A Compensation-based Reliable Data Delivery for Instant Wireless Sensor Network Yi-Ying ZHANG1, Xi Luo1, Laurence T. Yang2, Lei Shu3, Weiwei Fang4, Myong-Soon PARK1, ∗ 1 Department of Computer Science and Engineering Korea University, Seoul, Korea 2 St. Francis Xavier University, Antigonish, NS, Canada 3 Digital Enterprise Research Institute, National University of Ireland, Galway, Ireland. 4 Sino-German Joint Software Institute 1 {zhangyiying, rosa-xi, myongsp}@ilab.korea.ac.kr, [email protected], [email protected], 4 [email protected] Abstract Instant wireless sensor network (IWSN) is a type of WSN deployed for a class of special applications which have the common requirement on instantly responding for collecting and transmitting sensory data, e.g., volcanic eruption monitoring or nuclear leakage detection. In this paper, having a clusterbased WSN, we present a compensation-based reliable data delivery protocol (CRDD) to collect and transmit sensory data timely, reliably, and energy-efficiently. The CRDD consists of three important parts: 1) reliability calculating mechanism, 2) information classifying mechanism, and 3) intelligent balancing mechanism. By using these three mechanisms, the CRDD can reduce redundant messages for enhancing the transmission performance and compensate the deficient messages for reliability. The simulation results show that CRDD can outperform both LEACH and ECDG and significantly improve sensory data collection speed and system dependability. Keywords Wireless sensor networks, Reliability, Instant, Reliable data transfer.

1. Introduction Wireless sensor networks (WSNs) are undergoing rapid progress and inspiring numerous applications, especially for monitoring some hazardous situations. WSNs can be densely deployed in battlefields, disaster areas and toxic regions to handle various emergencies, and implement battlefield or volcanic eruption monitoring and nuclear leakage detection [4, 5, 6]. These short-lived and subversive applications require ∗

The corresponding author.

collecting sensory data instantly and reliably, because the major characteristics of these applications are: 1) sudden outbreak, such as blast 2) short duration without human intervention, which makes it very difficult to collect the needed information in time. WSN gathering sensory data in these applications is defined as an instant wireless sensor network (IWSN). Due to the ruinous environment, e.g., bombings, sensor nodes can possibly be damaged at any time by any occurrence, and die suddenly, and then the rapid and correct message transmission is an important facet of dependability and quality of service. Hence, an architecture by which the sufficient information can be collected in despite of some nodes’ failure is needed. In other words, the architecture should be fault tolerant. One or some nodes’ death cannot influence the overall task of sensor networks [4, 5]. As mentioned above, the reliability becomes one of the most important issues for IWSN, since sensor nodes are usually deployed in unattended and unfavorable environments, which makes each component of sensor nodes fault or crash easily [20, 21, 22, 23, 24]. The techniques and mechanisms for the operations of sensing, processing, and communication are necessarily aware of this essential fact to maximize the reliability of WSN [7]. In this paper, sensing (detection) reliability indicates the accessibility and validity from sender to receiver. Different from most of previous researches which mainly try to reduce or aggregate the data for energy efficiency of WSN, this paper not only focuses on the aggregation of the packets but also the compensation for the reliability when the packets are deficient. We preferentially consider how to organize WSN to collect information in a real time manner as well as

