Node Deployment and the Impact of Relay Nodes in Wireless Sensor Network

Transcription

1 Node Deployment and the Impact of Relay Nodes in Wireless Sensor Network Rudranath Mitra 1, Diya Nandy 2 Department of Information Technology Heritage Institute of Technology Anandapur,Kolkata INDIA. mitra.rudra@gmail.com diya1nandy@gmail.com Abstract Node deployment is a fundamental issue to be solved in wireless sensor network. A proper node deployment can reduce the complexity of problems in WSN as for eg, routing, data fusion, communication etc. Different node deployment models have been proposed to reduce the complexity. In this paper we will discuss about the three models- Tri-hexagon-tiling (THT), Uniform Random and Square Grid. The nature of deployment of sensor nodes depend on the type of sensors, application and the environment where the networks will operate. Deployment of sensor nodes can be random or pre-determined. In random deployment nodes are deployed in a random manner. In pre-determined deployment, location of the nodes are specified. Most commonly used cell structure is Regular Hexagonal Cell Architecture. Here we will discuss the concept of relay nodes and its impact in sensor nodes. Keywords: Node deployment, relay nodes, sensor nodes. 1.Introduction A wireless sensor network is a collection of sensor nodes which are deployed in a given area of interest[1]. A sensor node is made up of components such as sensing unit, processing unit, transceiver unit and a power unit. The sensor nodes collect the data from its neighbouring nodes and send the collected data to the neighbouring nodes that are situated in a single hop. Those nodes send in turn the collected data to the sink node. The sink node is responsible for collecting data from inside the network processing them and sending them to the outside world. The sensor nodes are equipped with battery and whose battery cannot be replaced after deployment, so the major concern is to conserve the energy of the sensor nodes. The nodes that are placed closer to the sink node have to do more processing power than the other neighbouring nodes and so their battery power also drains out faster than the other nodes causing possible shortening of network lifetime. Therefore, deployment of sensor nodes has serious impact on network lifetime. The nature of deployment of sensor node depends on the type of sensors, application and the environment where the network will operate. Deployment of sensor nodes can be random or pre-determined. In random deployment nodes are randomly deployed generally in an inaccessible terrain. For example, in the application domain of disaster recovery or in forest fire detection sensors are generally dropped by helicopter in random manner[2]. In pre-determined deployment, the locations of the nodes are specified. It is mainly used in indoor applications. For example, manual placing of sensor nodes in pre-determined locations is done to monitor manufacturing plants, detection of corrosions and overstressed beams in large old buildings etc. In WSN we try to deploy a minimum no. of relay nodes so that between every pair of sensor nodes there is a connecting path and such that each hop of the path is no longer than the common transmission range of the nodes. To tackle the random deployment issue from a lifetime perspective, we must account for the biased energy consumption rate (BECR) phenomenon, described as follows. Consider a heterogeneous wireless sensor network composed of relay nodes (RNs) and sensor nodes (SNs).Traffic originates at sensor nodes(sns) and is send to the base station via the relay nodes(rns). In a single hop heterogeneous WSN, where RNs transmit data to the BS in one hop, such as in Fig.1(a), the RNs which are farther away from the BS will deplete energy faster than the RNs closer to the BS due to the larger transmission distance. In contrast, in a multiple hop heterogeneous WSN, where RNs transmit data to the BS via multi-hop relay, such as Fig.1(b), RNs which are closer to the BS will consume energy 1 358

