And thirdly, presence of redundant data which need to be filtered or aggregated along their path towards the sink node. These issues motivate to divert from traditional architectures.

Figure 1.2: Layered architectures

Transport layer functionalities include end-to-end data delivery, acknowledged and unacknowledged services, and flow control [17]. Transport layer is required if the system has to talk to the Internet or any other communication network [14]. The argument presented is that: as most of the communication is done hop by hop in WSNs, and there is no notion of end-to-end delivery, transport layer may not be required. But for low power sensor networks where encryption algorithms cannot be used for complexity reasons, authentication server services may be implemented for security reasons.

Application Layer contains different protocols required by the end user [17]. WSNs are highly application specific1 and require reconsideration in protocol architecture.

The Mac layer can be altered or modified with higher data bit so as to enhance the system throughput and of course the network lifetime.

1.1.3 Cross Layer Approach

Cross layer design may be defined as, "the breaking of OSI hierarchical layers in communication networks" [20] or "protocol design by the violation of reference layered communication architecture is cross-layer design with respect to the particular layered architecture" [15]. The breaking of OSI hierarchical layers or the violation of reference architecture includes merging of layers, creation of new interfaces, or providing additional interdependencies between any two layers as shown in Figure 1-3.

Figure 1.3: Example reference architecture with defined interfaces (Fig a.) and its violation (Fig. b)

For resource constrained systems, such as WSNs, optimization has to be done across all layers to obtain lifetimes of years [21] and this optimization can be achieved by exchanging information across layers. The unique problems and opportunistic exploitation of wireless links, and the new modalities offered by wireless communication paradigm make a strong case for cross layer design and optimization [15]. An example of a unique problem is that; the TCP in layered architectures implicitly assumes that a packet loss is caused due to collision which is not true for ad-hoc wireless networks where a packet loss may occur because of other phenomenon like fading or varying link quality [22]. Potentially harsh environmental conditions, unattended operation, and operating in free frequency band make WSNs even more prone to errors by interference or fading. An example of opportunistic exploitation is increasing the transmission rate when the channel quality is good [23]. This exploitation would mainly depend on the application (if it requires high transmit rate) and many challenges are associated with it, for example, how to compute the link quality so that the transmit rate can be decreased or increased. But still, it is considered as one of the motivations to apply cross layer approach. An example of new modality in wireless communication systems is to use a wake-up radio [24] with a main radio to reduce the duty cycle of the main radio and hence save energy [25]. All these issues provide a basis to explore cross layer approach for WSNs.

1.1.3.1 Cross Layer Design

Resource constraints in WSN node require energy efficient and energy aware schemes on all layers of the protocol stack to increase the lifetime of the network [19]. The traditional layered networking approach has several drawbacks from WSNs perspective, improvements in performance and energy efficiency are possible if significant amount of information is passed across protocol layers and hence network lifetime can be improved [26].

1.1.3.1.1 Introduction

Cross layer design may be defined as, "the breaking1 of OSI hierarchical layers in communication networks" [20] or "Protocol design by the violation of reference layered communication architecture is cross-layer design with respect to the particular layered architecture" [15]. The protocols relying on interaction between various layers of the protocol stack can generally be termed as cross layer design [27]. The breaking of OSI hierarchical layers, the interaction between various layers or the violation of reference architecture include merging of layers, creation of new interfaces, or providing additional interdependencies between any two layers and are discussed in [15]. It is required that optimization has to be done across all layers to obtain lifetimes of years [21]. Consider protocol architecture with four layers as shown in Figure 2-5.

Figure 1.3: Illustrative reference architecture

Assume that each layer has defined interfaces for communication with any other layers. For example, in Figure 1.3, layer 3 can only communicate with adjacent layers (layer 4 and layer 2) via defined interfaces and it cannot communicate with layer 1 as no such interface is available for information exchange. It should be noted here that it is only example-layered-architecture and has nothing to do with well known protocol architectures like TCP/IP protocol suit or LonTalk protocol [13].