energy efficiency simultaneously. Since hierarchical wireless sensor networks (HWSNs) [1, 2, 14] have been shown to improve system performance, reduce messages redundancy, and economize energy, in our solution, we present a novel intelligent architecture, compensation-based reliable data delivery architecture (CRDD) which is based on clustering architecture for IWSN by reducing unnecessary or redundant messages as well as compensating insufficient messages for reliability. CRDD achieves good reliability and network performance in terms of transmission latency and energy efficiency. CRDD includes three parts: Credible probability calculation (CPC), Classification Mechanism (CM) and Intelligent Balance Mechanism (IBM). CPC provides the credible probability which draws conclusions about the likelihood of sending a message from sender to receiver correctly, and then CRDD can calculate the accurate probability of transmission from the current cluster head (CH) to base station (BS). Based on the credible probability value, CRDD can distinguish and disseminate the messages simultaneously through classification mechanism (CM). Furthermore, CRDD performs compensation judgment and decides to filter or compensate the messages by using the intelligent balance mechanism (IBM). Triggered by different events, CRDD invokes different mechanisms. We define the frequency of message as the number of same messages being sent per unit time. CJC judges the frequency of message and invokes different components according to the messages saturation which is presented by some threshold. If the messages are saturated, IBM invokes RFC to filter the redundant messages which have the spatiotemporal correlated characteristic; otherwise, it invokes message compensation component to increase message sending frequency to ensure that enough messages can reach the BS correctly. The advantages of our work are as follows: 1) To the best of our knowledge, CRDD architecture is the first scheme that focuses on providing compensationbased reliability for IWSN; 2) CRDD supports a highly efficient credible probability calculation algorithm based on probability theory, which can minimize message transmission while with high reliability. 3) CRDD associates a novel and unique compensation scheme for insufficient message source, which efficiently insures the message reliability. The rest of the paper is organized as follows: In section 2, the related work is briefly reviewed. Section 3 will introduce the system architecture and some notations in the paper. Section 4 will introduce the CRDD architecture in detail firstly, and then describe functions of each mechanism. In section 5, it gives

simulation results. In section 6, we conclude the paper with a summary and directions for the future work.

2. Related Works Various reliability solutions for WSN have been presented in the literature [15, 17, 18, 22]. In [15], it proposes the Flush, a reliable message transport protocol for multi-hop wireless, which advocates directly measuring intra-path interference at each hop and using this information to compute an upper bound on the credible transport probability. During the transfer, flush continually estimates and communicates with the bottleneck bandwidth using a dynamic probability control algorithm. However, the paper mainly focuses on the point-to-point routing and does not consider the energy consumption. In [17, 22], a reliable data dissemination services for wireless embedded devices which are constrained in energy, processing speed, and memory. Sprinkler embeds a virtual grid over the network whereby it can locally compute a connected dominating set of the devices to avoid redundant transmissions, and a transmission schedule to avoid collisions. However, Sprinkler has to waste more energy to establish a regular grid structure. In [18], a sender initiated path switching algorithm is presented, by which it enables the immediate sender to change the packet's route dynamically. The overall effect of path switching on the survivability is analyzed as a measure of reliable event delivery. The model is used to predict the maximum network lifetime in terms of total transmitted messages. But it wastes more energy during establishing the path, and does not deal with the redundant messages. Although these works can improve the reliability of WSN, all of them just focus on reliability in normal status and do not consider protection of dependability when the sensor network is damaged partially.

3. System Model and Notations 3.1. Network Models and Assumptions For improving the performance especially in energy consumption, the hierarchical architectures are very suitable for wireless sensor network. In [1], LEACH (an energy efficient adaptive clustering protocol) is presented. This organization-efficient communication protocol employs a clustering scheme, in which the nodes organize themselves into clusters using a distributed algorithm. Each cluster is controlled by a CH which collects and aggregates the messages from

the cluster members and forwards information to the BS. In [2], a clustering protocol, ECDG (Efficient Clustering of Data Gathering protocol) is proposed. The protocol groups sensor nodes as clusters, builds a routing tree among evenly distributed CHs of the clusters and only lets the root node communicate with the base station directly. Comparing with homogenous LEACH, ECDG applies a round-by-round operation scheme to establish the cluster head (CH), whereas it uses a routing tree to route messages to BS by multihop as shown in Figure 1.

randomized rotation of the high-energy consuming cluster head position among the sensor nodes to avoid draining the battery of any sensor in the network.

3.2. Notations Some notations used in this paper are shown in table 1: Table 1: The notations the amount of packets received correctly by Ar receiver, e.g. BS. Ci the ith cluster in WSNs. Ai the amount of packets sent from Ci to BS Hij the hop count from node i to node j. Hi the hops count from CHi to base station. Pe the default probability of error in WSN [13]. Pij the correct probability between node i and j. Pt the correct probability from CH to base station.

4. CRDD Architecture

Figure. 1 LEACH and ECDG.