2 faster than RNs farther away from the BS. The reason is because traffic is built up on RNs closer to the BS as it is relayed from far to near. As such, the RNs nearer to the BS will become unusable earlier than those far from the BS. between layer 2 & 3 there are two minimum distant vertices associated with a Cp cell ( C(5,2) ). These two vertices are priority-1 vertices. Similarly the minimum-distant vertices associated with the Cs cell on the same boundary are categorized as priority-2 vertices (Vprior 2 ). There is only one minimum-distant vertex with priority-2 associated with a s C cell ( C(4,2)) on this boundary. The rest of the vertices (if any) on the boundary are with priority-3 (Vprior 3 ). The following relevant notations are used to describe the architecture: r radius of a cell. Rs sensing range of a sensor node. Rc communication range of a sensor node. In this paper we will discuss about the Regular hexagonal structure, node deployment models, Sink routing hole problem and the impact of sink routing hole problem, various system models and deployment strategies. 2.Regular Hexagonal Cell Architecture[4][5] We consider regular hexagonal cell (RHC) [4] architecture where the network coverage area is divided into regular hexagonal cells as shown in Figure 1. A cell indicated by C(j,i) denotes the jth number cell of the ith layer. For example, cell C(6,2) is located in layer 2 and the cell number within this layer is 6. The sink node is located at the centre cell of the regular hexagonal cell architecture. The sensor nodes are placed in cells of different layers surrounding the centre cell. The cells of the layers are further categorized into two groups- primary and secondary. Primary cells ( Cp ) in a layer are those cells where the layer takes a turn of 60 and share a common boundary with more number of cells of the adjacent layer. Primary cells in the architecture are shown as shaded hexagonal cells. Secondary cells ( Cs ) are those which share a common boundary with relatively less number of cells of the adjacent layer. Secondary cells are shown as non-shaded hexagonal cells. The number of cells in each layer is 6 * i, where i=1, 2,, N and N is the number of the farthest layer from the sink. The minimumdistant vertices associated with the Cp cells of ith layer on the boundary between ith & (i+1)th layers are categorized as priority-1 vertices (Vprior 1 ). For example in Figure 1, on the boundary line Fig. 2 Regular hexagonal cell Architecture The relationship between cell radius r and node s sensing range Rs must satisfy r<= Rs/2 cover the whole cell area and the relationship between r and communication range Rc must be r<=rc/ 13 for ensuring the connectivity between neighbouring nodes. 3. Node Deployment Models [6][7] The designer knows how many sensors he has with him,only he has to decide to throw it in random or deterministic fashion. A circular feild of radius R is considered in these models. Here we will consider three models : Uniform Random, Square Grid and Tri-Hexagon-Tiling (THT)

3 3.1.Uniform Random In the uniform random deployment, each of the n sensors has equal probability of being placed at any point inside a given field, as shown in Figure 3. Consequently, the nodes are scattered on locations which are not known with certainty. For example, such a deployment can result from throwing sensor nodes from an airplaine. In general, a uniform random deployment is assumed to be easy as well as cost-effective. As we mentioned before, WSN applications often prefer random node deployment, which is why we assess its performance metrics here. 3.2.Square Grid There are popular grid layouts like unit square, an equilateral triangle, a hexagon, etc. Among them we find square grid to be the most feasible one because of its natural placement strategy over a unit square. Figure 1(b) shows a grid deployment of n sensors in a circular field, where each of the n grid points hosts a sensor. The approximate length of a unit square, d, can be calculated in the following way: First, the approximate area of a unit square with length d can be computed by dividing the whole area of a given field having radius R, with the number of cells, k. We do not know the value of k, but it is approximately equal to ( n 1)^2 for the square grid. From this relation, we derive Equation 1 for rsense, the sensing radius. However, since we consider an initial adjustment for a starting point, Equation 1 cannot be applied directly. According to simulation results, Equation 2 gives more precise values than Equation 1. Although we use these equations to find out the rsense (i.e., the length of a square, d ) given n and R, this formula allows the approximate computation of any one parameter out of n, rsense, and R given the other two parameters. 3.3.Tri-Hexagon-Tiling (THT) Here we are using the concept of tiling.a tiling is the covering of the entire plane with figures which do not overlap nor leave any gaps. Tilings are also sometimes called tesselations. Among different tilings we use a semi-regular tiling (which has exactly eight different tilings) where every vertex uses the same set of regular polygons. A regular polygon has the same side lengths and interior angles. We consider a semi-regular tiling that uses triangle and hexagon in the two dimensional plane, the so-called Tri-Hexagon Tiling. Here we are using the advantages of hexagon grid and triangle grid. A triangle grid uses a larger rsense than a square grid for the same n and R. In particular, the square grid uses about 5% of rsense less than the triangle grid. In a hexagon grid, rsense is about 17% less than in the triangle grid. In this aspect, the hexagon grid seems better than others, but with respect to other performance metrics it does not behave well. For this reason, we consider THT deployment, which uses 13% of rsense less than the triangle grid. In a way similar to the square grid, an approximate formulation for rsense can be found for THT. Fig. 3. (a)random (b)square Grid (c) THT Node Deployment 4.Sink Routing Hole Problem [8][9] An interesting observation is that, the forwarding workload of sensors increases inversely with their distances (hop count) to the sink a node closer to the sink usually has a higher relay workload than those of farther nodes. Accordingly, direct neighbours of the sink have the greatest transmitting workload. This incurs the failure of nodes near the sink at the very beginning because of energy depletion. Such scenario is referred as the Sink routing-hole problem. Multiple sinks or mobile sinks may help somewhat, but cannot eradicate this problem by multiple generated Sink routing-holes. Data-funnelling and aggregation techniques may alleviate the problem to some extent, but cannot eliminate the problem either. This work is solely related to the connectivity issue of networks but not the sensing coverage yet. The impact of Sink Routing Hole Problem 3 360

