[ieee 21st international conference on advanced networking and applications - niagara falls, on,...

Agent Based Approach to Minimize Energy Consumption for Border Nodes in Wireless Sensor Network

Elhadi Shakshuki, Haroon Malik

Jodrey School of Computer Science Acadia University

Wolfville, Nova Scotia, Canada B4P 2R6 {Elhadi.Shakshuki; 078337m}@acadiau.ca

Abstract

This paper presents an agent-based system to minimize the energy consumption for border nodes in Sensor-MAC (S-MAC), a Cluster based contention protocol. The S-MAC protocol is based on unique feature; it conserves battery power at nodes by powering off nodes that are not actively transmitting or receiving packets. In doing so, nodes also turn off their radios. Inspired by the energy conservation mechanism of the S-MAC, we unmitigated our efforts to augment the node life time in sensor network. Border nodes act as shared nodes between virtual clusters. Virtual clusters are formed on the basis of sleep/listen schedule of nodes. Towards this end, we propose a multi-agent system that allows nodes to join cluster where they experience minimum energy drain. This system includes two types of agents: stationary and mobile agents. A prototype implementation and simulation results compared with S-MAC are presented. 1. Introduction

Wireless sensor networks (WSN) enable pervasive, ubiquitous, and seamless communication with the physical world. A few common applications are military, security, habitat monitoring, industrial automation, and agriculture [1]. WSN comprises numerous sensor devices, commonly known as motes, which can contain several sensors to monitor the physical entities such as temperature, light, motion, metallic objects, and humidity [16].

This paper presents our approach to the expansion of Sensor-Medium Access (S-MAC) protocol [13]. This is a cluster based contention protocol to minimize energy consumption among Border Nodes (BN). The S-MAC protocol is based on unique feature; it conserves battery power by powering off nodes that are not actively transmitting or receiving packets. In doing so, nodes also turn off their radios. The manner in which nodes power themselves off does not influence any delay or throughput characteristics of the protocol. Inspired by the energy conservation mechanism of the S-MAC, we unmitigated our efforts to augment the node life time in sensor network. In WSN, border node acts as shared node between two or more virtual clusters. Virtual

clusters are formed on the basis of sleep-listen schedule of nodes. Border nodes follow the schedule of one or more virtual clusters, hence consume more energy due to double duty load. In our proposed approach, border nodes are allowed to intelligently switch between virtual clusters where they experience minimum energy drain. Towards this end, we proposed a multi-agent system at each node. This system includes two types of agents, namely: stationary agent and mobile agent. The stationary agent is a static agent that has the capability of monitoring the changes of its energy consumption for a predefined period of time. The mobile agent is able to travel and interact with stationary agents at neighbouring clusters and benefit from their acquired knowledge. 2. MAC: Back Ground

The characteristics of WSNs differ from traditional wireless networks in several ways [1]. Firstly, sensor networks consist of a number of nodes and have high network density that competes for the same channel. Secondly, most nodes in sensor networks are battery powered, and it is often very difficult to change their batteries. Thirdly, nodes are often deployed in an ad hoc fashion rather than with careful pre-planning; thus they are self-organized to form communication network. Fourthly, sensor networks are adaptable to local failures. Fifthly, broadcasting is the main mode of communication in sensor networks and this may cause channel contentions. Finally, most traffic in the network is triggered by sensing events, and it can be extreme at times. These characteristics of WSNs suggest that traditional MAC protocols are not suitable for wireless sensor networks without modifications. The basic idea of these protocols is to avoid interference by scheduling nodes onto different sub-channels that are divided either by time, frequency or orthogonal codes. Since these sub-channels do not interfere with each other, MAC protocols in this group are largely collision-free. These protocols are called scheduled MAC protocols, with the exception of contention-based protocol. Contention-based MAC protocols are based on channel contention. Rather than pre-allocating transmissions, nodes compete for a shared channel resulting in probabilistic coordination. Collision happens during the contention procedure in such systems. Classical examples of

21st International Conference on Advanced Networking and Applications(AINA'07)0-7695-2846-5/07 $20.00 © 2007

contention-based MAC protocols are Carrier Sense Multiple Access (CSMA) [7] and ALOHA [10]. In CSMA, for example, a node listens to the channel before transmitting. If it detects a busy channel, it delays access and retries to transmit later. The CSMA protocol has been widely studied and extended, and became the basis of several widely used standards, including IEEE 802.11 [7].