Considering the applicability for a large scale WSN, CRDD architecture is deployed based on a protocol similar to ECDG [2]. We assume a sensor network consisting of a BS and numerous sensor nodes which self-organize into clusters. Each member in the cluster is supervised by its CH which can broadcast messages to all sensors in the cluster. We assume that the sensors and CHs are stationary after deployment. Each CH is assumed to be reachable to all sensors in its cluster, either directly or through multi-hop, as shown in Fig.1. Cluster members perform two main functions: sensing and relaying. The sensing component is responsible for monitoring the environment around them. The collected data then can be relayed to CHs. In order to save constrained energy, the communication among cluster members or between cluster members and CHs is only realized via shorthaul radio communication. The CHs, after receiving the sensed data from their members, fuse and process the data to extract relevant information and transmit it to the BS via long-haul transmission. However, being a CH node is much more energy intensive than being a non-cluster head node because it needs to perform more energy consuming tasks. If the cluster heads are chosen priori and will never change throughout the network lifetime, these nodes will quickly die for running out of energy. Once the cluster head runs out of energy, it is no longer operational, and all the nodes that belong to this cluster will lose communication ability. So we assume that the network incorporates

In this section, we introduce the main construction of CRDD architecture and its mechanisms. As illustrated in Figure 2, CRDD consists of three main mechanisms: credible probability calculation mechanism, classification mechanism and intelligent balance mechanism.

Figure. 2 The Architecture of CRDD

During the generation process of cluster head, the CRDD architecture will be established in CH synchronously. Before dealing with the messages, CRDD firstly calculates the accurate probability of transmission from the current CH to BS using probability estimation algorithm. Thereafter, the mechanism starts to handle the messages. This process is divided into two phases: classification phase and intelligent balance phase. In the first phase, CRDD distinguishes and disseminates the messages simultaneously through classification mechanism (CM). Then in the second phase, CRDD performs compensation judgment and decides to filter or compensate the messages by using the intelligent

balance mechanism (IBM). CJC judges the frequency of message and invokes different components according to the messages saturation.

Ai =

1 = Pt

1 H i −1

∏P j =1

4.1. Credible probability calculation (CPC) Due to the inherent error-prone characteristic of WSN, the reliability cannot be provided predominantly by the natural network. It is necessary to estimate the probability that the messages can arrive at BS correctly. And then, the CRDD decides the amount of messages which are sent to BS successfully. According to [2], ECDG consists of many clusters, and when the CHs send the messages to BS, most of them send messages through multiple CHs as well as multi hops. Through each hop, messages will be sent in accordance with the same credible probability. In this section, we present an approach to calculate the credible probability from CH to BS. We assume that WSN is set with default probability of error message. For predigesting the computation, we can use the formulas (1), (2) to get the relationship between the amount of messages and probability as well as the probability between two nodes:

(1)

Ar = Ai (1-Pe)

According to descriptions above, we can get the amount of hops for every Ci to base station in the early stages of cluster formation. The amount of hops is not fixed but depends upon the spatial location of sensors [9], we call it as Hi. Without loss of generality, we can get that the correct probability Pt as follows: H i −1

(2)

Pt = ∏ Pj , j +1 j =1

Firstly, it can set the default value of Pj,j+1, i.e. Pj,j+1 = 1- Pe, then equation (2) can be expressed as equation (3) as follows:

Pt = (1 − Pe ) H i −1

(3)

Under the Taylor series theory, the correct probability is about 1-Pe(Hi -1), in order to calculate simply, namely Pt ≈1-Pe(Hi -1) [12]. In accordance with the probability, we assume that Ci at least sends Ai messages to BS for the reliability. And Ar should follow the credible probability as equation (4): H i −1

Ar = Ai Pt = Ai ∏ Pj , j +1

(4)

j =1

Therefore, if CH wants at least 1 message, that is, Ar = 1, can arrive at BS, Ai should be as follows:

j , j +1

≈

⎤ ⎡ 1 1 ≈⎢ ⎥ H i −1 (1 − Pe ) ⎢1 − ( Pe H i − 1) ⎥

(5)

By (5), it can guarantee at least one message arrives at the base station successfully. After sending sufficient number of messages, e.g. Ai, CH informs member nodes retard sending message until a new cycle starts. Based on CPC, CRDD can prune many excessive messages and reduce the communication overhead as well. The status tables in the nodes record the trails of probing and recalling messages, and also provide a way to suppress redundant retransmissions. Under different circumstances, for ensuring the reliability of the information, it can appropriately adjust these duplicate messages in accordance with the percentage. Moreover, with WSN running, the power of nodes gradually decline while the error probability is increasing. For handling this situation, the base station can judge it according to the amount of messages and adjust the probability parameter.