4 To capture the efficiency of the energy utilization in different scenarios, we define the metric energy efficiency as the ratio of the consumed energy over the total initial energy when the Sink routinghole appears. That is: An ideal η of 1 can be achieved by directing data transmission such that all sensors use up their energy when the whole network is disassociated. 5.System Models In WSN we try to deploy a minimum no. of relay nodes so that between every pair of sensor nodes there is a connecting path and such that each hop of the path is no longer than the common transmission range of the nodes. There are different types of system models: Energy model, Network model, Usability and lifetime model and routing schemes[2]. the network. NRN RNs are to be randomly deployed according to some strategy. 5.2.Energy Model In this model the energy spent by a SN for transmitting one packet to RNsis fixed as the transmission radius and packet lengths are fixed. In one round of data collection, the energy spent by an active RN consists of two parts, i.e., the energy used for intra-cluster communication and data processing, denoted by Eintra, and the energy used for inter-cluster traffic relay, denoted by Einter. Consider a RN having n member SNs[1]. The energy Eintra is composed of three parts, namely, the energy cost of receiving n packets of length l, denoted by ERX(l,n), the energy cost of transmitting the aggregated packet of length lag to its next hop RN or the BS over a distance rrn (fixed transmission power/range), denoted by ETX(lAG), and the energy cost of aggregating n packets of length l, denoted by EAG(l,n).Adopting an energy model similar to that in, we have 5.1.Network Model We assume a large-scale heterogeneous WSN on a sensing field A is composed of three types of devices, Sensor Nodes (SNs), Relay Nodes (RNs) and a Base Station (BS). A SN senses the environment, generates data, and periodically transmits the data to an active RN1, which functions as a cluster head (CH), in a single hop. It has limited energy and a fixed transmission radius rsn. It has no relaying function or at least traffic relaying is not a routine function of a SN for the following reasons. First, relaying traffic demands high intelligence, such as security and routing, which leads to higher device cost. Second, extra communication leads to faster energy dissipation. A RN is also energy constrained and has fixed transmission range rrn, where typically rrn is a few times larger than rsn. A RN works as a CH when active, which groups the SNs in its proximity into a cluster. It also coordinates and schedules the MAC layer access within its cluster so that the energy overhead, e.g., retransmissions due to collisions, is minimized. After receiving the data from SNs, it aggregates the traffic.the aggregation diminishes the redundant information from multiple nodes and reduces the network traffic. In the end, it transmits the aggregated data to the next hop active RN according to the routing algorithm running on these active RNs. The aggregated traffic won t be aggregated again while passing through other RNs. We assume the traffic is light compared with the available bandwidth, or the traffic is well scheduled so that there is no traffic congestion in Where β, α1,α2, m, and γ are energy related parameters. Letting g be the aggregation ratio, the length of the aggregated packet from n packets is, Replacing lag in (2) by (4), and adding (1), (2), and (3), we have Eintra=c1nl (5) On the other hand, the energy spent on intercluster relay Inter E consists of two parts, namely, the energy cost of receiving packets of total length lrelay and transmitting them (as they are) over the distance rrn. Thus, the total energy spent by a RN in one round of data collection is, 4 361

