Telecommun Syst DOI 10.

1007/s11235-011-9597-y

Fuzzy-logic based routing for dense wireless sensor networks

Antonio M. Ortiz Fernando Royo Teresa Olivares Jose C. Castillo Luis Orozco-Barbosa Pedro J. Marron

Springer Science+Business Media, LLC 2011

Abstract The task of routing data from a source to the sink is a critical issue in ad hoc and wireless sensor networks. In this paper, the use of fuzzy logic to perform role assignment during route establishment and maintenance is proposed. An incremental approach is presented and compared with similar existing routing protocols. Efcient routing approaches provide network load balance to extend network lifetime, efciency improvements, and data loss avoidance. Experiments show promising results for our proposals and its suitability for operating with dense networks, obtaining quick route creation as well as energy efciency. Keywords Wireless sensor networks Routing Fuzzy logic

1 Introduction Wireless Sensor Networks (WSNs), as well as other wireless personal area networks, have stirred up the world of wireless communications since they present new challenges in terms of energy efciency and communication performance. Sensor nodes are resource constrained in terms of energy, processing capabilities and storage. This kind of networks also have to deal with problems such as mobility and reliability. All these issues make necessary some kind of network organization to create and maintain data paths and to ensure reliable and efcient communications among the network nodes. Routing data in networks composed in many cases of a high number of low-resourced nodes, is a difcult task since the algorithms and protocols have to save as much energy as possible whilst offering good performance. These protocols have to be designed by considering the optimization of parameters, such as the battery status. Not considering this information leads to problems in the network such as interrupted paths, data loss or isolated nodes, among others. These problems are directly related to latency and throughput values. Efcient routing approaches should balance the network load in order to extend the network lifetime, efciency improvements, and data loss avoidance. Network monitoring is also necessary to control topology changes and the addition or elimination of nodes in the network. This work presents an incremental approach. Firstly, NORA (Network rOle-based Routing Algorithm), a roleassignment-based routing algorithm is presented, and secondly its evolution, NORIA (Network rOle-based Routing Intelligent Algorithm), a novel routing algorithm for wireless sensor networks which combines different effective techniques in order to reduce the energy spending and improve data routes. These techniques are role assignment for

A.M. Ortiz ( ) F. Royo T. Olivares J.C. Castillo L. Orozco-Barbosa Albacete Research Institute of Informatics, University of Castilla-La Mancha, 02071 Albacete, Spain e-mail: amanuel@dsi.uclm.es F. Royo e-mail: froyo@dsi.uclm.es T. Olivares e-mail: teresa@dsi.uclm.es J.C. Castillo e-mail: josecarlos@dsi.uclm.es L. Orozco-Barbosa e-mail: lorozco@dsi.uclm.es P.J. Marron University of Duisburg Essen, Bismarckstr. 90 Building BC, 47057 Duisburg, Germany e-mail: pjmarron@uni-due.de

A.M. Ortiz et al.

distributing tasks over the network nodes and fuzzy logic for making decisions. Role assignment enable nodes with better resources to act as data routers in order to forward information from more distant nodes up to the base station. The comparison of node conditions in NORA is performed by mathematical operations, while NORIA is based on a fuzzy logic system that improves the decision making process, reducing the number of discovery packets and the number of gateways while keeping low the route-creation time, making the algorithm even more reliable and efcient. Our proposals are compared with two well-known routing approaches, a simple tree-routing algorithm based on Tree Routing protocol implemented in ZigBee [20], and a Connected Dominating Set, CDS-based routing protocol [1]. Routing using CDS has been widely used to perform routing over wireless ad hoc and sensor networks and will serve as a basis for checking the performance of our proposal. The experiments show the efciency of NORA and its suitability for working with dense networks in an effective and reliable manner and, additionally, the improvements achieved by NORIA, making the discovery process even more efcient. The rest of the paper is organized as follows: Sect. 2 details related work, Sect. 3 presents NORA, and its evolution, NORIA, as well as network problem resolution. Experiments, tests and comparisons are detailed in Sect. 4, and nally, Sect. 5 gives some conclusions and lines for future work.