4.2. Classification Mechanism The classification mechanism (CM) is located in the front part of the processing messages and responsible for message acquisition and classification. Without loss of generality, messages are typically divided into three categories: (1) Generated by the CH itself; (2) Generated by other non-CH nodes in the same cluster; (3) Forwarded from other CHs. Each node has a unique ID and registers its ID into CH. Thus, CH can distinguish different type of messages using the members ID table. In order to reduce network traffic latency and save energy, for the third type messages, our design is to forward them towards base station directly, because these messages from other clusters are usually important and already filtered by CH that transfers them. We need not to consume the precious energy and computation resource to deal with them any more. But for the first two types of messages generated by the non-CH nodes, we show how to build nearly optimal aggregation structures that can further deal with network reliability and compensation for duplicate data by exploiting probabilistic techniques in the next mechanisms.

4.3. Intelligent Balance Mechanism The intelligent balance mechanism (IBM) contains three components for balancing amount of sending messages: compensation judgment component (CJC),

redundancy filter component (RFC) and message compensation component (MCC). 4.3.1. Compensation Judgment Component (CJC). CJC responds for judging whether it needs compensation or not. We design a list named LogList to preserve the various latest sending messages, and we can distinguish and transmit messages by it. LogList contains an information table structured as . In the triple, MessageType is representative of the type of messages; BoolStatus is a boolean flag for message sent saturation status. If the value of BoolStatus is true, it indicates that this type of messages has been sent enough, similar messages should be staved to send temporarily. RFC then sends the saturation feedback to CJC, which decides whether invoke the messages compensation component or not. Frequency field is a symbol of the amount of a certain type of messages. When it exceeds a certain value (MAX: the amount of messages that we have to send at least for reliability under Pt), we insure that the messages must be sent enough. We set the BoolStatus value as TRUE, and update the LogList simultaneously. The algorithm is shown in Table.1. Table 1: The algorithm of CJC 01 Get new Message Mnew; 02 Transmit Mnew to BS via RFC directly; 03 If get the message insufficient feedback Then{ 04 Invoke compensation function; } 05 End

Furthermore, before CJC triages messages, it has to compare new message with history records in LogList to classify the type of new message. Based on the facts that nodes in sensor networks often encounter spatiotemporal correlation [1], we can assume that the messages generated in the same cluster normally are similar in a period, which means that in normal cases the probability that there are different messages is very small. Furthermore, the LogList needs just a small memory space and will be refreshed in next turn. Simplistically, let D = {(Mi, Wi)| (M1, W1); (M2, W2),…,(Mn, Wn)} denotes the latest records in LogList. And D is a dynamic set arranged according to the FIFO principle. Where Mi is the feature of MessageType, and Wi is the weight (Frequency) of Mi, 1 ≤ i ≤ n. When the new message Mnew coming from the cluster member arrives at CH, the CJC mechanism authenticates it with a similarity as below. V ( M new ) • V ( M i ) (6) sim ( M new , M i ) = | V ( M new ) | × | V ( M i ) | According to a frequency-based method based Kmeans algorithm [16], which refines a K-modes-CGC

algorithm using small memory and sample set, we can get (6) for simplicity and explanation. Where Mi ∈ D. The function sim(Mnew, Mi) denotes the similarity between Mnew and Mi. Mnew traverses the whole set D, if it matches with a Mi, that is, the similarity is not less then a threshold, which indicates the new message is a sent message, and increase the Wi value and go to next mechanism. Otherwise, it means Mnew is a new type message, and it will be registered into LogList and update it. Table 2: The algorithm for matching message 00 Get new Mnew; 01 for i = 1 to D.Length{ 02 if sim(Mnew, Mi) ≤ Threshod Then { 03 Mi.Frequency++; 04 Update LogList; 05 Break; // Match correctly 06 }} 07 if i≥ D.Length then{ 08 Mnew is new type message; Register Mnew into LogList; Update LogList; //Match correctly and register Mnew } 10 End