5 5.3.Usability and lifetime Model The usability of a WSN is determined by both coverage and connectivity. Coverage has two aspects, i.e., coverage area and coverage degree. Connectivity refers to how much of the generated data can ultimately arrive at the BS. It can be measured by the percentage of SNs that can connect to the BS via RNs. As such, coverage provided by SNs and connectivity provided by RNs ultimately determine the effective coverage. As RNs get drained of energy, the connectivity becomes gradually weaker and so does the effective coverage. This process is called coverage aging. 5.4 Routing Scheme sub-optimal deployment density function. Indeed, the lifetime is increased up to more than 3 times by using the heuristic weighted deployment than the uniform deployment[3]. The average deployment density in a given area should depend on two factors, namely the average total energy consumption rate in the area and the size of the area. The energy consumption rate of an area is the total energy consumed by RNs in the area per round of data collection. To overcome the BECR problem, the average density over an area should be proportional to the energy consumption rate and inversely proportional to the size of the area. For example, in Fig. 4, consider two arbitrary shells, B1 and B2, with the BS at the center. The size of B1 is larger than that of B2. Due to the BECR phenomenon, suppose that RNs in B1 and B2 have same energy consumption per round. Thus, B2 should have higher deployment density so that the expected numbers of RNs are the same in the two areas. In this routing scheme we only assume that relay paths between RNs are always from far to near to the BS in order to avoid the unnecessarily long paths. 6.Deployment Strategies In this section we propose and examine deployment solutions for the following problem 6.1.Connectivity oriented deployment: Compared with other random deployments, this strategy provides identical and maximal connectivity everywhere in the WSN. In other words, for a given connectivity requirement σ0, this strategy will require the least number of RNs. We therefore refer to it as the Connectivity- Oriented Deployment Strategy. However, due to the BECR phenomenon discussed above, it suffers fundamentally from an energy efficiency perspective[2]. 6.2.Lifetime oriented deployment [2][3] The other name of this deployment is referred to as the weighted random deployment. The motivation is that the number of RNs deployed at different locations, and so the deployment density function, should be proportional to the expected energy dissipation rates at these locations. Due to the addup effects of the traffic relay and the randomness of the geometrical distribution of the RNs, deriving a provably optimal density function is a non-trivial task. Next, we present the derivation of a heuristic Fig 4. A sensing site: the density function is proportional to the energy consumption rate, and is inversely proportional to the size of areas. 6.3.Hybrid Deployment The weighted random deployment of RNs according to the density function can counteract the BECR phenomenon. However, this benefit will be materialized only if the connectivity of SNs is satisfied in the network[1]. If the number of given RNs, NRN, is less than, the number of SNs without connectivity may be too high for a network to function at all. The objective of the hybrid deployment is to optimize RN deployment by balancing the concerns of connectivity and lifetime extension. If there is no way to guarantee the connectivity at the first place[3]. If then the weighted deployment can provide the satisfying connectivity. If the weighted random deployment alone will not be able to satisfy the connectivity. Now we will see that the NRN are optimally split into and 5 362

6 Allocation of RNs for the two steps is a constrained optimization problem. As increases, has to be decreased. However, if is too small, the connectivity of the sparse area of the network is at risk. Since the connectivity of SNs is not satisfied outside the circle of radius d0, we arrange the deployment so as to compensate for the weaker connectivity in the area. [6] Crossbow: Mpr-mib users mannual. June [7] Karl, H., and Wittig., A. Protocols and Architectures for Wireless Sensor Networks. Wiley, [8] D. Culler, D. Estrin, and M. Srivastava, "Sensor Network Applications," IEEE Computer, pp , [9] W. Xue, Q. Luo, L. Chen, and Y. Liu, "Contour Map Matching For Event Detection in Sensor Networks," In proceedings of SIGMOD, Conclusion The deployment of nodes is an important issue in Wireless Sensor Network. The important thing that has to be kept in mind is to maximize the network lifetime as the sensor nodes are equipped with battery and cannot be replaced once they are placed in a sensor field. We have discussed various deployment models so that the network lifetime can be increased. We have also discussed various system models and deployment strategies so that we can use minimum number of sensor nodes and increase the network lifetime 8.Acknowledgements We express our sincere gratitude to the Head of the Dept. and the members of the Research Team, Dept.-of-Information Technology,Heritage Institute of Technology.We do also acknowledge and thank Prof(Dr.)Somenath Mitra,Shri Souren Pathak and Shri Indranath Mitra for constant support and encouragement. 8.REFERENCES [1] K. Xu, H. Hassanein, G. Takahara, Q. Wang, Relay Node Deployment Strategies in Heterogeneous Wireless Sensor Networks: Single-Hop Communication Case, IEEE Globecom 2005, to appear. [2] S. Shakkottai, R. Srikant, N. Shroff, Unreliable Sensor Grids: Coverage, Connectivity, and Diameter, IEEE Infocom 2003, Vol. 2, pp [3] K. Chakrabarty, S. S. Iyengar, H. Qi, E. Cho, Grid Coverage for Surveillance and Target Location in Distributed Sensor Networks, IEEE Transactions on Computers, 51(12): , [4] Akyildiz, I.F., Su, W., Sankarasubramaniam, Y., Cayirci, E.: A survey on sensor networks. IEEE Communications Magazine 40(8), (2002) [5] DasBit, S., Ragupathy, R.: Routing in MANET and Sensor Network- A 3D position based approach. Journal of Foundation of Computing and Decision Sciences 33(3), (2008) 6 363