2 Related work Sensor nodes are small and energy-constrained devices with limited computational capabilities and memory resources. Because of that, the development of a new generation of network protocols with new characteristics (such as autodiscovery, self-organization, adaptation to dynamics, and generally a higher degree of distribution) is essential to fulll the new requirements of sensor networks. In [2] routing techniques for WSN are reviewed. This survey serves as a basis for future self-organization algorithms which will improve network efciency. A general denition and classication of self-organization techniques in ad-hoc and sensor networks can be found in [3]. In [4], virtual structures such as backbone and cluster are proposed, but the necessity of wired nodes discards it for wireless sensor networks. Previously, some general design paradigms for selforganized networking were proposed in [5] and [6]. These paradigms propose a new network organization model by introducing a new concept: the role. Role assignment is a network organization technique which allows nodes or groups of nodes to have different

functionality in order to improve network performance. Role assignment can be performed in several ways: rule-based paradigm [7, 8], application-based approach [9] or nodesplacement-based approach [10]. This work extends the rulebased paradigm in order to make routing decisions. Another interesting technique is the use of set algebra to calculate Connected Dominating Sets (CDS) in order to obtain efcient routing, as was proposed in [1]. A set is dominating if all the nodes in the system are either in the set, or neighbors of at least one node in the set. This technique has been used for wireless ad hoc and sensor networks [11, 12]. In this work, CDS technique is used as theoretical basis to compare our proposals. Towards this end, we have implemented the routing approach appearing in [13], and its performance is compared with our proposal. NORA performs decision making by using a series of mathematical operations; in order to improve this process, our second proposal, NORIA, follows a fuzzy-rule-based paradigm. Nodes store and evaluate a set of fuzzy rules, taking both its own and neighborhood parameters into account. Role assignment and routing decisions are performed according to the output of the fuzzy-logic-based system, providing global energy saving and focusing packet load to nodes with a better state. The idea of using articial intelligence techniques to support the decision-making process in order to get more efcient algorithms is widely used in the recent literature. Nowadays, there are several algorithms that apply these techniques to ad-hoc network organization algorithms and routing protocols. Approaches based on fuzzy-logic [28], machine learning [15], neural networks [16], genetic algorithms or ant colonies [17] can be found. Articial intelligence techniques reinforce the efciency and performance of routing protocols, by combining data from nodes and their interactions in order to make decisions to improve global network performance. Some approaches such as ant colonies require a large number of messages, being automatically discarded for its use in WSNs [17]. Neural networks requires a complete knowledge of the WSN prior to discovering, and genetic algorithms and machine learning requires high computational capabilities [15, 16]. Since we aim to evaluate node conditions in an efcient manner, our approach makes use of fuzzy logic, which requires low computation capabilities, and is able to support these decision-making processes to improve efciency while extending the overall network lifetime. Fuzzy-logic imitates the logic of human reasoning, which is much less rigid than the calculations computers generally perform. Diverse approaches using fuzzy-logic have been presented for cluster-based routing approaches improvement [14, 25, 26], and also for directed diffusion routing security [27], endorsing so the use of fuzzy logic in wireless sensor networks. These approaches take advantage of the use of

Fuzzy-logic based routing for dense wireless sensor networks

fuzzy information treatment in order to get efcient routing, and will be applied herein to our basic tree-based routing protocol.

3 Routing proposal The rst proposal of this work is the use of a distributed routing algorithm that assigns roles to the network nodes using both local and neighborhood information, and creates energy-efcient routes to the sink. Discovery and routing processes are complemented with mechanisms to maintain routes and to manage the addition and failure of nodes in the network. The different experiments performed have led us to propose an incremental approach. The proposal begins by performing role assignment in order to have nodes with different functionality in the network. To assign this functionality, it was necessary to implement an efcient method able to evaluate node conditions. The second proposal uses fuzzy logic in order to obtain a fast and effective technique to evaluate node conditions. Furthermore, the fuzzy logic implementation presents low computational requirements [30], so it is adequate to be executed in low-resourced nodes composing WSNs. 3.1 The role-based approach Our rst routing proposal assigns roles to nodes in the network in order to establish the tasks for each node or group of nodes. Role assignment lies in assigning different tasks to each node or group of nodes in the network in order to globally improve network performance. This technique uses both local and neighborhood data to make decisions. Role assignment paradigm is an efcient manner of optimizing specic parameters such as network lifetime, path length, QoS. . . , while data routing is performed. Our role-based approach to create routes in the network, NORA (Network rOle-based Routing Algorithm), evaluates node conditions and assigns roles depending on current node and neighborhood characteristics. The whole process begins at the base station (sink node or coordinator), and nishes at the furthest nodes. Intermediate nodes decide between being end devices (nodes that just send sensed data) or routers (which, in addition to the formers, also forward data coming from end devices). Each router or end device selects the best router to act as parent, inside their radio range, to forward its data to the coordinator. Following this scheme, every node in the network is able to send data to the sink either directly, or through router nodes. The protocol establishes minimum routes in terms of energy consumption and efciency, from every node in the network to the base station, this latter gathers data from all