According to aforementioned definitions of cross layer design, the violation of this architecture would lead to a cross layer design as shown in Figure 1.4 which is also outlined in [15]. In Figure 1.4 (a), two new interfaces (encircled in the figure) are created at layer 3 for information flow from layer 4 to layer 3 and layer 2 to layer 3. Figure 1.4 (b) is another example of cross layer design where firstly, layer 2 and layer 1 are merged to result in a super layer and secondly, the design of layer 3 is dependent on the design of layer 4 (layer 3 → layer 4) which means that any change in layer 4 would result in changes in layer 3 as well. Figure 1.4 (c) shows violation of reference architecture by introducing another "vertical layer", which is used for vertical calibration and fine tuning of parameters of one layer on the basis of feedback from any other layer. The cross layer design may include the mentioned violations of the referenced architecture in one form or the other.

Figure 1.4: Cross layer design of the reference architecture

1.1.3.1.2 Significance of Cross Layered Approach

The simplicity of design of a layered protocol stack with static interfaces between independent layers resulted in development of robust and scalable protocols for the Internet but performs poorly for wireless ad-hoc networks [28]. The inter-dependencies between different layers can be utilized to get statistically optimal performances for different network parameters like energy efficiency or end to-end delay. In the following paragraphs, examples which show how to enhance the overall performance with cross layer design are outlined.

Location of the nodes can be used to define area dominating set [29], so that group of nodes can go to sleep to save energy. For example, assume a wireless sensor network is deployed to monitor humidity in a particular region. As the nodes know their own positions and also the position of the neighbor nodes, some of the nodes may go to sleep state (low power) by considering two things; firstly the network connectivity should be maintained, and secondly, the area to which those particular nodes belong should be represented by other nodes (e.g., neighbors). The location can also be used in geographic aware routing [30]

The packet length can effect output power and bit error rate [30]. Short packet sizes results in inefficient energy usage because of large overheads while long packet sizes may experience higher number of errors, so energy efficiency can be maximized by optimal packet size [31]. The modulation at physical layer can be changed depending upon the remaining capacity of the battery [30]. The number of packets in the system (in buffer or queue or being in transmission) can affect the constellation size of the modulation scheme [32].

The lifetime of the network can be extended by using varying data rate at each node in the routing path. Reducing transmission rates at critical node (energy constrained) also results in extended network lifetime [33]. If data rate is increased, the probability of encountering errors also increases, so a higher value of SNR (Signal to Noise Ratio) would be required at the transmitting end to have an acceptable value of BER (Bit Error Rate) at the receiving end. Higher SNR means higher transmitting power [34] and therefore, more energy consumption. Based on the data rate requirements, modulation scheme can be selected [34]. The modulation scheme, with certain BER threshold values and SNR can be used to calculate the transmit power [34]. The optimal transmit power increases with increase in the data rate (vulnerable time is decreased but thermal noise is also increased) while a carefully chosen data rate can have high impact on transmit power and network lifetime [35].

The benefits of using cross layer information indicate that cross layer design can be used to enhance the performance of WSNs.

1.1.3.2 Energy Consumption and Cross Layer Design

Routing of data packets and exchange of control information consumes significant amount of energy [28]. Routing protocols can minimize energy consumption by adjustable transmit power approach, load distribution approach and duty cycling [36]. Transmit power and load distribution approaches attempt to minimize energy when the network is in active state (transmitting or receiving data) while duty cycling (sleep-power down cycles) maximizes energy performance when in inactive state. Adaptive routing protocols utilizing cross layer information and based on link status, congestion and traffic conditions can be used to minimize energy consumption [28]. The following strategies can be adopted to decrease energy utilization at the routing layer.

Minimize flooding, which will consequently minimize the number of transmissions.

• Decrease in the number of control packets, e.g., Table less position based Routing (TPR)

[5.3. p. 77] does not exchange any control data.

• Using energy aware metrics in path selection as is done in Energy Aware Distance Vector Routing (EADV) [37]. Use of cross layer information for instance, remaining battery capacity to determine routing metric [38] or duty is cycling as in TPR [39].

• Using adaptive transmit power to select next hop neighbors as is done in TPR [39].

• Data aggregation or data fusion can decrease in the amount of redundant data and which would reduce the number of transmission and as well as energy consumption [36]. Similar approach is followed in grid based routing.