TDMA based protocols are effective at avoiding collisions and have a built-in duty cycle extenuating idle listening. In contract, contention-based protocols simplifies the activities and do not require any dedicated access point in a cluster. MACAW is an example of contention-based protocol [9]. MACAW protocol is widely used in WSN and in ad hoc networks, due to its simplicity and robustness to hidden terminal problem. Another example of contention-based protocol is IEEE 802.11 Distributed Coordination Function (DCF) [7], mainly built on a MACAW protocol. Most of the research work on sensor MAC protocols showed that energy consumption is very high when nodes are in idle mode, because of idle listening [13]. Power Aware Multi-access Protocol (PAMAS) provided an improvement by avoiding the overhearing among neighbouring nodes [10].

S-MAC is a slotted-based MAC protocol specifically designed for wireless sensor networks. S-MAC is built on contention-based protocols similar to IEEE 802.11 [13]. This protocol retains the flexibility of contention-based protocols while improving energy efficiency in multi-hop networks. It implements an approach to reduce energy consumption from all the major factors, including idle listening, collision, overhearing and control overhead. 3. Related Work

The operation of wireless network depends, to a large extent, on the effectiveness of the low-level Medium Access Control (MAC) layer. MAC in WSN aims to ensure that no two nodes are interfering with each other’s transmissions. The S-MAC contention-based protocol addresses not only the transmission interfering issues, but also intends to minimize the protocol-overhead, overhearing and idle-listening. Its principle is based on locally managed synchronizations and periodic sleep-listen schedules. Neighbouring nodes form virtual clusters based on sleep schedules. If two neighbouring nodes reside in two different virtual clusters, they should be in duty at listen periods of both clusters. Nodes synchronize their schedules by exchanging Synchronization (SYNC) packets. A node broadcasts the SYNC packet that contains its planned sleep time. The time allotted for sending the SYNC packet is called synchronization period.

In S-MAC, the communication between nodes is typically achieved by exchanging packets that start with Carrier Sense (CS) to avoid collision. This is followed by Ready to Send and Clear to Send (RTS/CTS) packets

which are unicasted to win the communication media. Upon success, nodes start transmission of the desired data. Since all immediate nodes have their own sleep schedules, periodic sleep may result in high latency especially in multi-hop routing algorithm. The latency caused by periodic sleeping is called sleep delay. An adaptive listening technique is proposed to improve the sleep delay, and thus the overall latency [13]. In their approach, a node who overhears its neighbour’s transmissions listens for a short time at the end of the transmission. Hence, if a node is the next-hop node, its neighbour could pass data immediately. The end of the transmissions is known by the duration field of RTS/CTS packets. The energy waste caused by idle listening is reduced by sleep schedules. Broadcast data packets do not use RTS/CTS, which increases the probability of collision.

One of the major problems of S-MAC is the possibility of a node to follow two or more different schedules. It goes to listen-state frequently in order to relay packets from one virtual cluster to another. Consequently, it drains more battery power. Since energy consumption is a major issue in wireless sensor nodes, there have been many attempts to over come this problem. Ye and Heidemann [13] proposed an approach where a border node adapts the first received schedule. A node can still communicate with neighbours in other clusters using its scheduled table, which contains its neighbouring nodes’ schedules. In this approach, the cost of transmitting more than once to different clusters is overlooked. In another approach, a boarder node selects one schedule that belongs to the higher synchronizer node [15]. The boarder node creates and unicastes a unifying packet, which contains the highest synchronizer’ schedule to be used as a target schedule. As soon as the neighbour node in the respective virtual cluster receives this unifying packet, it becomes the boarder node. This neighbour will attempt to achieve its non-boarder status by further transmitting the unifying packet to its immediate neighbours. This process repeats for each neighbour receiving the unifying packet, until there are no more boarder nodes left in the cluster. The drawback of this approach, it forms small virtual clusters into one large virtual cluster with unified listen and sleep cycle at the trade of missed events.

We attempt to supplement our effort by using an agent-based approach, which allows a border node to adapt the schedule of its neighbouring nodes. The main objective of these agents is to help border nodes to move to virtual clusters that experience minimum energy drain. Each node is equipped with Stationary Agent (SA) that closely examines the energy consumption in sensor node, and a Mobile Agent (MA) that frequently visits the border node’s neighbours, with the ability to clone itself. The mobile agent analyzes its neighbour node’s energy model in a predefined period, and then reports back to border node’s stationary agent (i.e. the source agent). The stationary agent compares between the boarder node local energy model and others’ energy model reported by


MA to predict the energy efficient cluster. The virtual clusters may span over different geographical regions. The border node then can shift itself to virtual cluster and become non-border node for amount of time until it finds more energy efficient cluster. The mobile agent frequently visits the host neighbour to query its energy model and reports to the stationary agent with the updated energy model.