network nodes. The considered node conditions, in order to select nodes optimizing this parameters to route data from their neighbors are the remaining battery and the distance to the sink (number of hops). To perform role assignment and route selection, NORA uses ve kinds of message: IPM (Information Propagation Message): includes local information (node ID, number of hops to the sink, remaining battery, . . . ). RDM (Role Decision Message): includes the same information as IPM, and is interpreted by nodes as a trigger to initiate the discovery process. RRM (Router Request Message): this kind of message is used by nodes which do not have any router within their radio range (no node can forward their data), and urges an end device to become router. RCM (Role Changing Message): used by router nodes with low resources to notify role changing (from router to end device), and urge their dependent end devices to look for another available router. ACK: unicast message to acknowledge addressed messages. It is used to control router presence and allows nodes to look for another router in case of parent failures. Data are encapsulated in Data Messages (DM) which are independent of the control messages dened above. The route creation process followed by NORA is outlined in Fig. 1. In NORA, the discovery process begins when the base station sends an RDM. Nodes receiving this message send an IPM and start a timer. Along that time interval, nodes wait for information messages from neighboring nodes. Once the timer expires, nodes perform role decision and parent election, that is, selection of the best router among the known routers located at a lower level, i.e. lower number of hops to the base station. One hop nodes will choose the base station as parent. If no router is found, an RRM is sent to the best end device neighbor, i.e. the one characterized by the lowest number of hops, and highest battery level. Once nodes have selected role and parent, an RDM message is broadcasted in order to induce the next hop

Fig. 1 NORA organization phase transitions

A.M. Ortiz et al.

neighbors to start the organization process. This RDM also informs the selected router on the new child. This procedure is propagated hop by hop until reaching the furthest nodes and all network nodes have a route to reach the base station. The pseudo-code corresponding to this phase is shown below: when receive(RDM) then: update_neighbors_table(RDM) send(IPM, BROADCAST) wait(temp) foreach received message during temp do update_neighbors_table(message) end foreach end wait temp_coord = select_coordinator() if get_role(temp_coord)==leaf then send(RRM, temp_coord) end if set_coord(temp_coord) local_role = calculate_role() send(RDM, BROADCAST) Lets formally dene the route-creation process: Considering Ha,b as the number of hops between nodes a and b , and as the set of all network nodes. Equation (1) denes the set of neighbor nodes, n , of a given node n with lower hops number to the base station, B . n = n {i }/i , H{i,B } < H{n,B } (1)

routers available, an RRM is sent to the highest battery level end device to induce it to switch its role from end device to router. In order to make own role decision, each node compares its battery level with same level (number of hops) neighbors. If local battery level is the highest, then the node sets its role as router. End device role is assigned otherwise. Equation (5) denes this process. R(n, t) = ROUTER, END DEVICE, if Ln = max{Ln } otherwise (5)

Similarly, (2) denes the set of same level neighbors (nodes with same number of hops to the base station), n . n = n {i }/i H{i,n} = 1 and H{i,B } = H{n,B } (2) With these previous denitions, router (parent) selection process as well as role decision can be dened. If we consider Li and Ri as the battery level and role of the node i , respectively, parent of node n, P (n, t), at time t is selected by using (3). P (n, t) = i n /Ri = ROUTER and Li = max{Ln } and H{i,B } = min{H{n ,B } } (3)

If current node has equal number of hops and battery level than other(s), the received signal strength is used to decide. To sum up, using NORA all network nodes are able to nd a path to reach the base station. Furthermore, forwarding nodes (routers) are those with lower number of hops to the sink and highest battery level. These assumptions avoid node failures and ensure a reliable, and energy-efcient network operation. Depending on particular application requirements, e.g. real time, extended network lifetime. . . , it is possible to select the parameters to be considered during route creation process such as available sensors, memory and processing capabilities, number of nodes connected to a coordinator and so on. Since wireless sensor networks need simple and fast methods to make decisions, fuzzy logic appears as an appropriate approach due to its ability to calculate results fast and precisely. Moreover, the user-friendly nature to dene node conditions provided by this approach and the need of low processing resources make this technique a suitable method to make decisions in wireless sensor networks. In order to improve the efciency and accuracy of the route creation process and to speed it up, the evaluation of node conditions through fuzzy logic is proposed. We have incorporated a fuzzy-logic engine in the decision process of NORA. 3.2 Fuzzy-logic principles Fuzzy logic consists of a decision system approach which works similarly to human control logic. It provides a simple method to reach a conclusion from imprecise, vague, or ambiguous input information. The execution of a fuzzylogic system requires less computational power than conventional mathematical computational methods [18]. Furthermore, only a few data samples are required in order to extract the nal accurate result. Besides, fuzzy logic is a handy technique since it uses human language to describe inputs and outputs [21]. In a fuzzy-logic-based system, calculations are performed by an inference engine. In order to select the inference engine, we have studied two widespread approaches