In short, the design of a routing protocol can play an important role in maximizing the performance of energy consumption in WSNs. It should be noted that there is always a tradeoff between minimizing energy consumption at the routing layer and maximizing other quality of service parameters like delay or throughput. For example, assume one wants to decrease end-to-end delay between two sensor nodes. Now to achieve this, the packets have to be routed through a path which has either less congestion or fewer hops towards the destination node. Continuously sending information on that particular path will drain energy of the intermediate nodes on that path and some of the nodes may die earlier resulting in reduced network lifetime which otherwise could have been extended. But if reduced energy consumption is kept as a primary motive, then the information may follow paths which always result in higher delays or lower throughputs. This trade off totally depends on specific application and would require due consideration at the design time.

Four major sources of energy wastage in MAC schemes [18] are as follows. When two or more WSN nodes within each other’s transmission range transmit at the same time, collision occurs. Collision wastes energy because the collided packet becomes corrupt and requires re-transmission. Idle listening is another source of energy consumption, where a node listens to the channel hoping to receive some data. It can result in high energy wastage in event driven applications, where nodes are awake to relay data while no event occurs. Overhearing is third source of energy wastage identified in [18] where nodes receives and analyze packets that have a different destination address.

WSNs with high network degrees and heavy traffic are mostly affected by this phenomenon. The last source of energy wastage identified is control packet overhead. The following strategies can be adopted to decrease energy utilization at MAC sub-layer.

• To turn on and off the radio to avoid idle listening and overhearing as is done in S-MAC [18] or CSMA-MPS [37].

• The use of a low power wake-up radio [24] to wake-up the main radio can save significant amount of energy1.

• Cross layering MAC and physical layer and using adaptive transmit power can result in improved energy efficiency, delay, and reception rate [40].

• Decrease in control packet overhead.

• Reduce idle listening as in timeout MAC [41] by adaptive duty cycling instead of fixed duty cycling [18].

• Adaptive rate scheme with CSMA and backing off in application layer rather than MAC layer can be used to achieve fairness in an energy efficient way [42].

In short, to support low duty cycle operations, MAC scheme has to be cross designed with the application layer [43]. Also the RTS-CTS mechanism of the MAC schemes can be coupled with the routing protocols which also require RTS-CTS packets, for example, to find out the neighbor with lowest cost towards the sink node among many numbers as is done.

1.1.4 Medium Access Control (MAC) Strategy:

Medium Access Control (MAC) protocols for WSNs are especially challenging because the receiver of a sensor node in a distributed multi-hop network should be turned off most of the time to save energy. However, to be contacted by a neighboring node, the receiver must be turned on and it becomes difficult for the nodes to discover and synchronize because of low duty cycle, limited active time [44][63], and clock drifts of the participating nodes [18]. Existing solutions try to solve this problem by different means of synchronization using methods such as preamble sampling or periodic synchronization and trading off delay for power consumption.

As discussed above, to achieve low duty cycle operation the sensor node has to operate in high and low power modes, for example, active and sleep mode. Assume that a sender wants to transmit data to a receiver as shown in Figure 1.5.

Figure 1.5: Duty cycling in medium access control schemes

For a receiver to receive data, it must be in active/listen state, as in sleep state, the radio is in low power mode with the receiving circuitry switched off. If the receiver operates at 100% duty cycle, which would mean that its transceiver is always on, then it would be able to receive the data at the cost of high energy consumption. To reduce the power consumption low duty cycle operations are applied. The lower is the duty cycle, the difficult it is to synchronize with the neighbor nodes at the time of transmission as there is no global clock available in most of WSN applications, and the clock drifts in local clocks can be such that it would make the synchronization expensive. Although low duty cycles, reduces the power consumption, the sender has to send the preamble until it is hit by the receivers windows. Though techniques like CSMA-MPS [37] have reduced the preamble length, the energy performance can be further enhanced if the need to synchronize is avoided in first place. Here, a wake-up radio, in addition to the main radio can play its role. The wake-radio works at 100% duty cycle and consumes energy in the range of 50 to 100 μW.