4.3.2. Redundancy Filter Component (RFC). RFC focuses on dealing with the redundant messages. Transferring sensed messages in an energy-efficient manner is critical in operating the sensor network. It is well known that, for nearly all node technologies, data transmission over wireless links consumes hundreds to thousands of times more energy than performing local computation on the same data [11]. Practically, when an event happens, most of the nodes in the same cluster detect and try to send the messages which almost contain the same content. The possibility that the individual node simultaneously sends different messages is very small as mention above. Therefore, it is not necessary for all nodes which detected the event to sent message to base station. That is, we should aggregate the messages and just send a certain amount of messages for sufficing the reliability. According to probability calculation, during a certain period, if some events happen, each CH which detected them sends at least MAX messages to ensure the messages can arrive at BS credibly. And the MAX is different for different CHs for different hops to BS. For protecting the reliability, the BS can send the query requirement to CHs to request them stay or increase the transmission. After having sent sufficient messages, the CH does not compensate in next period, even the messages are insufficient again, because the BS has gotten the reliable information. At the same time it avoids the unnecessary and virulent

compensation. The algorithm is illustrated in Table 3. After that, the CHs suspend the message transmission. Table 3: The algorithm of RFC 00 Get Mnew from CJC; 01 If Mnew∈LogList and Frequency >= MAX Then { 02 Count the amount of the type of Mnew; 03 Filter redundant messages during a time period; 04 Update the LogList ; 05 } 06 Else { 07 Transmit Mnew to BS directly; 08 Update the LogList ; 09 } 10 If Sent message insufficient Then { 11 Send compensation requirement to CJC 12 Invoke MCC; 13 } 14 End

4.3.3. Message Compensation Component (MCC). Due to the inclement environment for IWSN, the network often suffers from many different degrees of damage. If the explosion, such as volcanic lava eruption, corrosive liquids erosion and so on, destroyed some sensor nodes, it will cause information collection insufficient. Thus, MCC is the solution. As shown in the figure 3, sensor network consists of 6 clusters, cluster A~F, in the explosive situation. The clusters A, D, E and F have been damaged in varying degrees, and especially clusters A, E are dynamited severely as well as the cluster F is damaged lightly. Once the CH detects the breach, it will check its cluster members to make sure the damage. E.g. in the cluster E, almost 60% nodes dead for the blasting, which causes the messages decrease dramatically in a short time, CH will decide to invoke MCC to reinforce messages for dependency. On the other hand, if the CH was demolished, the rest nodes will rebuild the cluster or join neighbor cluster. On contrast, the breakage of cluster F is not severe, which can not cause the message lack.

Figure. 3 The WSN in blasted area. A~F are clusters. And the CHs forward messages along the blue lines to BS.

According to the description above, when CH detects that the messages are inadequate by RFC, the

CJC invokes the compensation component to increase sending relevant information until achieving MAX messages for credibility. For achieving the reliability, MCC will be invoked in the follow situations: 1.the cluster is damaged in large proportion which causes the message insufficient; 2.the cluster catches transient events which happen and end within a certain period. At the same time, CH sends insufficient messages to BS. The algorithm is shown in table 4. Table 4: The algorithm of MCC 00 Get compensation command from CJC; 01 If Mnew∈LogList and Frequency < MAX Then { 02 Sent duplicate of Messagex; 03 Update LogList 04 } 05 If Mnew.Frequency < MAX Then goto 01; 06 End

5. Simulation and Results In this section, we evaluate the performance of CRDD architecture implemented in NS2. According to the requirement of the CRDD, we designed a sensor network simulation incorporating ECDG – essentially a multi-hop hierarchical sensor network as the foundational environment. In order to explain the relations among CRDD, ECDG and LEACH, we run each kind of simulation in these three different scenes. For increasing the comparability and feasibility, the parameters are the same as LEACH and ECDG [1]. The parameters for the simulations are shown in table 5. Table 5: Simulation parameters

Parameter

Value

Area Size Quantity of sensor BS Position Initial Energy Cluster Radius Packet size Eelec

100*100 100 (50,250) 2J 40m 500 Bytes 50 nJ/bit 10 pJ/bit/m2

εfs εamp

0.0013 pJ/bit/m4

EDA d0

5 nJ/bit/signal 86.2m

In table 5, the Eelec is for running the transmitter or receiver circuitry; theε amp is for the transmit amplifier. We use network’s lifetime as the evaluation criteria. Network’s lifetime means the period from the deployment of the network to the death of all the nodes.