If no router node is found, then (4) is used and an RRM is sent to the selected node to request role switch from end device to coordinator. P (n, t) = i n /Ri = END DEVICE and (4)

Li = max{Ln } and H{i,B } = min{H{n ,B } }

Parent node is chosen as the highest battery level router among neighbors with lowest hops number. In case of no

Fuzzy-logic based routing for dense wireless sensor networks Fig. 2 Mamdani and TSK data evaluation diagram

present in the literature: Mandani [22], and TSK [23]. Both of them proceed in a similar way, consisting of four phases: fuzzication, rule evaluation, combination or aggregation of rules, and deffuzication (see Fig. 2). The main difference is presented at the deffuzication stage in which TSK rule consequents are mathematical functions (not fuzzy), loosing so its interpretability [24]. Moreover, as will be explained below, in our proposal, rule outputs are independent from each other. For our implementation, it does not make any sense to aggregate different nature outputs with a weighted average as TSK does. In this work, we aim to get the best match (max-min inference) and for that approach, the use of the Mamdani makes perfect sense. Lets now detail the Mamdani inference system. The input of a Mamdani fuzzy-logic system is usually a crisp value. To allow this value to be processed by the system, it has to be converted to natural language, that is, it has to be fuzzied. In this way, the fuzzier method takes numeric values and turns them into fuzzy values which can be processed by the inference system. These fuzzy values represent the membership values of the input variables to the fuzzy sets. Once values have been fuzzied, the inference system processes the fuzzy rules to get a fuzzy output. In the case of a fuzzy rule having more than one antecedent (conditional element), an AND (minimum) or OR (maximum) operator is used to estimate the output value of rule evaluation. The third step in the Mamdani inference method is the aggregation of all outputs, where the outputs of each rule are combined to form a new fuzzy set. Finally, at the deffuzication stage, the new aggregated fuzzy set is converted to a number. Mamdani uses the centroid technique which tries to determine the point where a vertical line divides the combined set into two equal parts. 3.3 Routing using fuzzy logic In order to improve NORAs performance, the integration of fuzzy logic to the decision process is proposed. Here is where NORIA (Network rOle-based Routing Intelligent Algorithm) comes into play. Thus, parent election and role assignment are now based on the results of the evaluation of a fuzzy rules set. The input variables to be considered in the experiments are: number of hops to the base station, and the remaining

Fig. 3 Number of hops fuzzy sets

Fig. 4 Battery level fuzzy sets

Fig. 5 Output fuzzy sets, node suitability

node energy. These parameters are just a subset within the full set of parameters which can be included in the decision process (delivery probability, delay, or signal strength. . . among others). Route length and remaining battery have been selected herein since they represent an example of two of the most important optimization parameters in wireless sensor networks. The energy is a key parameter in WSNs since it represents one of the main problems of this technology, and the route length indicates the number of forwardings necessary to route data from the source to the destination. The output variable represents the suitability of the node to be a router and to be selected as parent. Figures 3, 4, and 5 show fuzzy sets for input and output parameters. To perform the role assignment and parent selection process, nodes will compare the evaluation output for each neighbor node. The variables and fuzzy sets used in this paper are an example of the multiple possibilities and have been selected after several checks for the application and

A.M. Ortiz et al. Table 1 Fuzzy rule base N. of hops Very low Very low Very low Very low Low Low Low Low Medium Medium Medium Medium High High High High Very high Very high Very high Very high Bat. level Low Medium High Very high Low Medium High Very high Low Medium High Very high Low Medium High Very high Low Medium High Very high Node suitability Low adequate Low adequate Adequate Perfect Not adequate Low adequate Adequate Perfect Not adequate Low adequate Adequate Adequate Not adequate Low adequate Low adequate Adequate Not adequate Not adequate Low adequate Adequate

detected during a pre-xed period of time, the new node will send a RRM to the best end device listened and will select it as parent. The role of the new node will be end device (by default) being able to be changed into router if a RRM is received. Low-resourced router: when a router realizes that its battery drops under a determined threshold, it sends a CR (change role). Then, children nodes will look for another router following the route-creation process described in Sect. 3.1. Thus, nodes sending through the low-resourced router are re-organized and network connectivity is preserved.