1.1.5 Physical Layer and Energy optimization

A few of the major hurdles in the wide spread utilization of WSNs comprise cost and energy consumption, and the design of the physical layer is dominated by both these factors [44]. The physical layer has to meet the design requirements keeping in view the nature of WSNs. The modulation schemes, the use of a certain frequency band, and the coding techniques have significant impact on the cost and battery lifetime requirements [44]. The requirements and the constraints imposed on the design of the physical layer for WSNs are discussed in [Dan04] and summarized as: in general, the radio must be small, the price of the radio must be cheap as the number of sensor nodes may be too high, and it must be able to work with the higher layers to reduce the power consumption levels.

The important parameters at physical layer which can be used as cross layer information to improve energy efficiency include modulation scheme, transmit power, and transmission range [46]. In [47], the authors have shown that, if the higher layer protocols and algorithms are developed irrespective of the knowledge of the physical layer and if physical layer is considered as a black box, then such developments may lead to inefficient solutions. Increase in the number of bits per symbol, will reduce the transmit-on-time and consequently the energy consumption [48].

The power consumed by the radio while receiving or transmitting is one of the main power consuming components on the physical layer. Current radios can have different modes of operation with different power consumption levels, for example, CC25201 transceiver can have different power modes of operation and each mode has its own power consumption profile depending upon which particular components are running in that particular mode. As the power consumed in the active state by the transceiver is much greater than the power consumed in the low power state (e.g., sleep state), so it is always desirable to keep the duty cycle low to have extended network lifetime [49]. In low duty cycle operations, batteries always get an opportunity to undergo recovery effect which is beneficial for extended network lifetime. Bursty communication at physical layer can help attain low duty cycle and consequently extended network lifetime [44]. Many of the routing schemes depends on increasing or decreasing the transmit power. All these issues indicate that physical layer is involved in the cross layer design to much extent and can have significant impact on the energy performance.

1.1.6 Energy Efficient Routing Protocol:

Routing Protocols

A plethora of routing protocols has been developed for WSNs and can be studied in detail in survey papers [50] and [19]. One reason for such a huge number of protocols for WSNs is that such networks are application specific and a particular routing protocol can only satisfy a class of WSN application requirements. [50] Classified routing protocols into data centric routing, hierarchical routing, and location based routing, and quality of service (QoS) aware routing. Any of the routing protocol is either reactive (routes to destination are computed on demand) or proactive (route to destination is pre-determined) [51]. In the following sections, the routing protocols are discussed in their respective categories.

1.1.6.1 Proactive and Reactive

Proactive routing protocols attempt to maintain routing tables from each node to every other node in the network and topology-changes are propagated to ensure updated routing information [51]. Destination-Sequenced Distance Vector (DSDV) [52] is a proactive routing protocol which uses Bellman Ford algorithm to compute shortest paths and ensures loop-free routing tables. Every node maintains a routing table composed of entries for every other node with associated next hop and cost to that node. There are several drawbacks if these protocols are used for WSNs. Firstly, they maintain routing information even for unused links which results in unnecessary bandwidth utilization when topology changes frequently [53]. Secondly, if hundreds of thousands of sensor nodes are used in a particular application, the size of the routing tables will grow exponentially resulting in processing and memory overhead.

To overcome issues associated with proactive routing protocols, reactive routing protocols which compute routes on per need basis were developed [53]. Dynamic Source Routing (DSR) [54] and Ad-hoc On Demand Distance Vector (AODV) [55] are two examples of reactive protocols. AODV [55] is based on DSDV but it is reactive. If a source S wants to send data to destination D, S will broadcast a route request if S does not have any information about D in the routing table. The request is further broadcasted unless some intermediate node, which already has route information to D or the D itself, replies. In the process of the broadcast, every intermediate node establishes a reverse path to the neighbors from which the first copy of the broadcast is received. In DSR [54], S specifies the complete route to be taken by the packet towards D unlike AODV in which only next hop is specified. If S does not know about the complete path, it will broadcast a route request which is replied in the same manner as in AODV but with complete route path information. The entire path is then added to the header of the packet destined for D and hence it is called source routing. Reactive protocols introduce additional delays for the first packets from any source S to destination D as the routes are not available in advance [53]. If the topology-changes are frequent, then there are chances of packet losses when a packet is on its way towards the destination with complete route information and some intermediate nodes, part of the route, have already disappeared.