4. System Architecture

This section discusses our proposed agent-based approach for wireless sensor medium access protocol. The main aim of this approach is to address the problem energy conservation of border nodes in S-MAC. This problem arises when nodes try to adapt different schedules. If Radio Frequency (RF) signal of a node A with particular scheduler overlaps with a signal of another node B with different schedule; thus node A might adapt node B’ schedule. If a node adapts the schedule of two or more neighbouring nodes’ schedules, it is called a Border Node (BN), as shown in Figure 1.

Border nodes lose more energy as they have to listen for longer time than non-boarder nodes in a cluster. They listen with the listening schedule of nodes in both clusters, which causes more energy consumption. This raises the threat of minimizing their life time in sensor network.

��

A node can be restricted to adapt to one schedule only

of its neighbours [13][15]. This will stop a node from becoming a border node. Using these approaches, a node will remain in only one cluster for all its life time until it expires. Due to the nature of WSN, sensing in a particular region may increase or decrease over period of time. This change will effect the energy consumption of the virtual cluster in that region. Keeping in view of this novel fact, our agent-based approach helps a boarder node to find the energy efficient cluster. The agents provide a boarder node with the capability to move from one cluster to another during its life time. 5. MA and SA: Process flow

Figure 2 demonstrates agents’ flow control to reduce the energy consumption of the boarder nodes

�� The agent selects the first schedule from the list in its

schedule table. It will use this schedule to broadcast as a SYNC packet. MA interacts with SA agent on host node and reports back its energy model. Once all the energy models are reported back via MA, SA searches for the most energy efficient neighbour. If it finds one, it will move to it and adapts to its schedule. The SA will mark its old schedule as inactive; and it uses the new schedule for future broadcasts.

Wait for MA to return

Broadcast Schedule SYNC packet

Listen for SYNC packet

MA starts building Energy

Found Schedule

?

Node = Synchronizer

MA fires Timer to visit neighbouring nodes

Received ACK of

all neighbour

s

Formation of boarder node with 2 or more

schedules

SA schedule = first schedule received

MA clone itself

Ye

Yes

Ye

Start

Node = Follower

ACK Received

Increase retry counter by 1

No

MA reports energy model

SA updates energy parameters of Others

Model

All Cloned MA are back?

Energy Model

Selection? SA found the best

energy model

Yes

No

Yes

Comparison: SA does analytic comparison of Energy Models

1- Update Schedule Table 2- Energy efficient

neighbour Schedule marked Active

3- Other schedule marked Inactive

Schedule-1

Schedule-2

Virtual Cluster -1

Virtual Cluster-2

Border nodes


6. Agent’s Architecture The architecture of the proposed agents is based on

the agent model described in [12]. Each agent posses the basic structure, as shown in Figure 3.

��

The agent architecture consists of knowledge components and executable components. The knowledge component contains the information about the WSN environment such as the number of clusters, neighbour nodes, goals that need to be satisfied and possible solutions generated to satisfy a goal. The learning component provides the agent with the capability to learn; it uses the monitored observations stored in its knowledge to predicate the energy efficient cluster. The scheduler component provides the agent with a time agenda to start and stop certain activities such as monitoring and mobility. The communication component allows the agent to exchange messages with other agents and with event occurring in a node. The two proposed agents (SA and MA) are the subject of the following two subsections.

6.1. Stationary Agent

Each sensor node is equipped with a stationary agent. This agent monitors and records node’s activities (transceiver and inter-processing) and correspondingly updates its energy model in its knowledge, shown in Figure 4. The agent maintains the node energy model in a table format, as shown in Table 1.

This table contains a list of node activities and its corresponding energy consumption for a predefined time. . Predefined time refers to the amount of time SA will hold the table in its knowledge, and it is user defined.

Table 1. Node energy model Seq Node Id Activity Energy Time

Stamp 1 01 RTS 36mW. 12073201 2 03 CTS 14.4mW 12073109

3 01 Data 45mW 12073200

4 03 ACK 14.4mW 12073200

5 B SYNC 14.4mW 12073109

SA purges all the entries when this time expires and it

starts building new model.