4 Experiments NORA and NORIA efciency will be evaluated through its implementation in OMNET++ [19], an extensible and modular component-based C++ simulation library and framework for network simulations. As aforesaid, our proposals are compared to a Simple Tree Routing protocol (STR) [20] and to a routing scheme based on Connected Dominating Sets (CDS) [1]. Simple Tree Routing protocol was chosen because it works similarly to NORA in the sense that STR builds a treebased routing scheme, but considering different node and neighborhood conditions. STR operation can be summed up as follows: (1) base station announces its presence; (2) nodes that have received the base station announcement send their own announcement message and start a timer; (3) once the timer expires, each node decides its parent node based on the number of hops and link quality information. This procedure spreads hop by hop until reaching all network nodes. This approach is similar to NORA in the sense that STR builds a tree-based routing scheme but with the absence of roles. CDS uses roles dening nodes inside and outside the set. But the procedure to calculate roles is different: CDS calculate roles before calculating routes, while NORA and NORIA perform role assignment at the same time that routes are created. The operation of CDS-based routing can be summarized up as follows: (1) rst it computes a marking process in which the protocol calculates a connected dominating set among all network nodes; (2) after that, all nodes in the network are in the CDS, or are neighbors of at least one node in the set. This marking process selects nodes depending on their set of neighbors in order to keep all nodes in the network connected. The routing process is divided into three steps: 1. If the source is not a gateway host, it forwards the packets to a source gateway, which is one of the adjacent gateway hosts.

topologies used, in order to have a generic proposal able to work in a wide range of WSN applications. Notice that fuzzy sets (input and output) can be customized depending on the application, requirement and circumstances of each particular WSN. For example, in a surveillance network would be interesting the use of variables such as delivery probability in order to ensure alarm nodes to receive important data. In our Mamdani-based implementation, the fuzzy rules base includes rules such us: if the Number of hops is Low and the Battery Level is High then the node suitability is Adequate. Here, since there are 4 fuzzy sets for Battery level input and 5 for Number of hops input, there exist 20 rules, which are summarized in Table 1. 3.4 Solving problems in the network During network operation, NORIA is able to control several situations related to router failures, new nodes coming to the network, and low-resourced routers. The algorithm manages these situations as follows: Router failure: if a node does not receive the ACK message from its router for at least two consecutive times, it will proceed as if it would have received a CR from its coordinator. New node joining the network: the new node listens the channel. If a message from a router or from the sink node is received, it is selected as parent. If no router or sink is

Fuzzy-logic based routing for dense wireless sensor networks

2. This source gateway acts as a new source to route the packets in the induced graph generated from the CDS. 3. Eventually, the packets reach a destination gateway, which is either the destination host itself or a gateway of the destination host. In the latter case, the destination gateway forwards the packets directly to the destination host. The parameters considered in the experiments, for the evaluation of the three schemes, are: Number of packets: represents the amount of packets sent during route discovery phase, and measures the necessity of information exchange for each approach. A low number of sent packets indicates that a protocol will be more efcient in terms of the energy spent during route discovery. Energy spending: due to nodes forming WSNs are energy constrained, the energy spent during the operation of the protocols is very important. The lower energy spending, the higher efciency, and consequently the longer network lifetime. Network set-up time: represents the time spent by a protocol to discover routes from every node to the base station. It is the time since the rst discovery packet is sent until all nodes in the network have a route to the base station. Number of gateway nodes: this variable represents the number of nodes that have to forward data from other nodes, and so subsequently spending more energy. For the application implemented in this work, we suppose the use of data aggregation [29], an effective technique to reduce the amount of information travelling through the network, so a lower amount of forwarding nodes implies a lower global energy consumption. 4.1 Set-up and scenarios In order to evaluate the performance of our proposals and to compare it with other similar approaches, the designed scenario has considered circular network areas with radius from R (50 m) to 10R (500 m), maintaining constant node density (nodes per unit of area). Nodes are randomly deployed in these areas and Base Station is placed at the center. In the largest area (10R), 1959 nodes have been used. Table 2 shows number of nodes for each experiment depending on network radio. For each scenario and particular combination of parameters, we have run 100 simulations. 4.2 Results The rst results show the number of necessary messages to create the routes across the network for NORA, NORIA, CDS and STR. The amount of messages used to create the

Table 2 Number of nodes by radio Radio R (50 m) 2R (100 m) 3R (150 m) 4R (200 m) 5R (250 m) 6R (300 m) 7R (350 m) 8R (400 m) 9R (450 m) 10R (500 m) Number of nodes 21 81 177 317 489 709 973 1257 1597 1959