1.1.6.2 Data-Centric

According to [50, p. 5] IP-like global addressing scheme is difficult in sensor networks because of the sheer number of nodes. Therefore, attribute based naming is required to specify the data and that naming is then used to query the network. For example, if a sink node wants to know about humidity in a particular area, it can broadcast query mentioning "humidity > 40%". Sensor nodes with humidity greater than 40% can send their data back to the sink node. In this case, a particular node is not "addressed" by its ID but queried on the basis of certain "data threshold" or it can be queried based on its location information hence resulting in data-centric routing as opposed to address-centric routing. In the following sections, different data centric routing protocols are discussed briefly.

Flooding

Flooding is a classical way to propagate information by broadcasting it. But if it is done blindly, it always results in the broadcast storm problem (redundancy, contention, and collision) as discussed in [56]. The problems associated with flooding are implosion (duplicate messages sent to same node), overlap (e.g., two sensor nodes close to each other and report temperature of that area) and resource blindness (not considering resources of node, e.g., remaining energy). For wireless container management system (WCMS) where node energy is a scarce resource, flooding will prove energy inefficient if the number of nodes is high and number of messages per unit time is also high. Detailed analysis of flooding is discussed.

Gossiping

Gossiping is enhanced version of flooding which overcome the problem of implosion by selecting neighbors randomly [14]. The problem of implosion is avoided at the cost of introducing delays and more resource utilization as shown in Figure 1.6. Assume node 1 wants to send a packet to sink node. In Figure 1.6(a) (dotted lines with integers showing the sequence of packet flow), node 1 selected node 3 randomly and sent a packet to node 3 to relay it to the sink node, therefore, node 2 avoided implosion. In Figure 1.6 (b), for the same scenario, node 1 selected node 2 as a random node. Node 2 has only one path to the sink node and that is via node 1. So the packet is again sent to node 1, which in turn is sent to node 3 by node 1 and finally reaches the sink node hence resulting in unnecessary resource utilization at node 1 and node 2 and incurring delays when the packet was sent to node 2 initially. Although it is improvement over flooding, it still may not be used for WCMS because of inefficient operation and in some cases unprecedented forwarding delays.

Figure 1.6 : Gossiping

Sensor Protocol for Information via Negotiation

Sensor Protocol for Information via Negotiation (SPIN) [57] is an efficient data dissemination scheme for energy constrained sensor networks. The data of sensor nodes are named by descriptors known as meta-data to avoid redundancy of data in the network but SPIN does not specify this metadata. Each node in a sensor network is considered as a sink node. To overcome the problem of implosion and overlap, negotiation based dissemination is introduced. To overcome the problem of resource blindness, every node polls its resources with the help of resource manager to other nodes before actual data transmission

Figure 1.7: Sensor Protocol for Information via Negotiation (SPIN)

The operation of the protocol is shown in Figure 1.7 and explained as follows. Whenever new data are available to a sensor node A, it advertises these data to its neighbors with the help of ADV message which also contains the meta-data. Any node which is interested in these data (assume node B) sends back a REQ message to the node A. Upon receiving request for new data, node A sends a DATA message to node B. The received data are then advertised to other nodes (C and D) and the process is repeated. The problem with SPIN protocol is; assume that node E in Figure 1.7 is interested in data ‘X’ which is available with node B, and node E connects to node B with the help of intermediate nodes D and C. Further assume that nodes C and D are not interested in data ‘X’, then, there is no way that these data can be sent to node E. Although, it is claimed to be energy efficient, it cannot be applied to WCMS, because in WCMS, node may be queried by central controller, but they also need to send the information on periodic basis.

Directed Diffusion

Directed diffusion [58] is a new paradigm in data centric routing schemes. The main idea of directed diffusion is to aggregate data on their way to destination node to eliminate redundant data, reduce number of transmissions, and save energy. Data generated by sensor nodes are named by attribute-value pairs. The sink node broadcasts its interest to request some data. This interest is further broadcasted by the intermediate nodes until it is diffused in the network. Each node receiving interest caches it for future use, while it setups a gradient (data rate, duration and expiration time) towards the node from which interest was received. Once the criteria is matched (sources know that it can respond to the particular interest shown by sink node), sources start sending the data with low data rate as specified by the already established gradient. Once the sink starts receiving the required data, it requests the same interest from one particular neighbor again, but with different quality requirement (high data rate). The neighbor which already had cached that interest reinforces the gradient with the sink node and contacts a particular neighbor from the list of its neighbors and the process is repeated till the source is reached. In this way, a high data rate path is available to the source node which it can use to send the data to the sink node. The data on their way back to the sink node can be aggregated to reduce number of transmissions and redundant data.