��

To avoid degrading protocol performance, a

predefined time parameter should always be set at least 3 to 4 times greater than MA’s visit interval parameter. This parameter ensures that neighbouring MA’s can get fair chance to visit the node at least once, before SA purges the energy model. The SA frequently updates it knowledge by updating others models. When MA reports the energy model of its neighbour, SA compares it from the previously reported. If the new energy model is better than the old model, then old model is replaced by the new one. Once all the energy models have been reported for cloned instance of MA, SA makes analytical comparison between its local and reported schedules. If it finds a neighbour with better energy model, it will adapt its schedule by setting its sleep timer to that of neighbouring node next sleep. It will also mark this neighbour schedule as active, A, in others models, as shown in Figure 4. This adapted schedule will be used for future SYNC packet broadcast.

Tim

er Knowledge

Information and control flow Knowledge Process

Goals Scheduler

Communication

Local model

Problem Solver

Solutions

Learning Knowledge Update

Other’s models Schedule

Table

Monitoring

Energy Model

Radio (external) Internal


6.2. Mobile Agent

Mobile agent resides on every node in wireless sensor network and collaborates with SA to satisfy the assigned goal. MA has the capability to clone itself and to travel to neighbouring nodes. It benefits from knowledge acquired from SA that resides in the visited node. In addition to mobility, it is also able to learn about other nodes schedules. This makes MA scalable. As energy is vital issue in WSN, MA helps to learn the best energy efficient cluster for the border node. It should be noted that the MA continues building its knowledge by periodically visiting neighbour nodes, even when an energy efficient cluster is found.

Figure 5 shows nine sensor nodes marked with their node-Ids from 1 to 9, and scattered among virtual clusters 1 and 2.

��

The nodes 1, 2, 3 and 4 in Virtual cluster-1 have

different sleep-listen schedule than that of nodes 5, 6 and 7 in virtual cluster 2. Node 8 and 9 are border nodes and hence follows the sleep-listen schedule of both the virtual clusters 1 and 2. Implicating that node 8 and 9 will listen to both virtual clusters. This will reduce their regular sleep time in contrast to other nodes in both clusters. Hence, they will quickly drain energy as compared to all other neighbours, shortening their network life time. Towards this end, stationary and mobile agents become more practical. As the time elapses, the visit interval timer of MA on node 9 will start. MA utilizes others model maintained by SA to find other neighbours that have different schedule than its active schedule.

Now, MA is residing at node 9 and adjusts its clone-

parameter accordingly for the particular instance. Node 7 residing in virtual cluster 2 happens to be one of its neighbouring nodes, where MA visits. MA updates its knowledge from the energy model of node 7 and reports back to SA at source node (i.e., node 9), as shown in Figure 5.

��

The SA at node 9 (Border Node), updates the node 7

model in its others model table. Once all the cloned MAs have reported back, the SA at node 9 will compare the energy consumption for predefined period of the reported models with its local model. In this case, it finds node 7 consuming 115.2mW of energy which is least consumed energy for the predefined time in others model. Consequently, it will update its others model.

At this point, SA at node 9 marks node 7 schedules as active, as shown if figure 6. It also updates its local model correspondingly by updating its next sleep schedule similar to node 7. Then, it marks all the other schedules as inactive, X. This helps node 9 to follow the sleep-listen schedule of virtual cluster 2. The MA agent will keep visiting node 9 neighbour in virtual cluster 1, and report back to SA. If at any point, SA finds any neighbour in virtual cluster 1 that out performing its local energy model, it will move node 9 back to virtual cluster 1. Thus, node 9 will toggle between clusters whenever minimum energy consumption is predicted.

Sensing for a particular region may increase or decrease by demand, due to the nature of commercial applications of sensor networks [4]. Hence, it is vital to

A- Active X- Inactive B- Broadcast


highlight that nodes should always try to join the best energy efficient virtual cluster at times. This can be accomplished by MA of node 9 that visits its neighbours in virtual cluster 2. It will determine the energy efficiency based on the knowledge acquired by the neighbouring SA. If MA finds virtual cluster-2 outperforming virtual cluster 1, this results to the movement of node 9 to virtual cluster 2.

In doing so, SA will updates the others models, i.e. marking neighbours in cluster 1 as inactive and neighbours of cluster-2 as active, as shown in Figure 6.