Fig. 6 Packets sent during route discovery phase

routes is a good variable to evaluate the spent energy in this process. The least number of messages used, the lowest energy used. Figure 6 shows the average number of packets sent by nodes during the simulations of the route creation phase as a function of the number of network nodes, it is, the number of packets since the discovery process is started until all nodes in the network have found a route to the sink. Remark that STR uses the lowest number of packets to organize the network among different experiments (see Fig. 6) closely followed by NORIA. These results prove the energetic efciency of our proposal, spending just a little bit more energy than STR. This metric is very important when working with networks composed by a high number of nodes (dense networks) and for applications that require a proactive routing and cannot compute a route each time a node have to send data (reactive routing). Our experiments have also shown that while NORA, NORIA, and CDS are able to create routes for all network nodes, STR leaves out a large area of the network unorganized. The section varies from 1 to 5% of the total amount of network nodes. This is caused by the decision approach implemented in STR that leaves unorganized those nodes which cannot communicate with a signal power greater than

A.M. Ortiz et al.

Fig. 7 Energy consumption reduction for NORA, NORIA, and STR with respect to CDS

Fig. 8 Network self-organization time

50%. In contrast, our experiments have shown that nodes whose signal is received below 50% of power still can send successfully data to the sink. The energy consumption is highly related to the amount of packet sending. In order to illustrate the differences among the energy consumption of the different proposals, Fig. 7 shows the normalized energy consumption with respect to the highest value (CDS) for each approach as a function of the number of nodes. Notice that the energy consumption for NORA, NORIA, and STR has been normalized to the maximum value obtained (reached by CDS). This results not only shows the energy saving achieved by fuzzy-logic-assisted approach, but also that fuzzy-logic itself does not represent an increment in the energy consumption as might be thought. Another important metric of the routing algorithms is the time spent to set up a fully connected network, specially for applications with real-time requirements. A low route establishment time is also propitious when solving problems in the networks, to create new routes, and to perform periodical set-ups, necessary for some applications. Figure 8 shows the time spent by the different proposals to complete the route-creation phase. NORA and NORIA show good average results in the experiments, proving that both are very efcient in terms of route-discovery time, even spending half of the time than other proposals such as CDS. These results make NORA and NORIA suitable to be used in real time networks, in which the user needs a fast response from the network. It is also interesting to consider the number of router nodes in the network. This gives an idea of the number of nodes spending energy forwarding data from other nodes. Using data aggregation during data routing, the least number of router nodes, the least global energy consumption. Figure 9 shows the number of router nodes obtained in the experiments.

Fig. 9 Number of forwarding nodes

For that experiment, STR obtains the best value, closely followed by NORIA. But it is important to consider the problem of unorganized nodes left by STR. We then conclude that NORIA obtains the best average results in the experiments, making it suitable to be implemented in real dense wireless sensor networks and to be used with a wide range of applications. Furthermore, role migration is performed by using CR messages to get load balancing, avoiding node failure and data loss. Notice that no loops between nodes has occurred during simulations, that is, all created paths ows from any node in the network to the base station.

5 Conclusions and future work The desire to improve the existing routing approaches for wireless sensor networks has led us to design and experiment with the routing algorithm presented in this paper. NO-

Fuzzy-logic based routing for dense wireless sensor networks

RIA is a novel role-based routing protocol that makes use of fuzzy logic to make decisions. NORIA has been presented as an incremental approach, starting with a basic approach, NORA, improved with the addition of a fuzzy-logic-based system to be the basis of the decision making process. Simulation results show the correct operation of the protocol and its suitability to be used in a wide range of applications and scenarios. The performance of this proposal has been compared with two well-known routing methods and both NORA and NORIA have proved its efciency, by achieving better average results than the other proposals. The combination of fuzzy logic with role assignment to perform routing tasks has been proved to be a good association for working with dense WSNs in an energy-efcient, fast, and effective manner. Furthermore, the design of the fuzzy-logic system is simple and easy, allowing users to dene different variables, sets, and rules, depending on each particular application and node features. The use of fuzzy logic makes node feature denition easier, and the accuracy and the low amount of resources needed to run the system make this technique appropriate to be executed in the low-resourced nodes that make wireless sensor networks. The solution of network problems such as node failures, low-resourced routers and the addition of new nodes in the network is now being implemented and will be tested to evaluate the solutions proposed in Sect. 3. With this periodical monitoring, network operation will be extended and the efciency and reliability of the network will be improved. Our future work is now focused on the incorporation of other parameters to the decision system (end to end delay and delivery probability, for example) as well as the incorporation to the standard ZigBee [20] of the techniques proposed herein. The implementation and experimentation with the full system in a real WSN is also contemplated in our future plans.
Acknowledgements This work was supported by the Spanish MEC and MICINN, as well as European Commission FEDER funds, under Grants CSD2006-00046 and TIN2009-14475-C04.