Directed diffusion may not be applied to applications where data are required by the sink node on continuous basis. For example, in WCMS, if the locations of the containers are reported continuously, directed diffusion may not fit well. But the data aggregation concept can save energy by reducing the number of transmissions, and therefore, this concept is utilized in position base routing scheme.

Rumor Routing

Rumor Routing is a variation of directed diffusion. The basic idea of rumor routing is to flood events instead of flooding queries if the number of events is small compared to the number of queries. If a certain node detects some event, it stores it in its event table with a distance zero to that event. It then creates an agent (long lived packet), which is propagated in the network for some time. Every node receiving that agent store information of the event in its event table with distance (number of hop counts) to the node at which the event occurred. Any node which is interested in some query will either transmit query directly to the source node (if it has route information) or randomly transmit the query hoping that it will reach the source node. If a node that initiated the query realized that its query has not been answered, it can retransmit the query in the similar manner discussed above or use the flooding approach. In WCMS, though it is true that the number of events will be small as compared to the number of queries (although it depends on the railway track and on site conditions), and lets assume for a while that no data are sent on continuous basis, still the information is always sent to a sink node (the destination node is always known) and it will not be energy efficient to flood that event within the network.

Energy Aware Distance Vector Routing

Energy Aware Distance Vector (EADV) routing protocol is developed at Institute of Computer Technology for ultra low power WSNs. It is suitable for many to one communication paradigm. The basic operation of the protocol is as follows. Network initialization starts by sending Initial Broadcast Vector (IBV) by the sink node. Every node receiving the Initial Broadcast (IB) stores the IBV in its routing table and forwards the broadcast by replacing the source address with its own, incrementing the HC (hop count) by 1 and by updating the cost field. This process is called as "Initialize Broadcast Forward" (IBF).

Figure 1.8: EADV operation

Figure 1.8 shows IBV flood in the networks. The continuous lines represent those IBVs, which trigger an IBF at the receiving node. The dashed lines represent a transmission of the IBV without consequent IBF at the receiving node. As an example in Figure 1.8, if node 1 issues an IBF, nodes 0, 2, 4 and 5 receive it. The sink node (address = 0) issues no IBF since by default it forwards no messages, node 2 has a direct connection to the sink node (HC=1) hence it does not issue an IBF this time but since it also gets an IBV from the data sink with (HC=0) it issues an IBF triggered by the sink node IB message. For nodes 4 and 5, node 1 is the first connection towards the sink node hence both nodes issue an IBF. Node 4 also gets the IBV from node 2 and since the HC is lower than its own HC, it issues an IBF message. In this way, all nodes at the HC level 2 issue the IBF multiple times thus also receiving multiple IBVs with the same HC as their own hop count. Only when the HC of the IBV gets larger or equal to the HC of the receiving node, no IBF is issued any more. The break condition of termination of a rebroadcast can be stated as follow:

Where (0 £ X if HC Node > 1) as well as (X = 0 if HC Node £ 1). At the end of initialization step, each node has a routing table which can be used to route the data. The cost metric is based on remaining energy of the sensor node. The protocol consumes about 128 bytes of RAM and is well suited for ultra low power WSNs. Although, EADV is shown to be energy efficient but as in

WCMS, nodes know their location information and as it is shown in [59] that the protocols utilizing location information scale better than those which do not use this information. Therefore, it is not investigated further for the reference application.

1.1.6.3 Hierarchical

Hierarchical routing protocols arrange nodes in form of clusters or trees where every cluster has a cluster head used to transmit data from nodes within a cluster towards the sink node. Cluster heads may also aggregate data from cluster nodes to decrease the number of packet transmissions and hence conserve energy. Advantages of hierarchical routing protocols include scalability and efficient communication.