7. Implementation All graphical user interfaces, agents and visualizations are implemented as a simulator in Java. To demonstrate and assess our contribution, a comparison with Sensor Medium Access Control Protocol (S-MAC) is performed. The user interacts with the agents using graphical interfaces to tune their parameters. To assist the user in validating energy efficiency of boarder nodes, we implemented virtual cluster visualization. All the components of the agents are implemented as objects For example, the problem solver is implemented as a rule-based system with a set of rules with some supporting classes, such as Agent-class, Network-class and V_Cluster-class.

��

Agent-class describes the characteristic and behaviour of a sensor node. The data members of this class include x-y coordinates, available energy, and a role variable to differentiate between SA and MA. Moreover, SA contains a list of its corresponding neighbour nodes. A Network-class describes the characteristic and setup of the network. The data members of this class include the network size, initial number of nodes, amount of energy allocated to each node, and the topology of network (i.e. random or grid). V_Cluster-class provides services to a set of nodes with the same schedule. The data members

of this class include total energy available in the virtual cluster, list of x-y coordinates of its member nodes, and number of times a node moved to a cluster.

Figure 7 shows the main interface of our simulator, which consists of input and output sections. The input section allows the user to input the desired parameters, using text-fields. Using these text-fields, parameters can be adjusted dynamically by inputting new data. The output section is further classified into two subsections. The first subsection on the bottom of the interface displays the results-set of specific simulated iteration. These results-set consists of a list of all border nodes in the network for a given period. Each row in the results-set contains SA-id, its energy status, present schedule, time spent on network, number of visits it received from neighbour’s MA, number of schedules in its schedule table, and number of times switched between clusters.

The second subsection at the upper right corner of the interface provides virtualizations of the network topology for all virtual clusters. In the visualization subsection, nodes are distinguished by different shapes and colours. Nodes shown as solid circles are assigned different colours based on their energy levels. MAs those unable to return to their original nodes are shown as black triangles. Using this interface, users are allowed to graphically monitor the status of the network at any time by pressing ‘Start Visualization’ button.

8. Experimental Results

This section discusses our simulated results and comparison between our approach and S-MAC.

��

The first experiment was conducted on a network size

of 230 nodes, where energy parameter was kept constant through out the simulation. We used random rectangle areas to simulate the increase and decrease of sensing. All the nodes in a rectangle area were selected and exposed to heavy load.

We performed our analysis on energy consumption of the boarder node within the affected clusters. Figure 8 show that our approach outperforms S-MAC. Although S-MAC performed slightly better during the initial


simulation time, this is due to the time required for the SA to build its knowledge.

��!� ��"�� #��# � ��

��

The variation in energy consumed with respect to MA’s visit interval is demonstrated in Figure 9. This figure shows that as the value of visit interval decreases will result into a decrease in energy consumption. Smaller the visit interval, frequent MA can visit neighbours and faster SA can built others model. Consequently, BN can make early and frequent decision to adapt to better neighbour.

To illustrate the effectiveness of our proposed approach, a comparison with S-MAC is performed. Each protocol is run through a series of simple workload test forming an empirical characterization of protocol performance. The purpose of this experiment is to show how in typical wireless sensor network our proposed approach and S-MAC will achieve low to high latency and its correlation with power consumption. We utilized RTS-CTS and message fragmentation services similar to that of S-MAC protocol. The test is run on a 50-hop network topology shown in Figure 10. Hops are placed one meter apart. In each test, the source node sends 30 messages with a payload of 25-bytes without fragmentation.

��

The latency at each hop of the network is measured

as the difference between the sender(A) and receiver(B) time stamps. It should be noted that adaptive listening is not considered in our implementations. From the results generated in Figure 11, we can observe that both

approaches have an increase in latency as the energy consumptions increases. However, S-MAC has lower latency as compared to our approach.

��!� ��"��

��

Our agent-based approach enables BN to firmly bank its energy. In contrast to this tight energy cutback, it accepts trade-off to increased latency. Figure 12 reveals that latency is high at start, but it gradually decreases as nodes start to die. MA visit interval also affects latency.

��$ ��% � ��

9. Conclusion and Future Work

This paper proposed a multi-agent based system to reduce the energy consumption among border nodes in S-MAC protocol. This system consists of two types of agents, including Stationary Agent (SA) and Mobile Agent (MA). SA is able to monitor the events associated with a sensor node during communication and builds its knowledge. MA is able to travel and interact with stationary agents at neighbour node. It benefits from predict the most energy efficient cluster for border nodes. A prototype of this system is simulated using Java. The system was tested and successfully compared with S-MAC, using some experiments.