References
1. Wu, J., Dai, F., Gao, M., & Stojmenovic, I. (2002). On calculating power-aware connected dominating set for efcient routing in ad hoc wireless networks. Journal of Communications and Networks, 4(1). 2. Al-Karaki, J. N., & Kamal, A. E. (2004). Routing techniques in wireless sensor networks: a survey. IEEE Wireless Communications. 3. Dressler, F. (2008). A study of self-organization mechanisms in ad hoc and sensor networks. Computer & Communications Magazine, 31, 30173029.

4. Theoleyre, F., & Valois, F. (2008). A self-organization structure for hybrid networks. Ad Hoc Networks, 6. 5. Prehofer, C., et al. (2005). Self-organization communication networks: principles and paradigms. IEEE Communications Magazine. 6. Di Marzo, G., Foukia, N., Hassass, S., Karageorgos, A., Mostefaloui, S. K., Rana, O. F., Ulieri, M., Vackenaers, P., & Van Aart, C. (2008). Self-organisation: paradigms and applications. In IEEE personal symposium on personal indoor and mobile radio communications (PIMRC). 7. Marron, P. J., Lachen Mann, A., Minder, D., Hahner, J., Sauter, R., & Rothermel, K. (2005). TinyCubus: a exible and adaptive framework for sensor networks. In Second European workshop in wireless sensor networks (EWSN). 8. Castillo, J. C., Olivares, T., & Orozco-Barbosa, L. (2007). Implementation of a rule-based routing protocol for wireless sensor networks. In 2nd ACM workshop on performance monitoring and measurement of heterogeneous wireless and wired networks (PM2HW2N). 9. Kochhal, M., Schwiebert, L., & Gupta, S. (2004). Role-based middleware for sensor networks. WSU-CSC-NEWS. 10. Dasgupta, K., Kukreja, M., & Kalpakis, K. (2003). Topologyaware placement and role assignment for energy-efcient information gathering in sensor networks. In 8th IEEE international symposium on computers and communications. 11. Vahdatpour, A., Dabiri, F., Moazemi, M., & Sarrafzadeh, M. (2008). Theoretical bound and practical analysis of connected dominating set in ad hoc and sensor networks. In Proceedings of the 22nd international symposium on distributed computing. 12. Min, M., Du, H., Jia, X., Huang, C. X., Huang, S. C.-H., & Wu, W. (2006). Improving construction for connected dominating set with Steiner tree in wireless sensor networks. Journal of Global Optimization, 35, 111119. 13. Stojmenovic, I. (2002). Handbook of wireless networks and mobile computing. New York: Wiley. 14. Gupta, I., Riordan, D., & Sampalli, S. (2005). Cluster-head election using fuzzy logic for wireless sensor networks. In Proceedings of the communication networks and services research conference. 15. Wirstrm, N. (2006). Optimization of wireless sensor networks using machine learning. Master of Science Thesis, KTH Computer Science and Communication. 16. Barbancho, J., et al. (2007). Using articial intelligence in routing schemes for wireless networks. Computer Communications Magazine. 17. Wang, J., et al. (2008). HOPNET: a hybrid ant colony optimization routing algorithm for mobile ad-hoc networks. Ad Hoc Networks. 18. Su, W., & Bougiouklis, T. C. (2007). Data fusion algorithms in cluster-based wireless sensor networks using fuzzy logic theory. In Proceedings of the 11th WSEAS international conference on communications. 19. OMNeT++ Event Discrete Simulator. http://www.omnetpp.org. 20. Zigbee Alliance, Zigbee Specication (2006). http://www.zigbee. org. 21. Reznik, L. (1997). Fuzzy controllers. Oxford: Newnes Publishing. 22. Jang, J.-S. R., et al. (1997). Neuro-fuzzy and soft computing: a computational approach to learn and machine intelligence. New York: Prentice Hall. 23. Takagi, T., & Sugeno, M. (1985). Fuzzy identication of systems and its applications to modeling and control. IEEE Transactions on Systems, Man, and Cybernetics, SMC-15(1).