LEACH

Low Energy Adaptive Clustering Hierarchy (LEACH) is a clustered based routing protocol proposed for sensor networks with energy constraints. LEACH assumes a fixed base station and homogenous energy constrained sensor nodes. The basic idea of LEACH is to set-up clusters based on local information. Then a cluster head is selected on random basis and rotated so that a particular cluster head does not drain all its energy. Data fusion is performed at each cluster head to compress the information to be sent to reduce communication overhead. LEACH assumes that all the nodes have enough power to reach the base station (2 hops routing, i.e., from sensor nodes to cluster heads and from cluster heads to base station) and hence it cannot be deployed in geographically large areas. Also overhead is introduced when cluster heads are selected and rotated dynamically. The assumption of fixed base station limits applicability of LEACH in sensor network applications where base station is mobile.

1.1.6.4 Evolutionary Computing based Protocols:

PSO Algorithm

Optimization is the technique of finding the maximum and minimum parameters of a particular function or process. This technique is implemented in scientific applications like physics, chemistry, economics and engineering where the optimum goal is to maximize the effectiveness, efficiency and productivity etc.

Particle Swarm Optimization (PSO) is a method that is required to explore the search spaces for a provided problem so as to find the settings or the parameters required to optimize the a particular objective. Usually this method begin from two separate concepts: the thought of swarm intelligence stand off the assessment of swarming behavior through definite kind of natural world like birds and fish; and the specialized study of the evolutionary computation. The PSO algorithm executes by maintaining numerous candidate solutions in the search space simultaneously. In the duration each iteration of the protocol, each candidate solution is calculated by the objective function that is being optimized by estimating the fitness or threshold of that solution. The individual contender resolution is able to be measured as a particle "flying" by the fitness landscape and verdict the greatest or least amount of the purpose function. At inception, the PSO algorithm chooses the candidate solutions randomly within the search space. The search space is comprised of all the possible optimal solutions along the x-axis and the originating curve represents the objective function. Furthermore it ought to be noted that the Particle Swarm Optimization algorithm have no in sequence of the fundamental purpose function, and therefore it is not contain a method to identify if some of the applicant solutions are close to to or far away from a local or global utmost. The Particle Swarm Optimization algorithm basically utilizes the purpose function to estimate its candidate explanation, and operate upon the ensuing strength values.

Every particle constituent maintains its position that is encompassed of the applicant solution beside with its intended fitness limitation and its rapidity of roaming. On the other hand, it acknowledges the best possible fitness value that it has obtained during the entire operation of the algorithm and is generally referred as individual best fitness value and the candidate solution. Ultimately, the Particle swarm optimization (PSO) algorithm keeps its fitness value maintained among all particles in that particular swarm and thus is referred as the international greatest fitness, and the application solution that productively accomplishes that exacting fitness or threshold is passing on as the universal most excellent position or international greatest candidate solution.

The normal particular swarm optimization algorithm usually encompass of the subsequent ladder, which are frequently sustained till the strength circumstances are met:

1. Calculate the fitness value or fitness parameter for each particle.

2. Update the achieved individual and global best fitnesses parameters and its positions

3. Update the velocity and the location of the participating particles.

The fitness parameters are calculated by providing the candidate solution to the predefined objective function. Meanwhile the individual as well as global optimum fitnesses parameters and locations are rationalized by comparing the newly raised threshold or the fitness parameter against the previous fitness parameters. Uniformly, the swarm or particle rapidity and its position inform steps are accountable for the optimization competence of the Particle Swarm optimization algorithm.

The rapidity of every subdivision in the swarm is reorganized using the subsequent equation:

The notation of the particle is presented by. Therefore indicates the velocity of particle at a particular time and represents the position of particle at a particular time.

The parameters,, and are user-assigned coefficients. And are random parameters that are regenerated for each velocity update.

The value is the individual optimum candidate solution for a swarm particleat time, and represents the swarm’s global most optimum candidate solution at a particular time.

PSO algorithm has exhibited a potential role of a very robust system architecture where the system is optimized so as to deliver the best optimum output. A number of systems have been designed for PSO based network QoS optimization and even these technology have provided a better results as compared to the existing systems. Then while there are certain limitations in PSO algorithm, few of the limitations of Particle swarm optimization are like premature convergence that results into a huge degradation in performance as well as sub-optimal solutions. In order to illuminated the possibility of premature convergence, diversity guided PSO was utilized whereas mutation is also implemented to the swarm particles.