Source

50 49

48

2

1

Sink


Our future work will include the incorporation of Mica [18] motes running on TinyOS operating system [17]. We also plan to deploy our agents that occupy minimal memory in motes and can be easily transverse in sensor network with fewer transmissions. Finally, we are considering adding an end-to-end security between the agents.

10. References [1] Akyildiz, I.F., Su, W., Sankarasubramaniam, Y.,

Cayirci, E., “A survey on sensor networks, IEEE Communications Magazine”, vol. 40, no. 8, pp, 102-116, 2002.

[2] Woo, A., Culler, D., “A transmission control scheme

for media access in sensor networks”, in Proceedings of the ACM/IEEE International Conference on Mobile Computing and Networking, Rome, Italy, July 2001.

[3] ASH Transceiver TR3000 Data Sheet [Online],

http://www.rfm.com/ [4] Estrin, D., Govindan, R., Heidemann, J. and Kumar,

S., Next century challenges: scalable coordination in sensor networks, ACM MobiCom’99, Washington, USA, pp 263–270, 1999.

[5] Bennett, F., Clarke, D., Evans, J., Hopper, A., Jones,

A., Leask, D., “Piconet: Embedded mobile networking,” IEEE Personal Communications Magazine, vol. 4, no. 5, pp. 8–15, Oct. 1997.

[6] Sohrabi, K., Pottie, G., “Performance of a novel self

organization protocol for wireless ad hoc sensor networks,” in Proceedings of the IEEE 50th Vehicular Technology Conference, pp. 1222–1226, 1999.

[7] LAN MAN Standards Committee of the IEEE

Computer Society, Wireless LAN medium access control (MAC) and physical layer (PHY) specification, IEEE, New York, NY, USA, IEEE Std 802.11-1999 edition, 1999.

[8] Kleinrock, L., Tobagi, F., "Packet Switching in

Radio Channels: Part I – Carrier Sense Multiple-Acces Modes and Their Throughput-Delay Characteristics", IEEE Transactions on Communications, Vol. COM-23, no. 12, pp. 1400-1416, December 1975.

[9] Stemm, M., Katz, R., “Measuring and reducing energy consumption of network interfaces in hand-held devices,” IEICE Transactionson Communications, vol. E80-B, no. 8, pp. 1125–1131, Aug. 1997.

[10] Norman, A., "Development of the ALOHANET",

IEEE Transactions on Information Theory, vol. IT-31, no. 2, Mar. 1985.

[11] Singh, S., Raghavendra, C.S., “PAMAS: Power

aware multi-access protocol with signalling for ad hoc networks,” ACM Computer Communication Review, vol. 28, no. 3, pp. 5-26, July 1998.

[12] Shakshuki, E., Ghenniwa, H. Kamel, M, “Agent-

based System Architecture for Dynamic and Open Environments”, International Journal of Information Technology and Decision Making, vol. 2, no. 1, pp. 105-133, 2003.

[13] Ye, W., Heidemann, J., “Medium Access Control in

Wireless Sensor Networks”, In Wireless Sensor Networks, C. S. Raghavendra, Krishna Sivalingam, Ty Znati, Editor, Kluwer Academic Publishers, pp. 73-92, 2004.

[14] Heinzelman, W., Chandrakasan, A., Balakrishnan,

H., “Energy-efficient communication protocols for wireless microsensor networks,” in Proceedings of the Hawaii International Conference on Systems Sciences, Jan. 2000.

[15] Lee, W., Lee, D., Lee, H., “Lifetime extension of

border nodes in SMAC-based wireless sensor networks by unifying multiple sleep schedules among adjacent virtual clusters”, PE-WASUN 2005, pp. 267-268, 2005.

[16] Culler, D., Estrin, D., Srivastava, M., “Overview of sensor networks,” IEEE Computer, vol. 37, no. 8, August 2004.

[17] Hill, J., Szewczyk, R., Woo, A., Hollar, S., Culler, D. E., Pister, K. S. J., “System architecture directions for networked sensors,” Architectural Support for Programming Languages and Operating Systems (ASPLOS), no. 4, pp. 93-104, 2000.

[18] Hill, J., Culler, D., “Mica: a wireless platform for deeply embedded networks,” IEEE Micro, vol. 22, no. 6, pp. 12-24, 2002.


[ieee 21st international conference on advanced networking and applications - niagara falls, on,...

Documents