A.M. Ortiz et al. 24. Jassbi, J. J., Serra, P. J. A., Ribeiro, R. A., & Donatti, A. (2006). A comparison of Mandani and Sugeno inference systems for space fault detection application. In Proceedings of the automation congress 2006 (WAC06 World). 25. Ran, G., Zhang, H., & Gong, S. (2010). Improving on LEACH protocol of wireless sensor networks using fuzzy logic. Journal of Information and Computational Science, 7(3), 767775. 26. Haider, T., & Yusuf, M. (2009). A fuzzy approach to energy optimized routing for wireless sensor networks. The International Arab Journal of Information Technology, 6(2). 27. Chi, S. H., & Cho, T. H. (2006). Fuzzy logic anomaly detection scheme for directed diffusion based sensor networks. In Lecture notes in articial intelligence (Vol. 4223, pp. 725734). 28. Ortiz, A. M., Olivares, T., Castillo, J. C., Orozco-Barbosa, L., Marron, P. J., & Royo, F. (2010). Intelligent role-based routing for dense wireless sensor networks. In Proceedings of the third joint IFIP wireless and mobile networking conference (WMNC). 29. Heidemann, J., et al. (2001). Building efcient wireless sensor networks with low-level naming. In 18th ACM symposium on operating systems principles. 30. Batuwita, K. B. M. R., & Bandara, G. E. M. D. C. (2005). An online adaptable fuzzy system for ofine handwritten character recognition. In Proceedings of fuzzy logic, soft computing and computational intelligence. 11th world congress of international fuzzy systems association (IFSA 2005). Teresa Olivares was born in Granada, Spain, in 1970. She received the Ph.D. degree in computer science from the University of Castilla-La Mancha, Spain, in 2003. She is an Associate Professor in the Computing Systems Department of the University of Castilla-La Mancha, Albacete, Spain, and teaches computer networks at the Polytechnic School. Her main scientic research interest includes environmental sensor networks, wireless communications and network architecture and protocols for sensor and actor networks. She has very interesting publication in these areas and she participates on interesting sensor-based research projects.

Antonio M. Ortiz was born in Madrid, Spain in 1981. He is a Computer Science Engineer from the University of Castilla-La Mancha. He is currently a Ph.D. Candidate in the Albacete Research Institute of Informatics (University of Castilla-La Mancha). His research interests is centered in the area of wireless sensor networks including self-organization and integration of cross-layer protocols and algorithms. Rule and role-based routing protocols and the incorporation of Articial Intelligence for WSNs are some of his main topics. At present, he is involved in several projects including the specication, and implementation of an intelligent routing algorithm for WSNs and its integration with MAC and physical protocols as well as the experimentation with real motes. Fernando Royo was born in Albacete, Spain in 1983. He holds a Computer Science Engineer degree from the University of Castilla-La Mancha. He is currently a Ph.D. Candidate at the Department of Computing Systems. His main scientic research interests include the design and implementation of MAC protocols for wireless sensor networks, and the specication and design of architectures for wireless sensor networks. At present, he is involved in several projects dealing with power-aware management techniques for wireless sensor nodes. He is student member of the IEEE.

Jose C. Castillo holds a Computer Science Engineer degree from the University of Castilla-La Mancha. He is currently a Ph.D. Candidate in the Albacete Research Institute of Informatics. His research interests include the elds of routing protocols for wireless sensor networks and multisensory intelligents systems for monitoring and interpretation of behaviors. At present, he is involved in several research projects including the combination of different kinds of heterogeneous sensor networks for intelligent surveillance purposes, from the network data acquisition to the detection of objects and the interpretation of their behaviours.

Luis Orozco-Barbosa received the B.Sc. degree in electrical and computer engineering from Universidad Autonoma Metropolitana, Mexico, in 1979, the Diplome dtudes Approfondies from cole Nationale Suprieure dInformatique et de Mathmatiques Appliques (ENSIMAG), France, in 1984 and the Doctorat de lUniversit from Universit Pierre et Marie Curie, France, in 1987, both in computer science. From 1991 to 2002, he was a Faculty member at the School of Information Technology and Engineering (SITE), University of Ottawa, Canada. In 2002, he joined the Department of Computer Engineering at Universidad de Castilla La Mancha (Spain). He has also been appointed Director of the Albacete Research Institute of Informatics, a National Centre of Excellence. He has conducted numerous research projects with the private sector and served as Technical Advisor for the Canadian International Development Agency (CIDA). His current research interests include Internet protocols, network planning, wireless communications, trafc modeling and performance evaluation. He is a member of the IEEE.

Fuzzy-logic based routing for dense wireless sensor networks Pedro J. Marron received his bachelor and masters degree in computer engineering from the University of Michigan in Ann Arbor, USA in 1996 and 1998 respectively. At the end of 1999, he moved to the University of Freiburg in Germany to work on his Ph.D., which he received in 2001. In 2003, he started working on his habilitation at the University of Stuttgart, which he nished in December 2005. In Stuttgart, he lead the mobile data management and sensor network group. Since 2007, he is the head of the Sensor Network and Pervasive Computing group at the University of Bonn. His current research interests are distributed systems, mobile data management, location-aware computing, sensor networks and pervasive systems. He is a member of ACM and GI.