Figure. 4 The amount of error messages As shown in Figure 4, our CRDD architecture performs better than ECDG and LEACH under the same environment. Having sent the same amount of messages, CRDD generates the least amount of error messages. The error message amount generated by CRDD is about 3 times less than LEACH’s and about 5 times less than ECDG’s. This indicates that CRDD can achieve much higher reliability and efficiency in energy consumption. On the other hand, because the correct messages and error messages in direct proportion, the CRDD sends fewer messages than LEACH and ECDG too, which predicates the CRDD can get information faster. We can collect the useful information for IWSN instantly.

In this paper, we present a novel, scalable and intelligent architecture, a compensation-based reliable agent architecture (CRDD) for wireless sensor network. In addition to reducing the amount of messages transmitted in the sensor network as well as compensating the insufficient messages, CRDD can also improve the reliability of the sensor network. The simulation shows, with CRDD architecture, WSN can perform much better than LEACH and ECDG in terms of energy-efficiency and reliability. One limitation is that our system is only implemented in the CH but not in other nodes of the sensor network. As a result, in the near future we are going to work on improving the ability of message processing in base station and make overall sensor network reflect faster.

Acknowledgement Mr. Lei Shu’s work in this paper was supported by: (in part) the Lion project supported by Science Foundation Ireland under grant no. SFI/02/CE1/I131, (in part) by the European project CONET (Cooperating Objects NETwork of excellence) under grant no. 224053.

References [1] S. Bandyopadhyay and E. J. Coyle, “An Energy Efficient Hierarchical Clustering Algorithm for Wireless Sensor Networks”, in Proceeding of IEEE INFOCOM’03, San Francisco, April 2003. [2] Zhen Fu, Yuan Yang and Myong-Soon Park, "Efficient Clustering of Data Gathering in Wireless Sensor Networks", The 8th International Conference on Electronics, Information, and Communication (ICEIC 2006), pp.351-354, Ulaanbaatar, Mongolia, June, 2006 [3] Tynan, R. Marsh, D. O'Kane, D. O'Hare, G.M.P., "Agents for wireless sensor network power management", in Parallel Processing, 2005. ICPP 2005 Workshops. International Conference,pp.413- 418,June 2005

Figure. 5 Deformity VS Reliability Figure.5 shows that our CRDD can compensate the insufficient messages in the WSN deformity period. And the reliability of WSN is still supported by CRDD. On the contrary, the LEACH and ECDG can not support more messages at the same situation. With the increasing deformity, the network has become more discredited, which shows that our CRDD has higher information reliability as well as the situation with sufficient messages.

6. Conclusions and Future work

[4] David Culler, Deborah Estrin, and Mani Srivastava. “Overview of sensor networks”, IEEE Computer, Special Issue in Sensor Networks, Aug. 2004. [5] N. Heo, P.K. Varshney, An intelligent deployment and clustering algorithm for a distributed mobile sensor network, IEEE International Conference on Systems, Man and Cybernetics, October 2003 [6] Akyildiz LF, Su WL, Sankarasubramaniam Y, Cayirci E. A survey on sensor networks. IEEE Communications Magazine, 40(8):102~114,2002. [7] Jie Yin,Qiang Yang,” Integrating hidden Markov models and spectral analysis for sensory time series clustering",in Data Mining, Fifth IEEE International Conference,Nov. 2005 [8] Ronan Mac Ruairí and Mark T. Keane, “An EnergyEfficient Multi-Agent Sensor Network for Detecting Diffuse

Events”, International Joint Conference on Artificial Intelligence (IJCAI’07), Hyderabad, India, 2007 [9] Navin Kumar Sharma,Mukesh Kumar,"An Energy Constrained Multi-hop Clustering Algorithm for Wireless Sensor Networks", International Conference on Networking (ICN'05), LNCS 3420, pp. 706 – 713, 2005. [10] J. Hill, R. Szewczyk, A. Woo, S. Hollar, D. Culler, and K. Pister, “System architecture directions for networked sensors”, Proc. ACM ASPLOS IX, Nov. 2000. [11] Kyle Jamieson, Hari Balakrishnan, Y. C. Tay “Sift: a MAC Protocol for Event-Driven Wireless Sensor Networks”, Proceedings of the Third European Workshop on Wireless Sensor Networks (EWSN), pp. 260-275, February 2006. [12] http://en.wikipedia.org/wiki/Taylor_series [13] Zude Zhou, Sheng Wang, Quan Liu, "Local Hop-Count Probability Grid: An Improvement Nodes Localization Scheme in WSN," icicic, pp. 64-67, First International Conference on Innovative Computing, Information and Control - Volume I (ICICIC'06), 2006

[19] Chang, Y.-S., Hsu, M.-T., Liu, H.-H., Juang, T.-Y.: Dependable Geographical Routing on Wireless Sensor Networks. In: LNCS, vol. 4523 (2007) [20] Chang, Y.-S., Juang, T.-Y., Lo, C.-J., Hsu, M.-T., Huang, J.-H.: Fault estimation and fault map construction in Cluster-based Wireless. In: The IEEE International Workshop on Ad Hoc and Ubiquitous Computing (AHUC 2006), Taichung, Taiwan, June 5–7 (2006) [21] Clouqueur, T., Saluja, K.K., Ramanathan, P.: Fault Tolerance in Collaborative Sensor Networks for Target Detection. IEEE Trans. on Computers 53(3), 320–333 (2004) [22] Hsu, M.-T., Lin, F.Y.-S., Chang, Y.-S., Juang, T.-Y.: The Fault Probability Estimation and Decision Reliability Improvement in WSNs. In: Proceedings of the IEEE Region 10 Annual International Conference (TENCON 2007), October 30–November 2 Taipei, Taiwan (2007) [23] Sun, T., Chen, L.-J., Han, C.-C., Gerla, M.: Reliable Sensor Networks for Planet Exploration. In: The 2005 IEEE International Conference On Networking, Sensing and Control (ICNSC 2005) (2005)

[14] M. Ye, C. F. Li, G. H. Chen, and J. Wu, “EECS: An Energy Efficient Clustering Scheme in Wireless Sensor Networks”, in Proceedings of IEEE Int’l Performance Computing and Communications Conference (IPCCC), 2005, pp. 535-540.

[24] Ye, F., Zhong, G., Cheng, J., Lu, S., Zhang, L.: PEAS: A Robust Energy Conserving Protocol for Long-lived Sensor Networks. In: Proceedings of the 23rd International Conference on Distributed Computing Systems (ICDCS 2003) (2003)

[15] Sukun Kim, Rodrigo Fonseca, Prabal Dutta, Arsalan Tavakoli, David Culler, Philip Levis, Scott Shenker, and Ion Stoica. "Flush: A Reliable Bulk Transport Protocol for Multihop Wireless Networks." In Proceedings of the Fifth ACM Conference on Embedded Networked Sensor Systems (SenSys), 2007.

[25] Zhang, H., Hou, J.C.: Maintaining Sensing Coverage and Connectivity in Large Sensor Networks. Wireless Ad Hoc and Sensor Networks: An International Journal 1(1–2), 89–123 (2005)

[16] Sun Y, Zhu QM, Chen ZX. An iterative initial-points refinement algorithm for categorical data clustering. Pattern Recognition Letters, 2002,23(7):875-884. [17] V. Naik, A. Arora, P. Sinha, and H. Zhang. Sprinkler: A reliable and energy efficient data dissemination service for wireless embedded devices. In Proceedings of the 26th IEEE Real-Time Systems Symposium (RTSS 2005), 2005. [18] Sebnem BAYDERE, SAFKAN, Ozlem DURMAZ, "Lifetime Analysis of Reliable Wireless Sensor Networks",JIEICE Trans Commun,VOL.E88-B(6),24652472,2005

[26] S. Ganeriwal, R. Kumar, and M. Srivastava, “Timing Sync Protocol for Sensor Networks”, The First Int. Conf. on Embedded Networked Sensor Systems, Los Angeles, California, November 2003. [27] J. Elson, L. Girod and D. Estrin, “Fine-Grained Network Time Synchronization using Reference Broadcasts”, May 17, 2002, University of California, Los Angeles. [28] H. Dai, R. Han, “TSync : A Lightweight Bidirectional Time Synchronization Service for Wireless Sensor Networks”, ACM SIGMOBILE Mobile Computing and Communications Review, Special Issue on Wireless PAN & Sensor Networks, vol. 8, no. 1, University of Colorado, Boulder, January 2004, pp. 125-139.