Vous êtes sur la page 1sur 8

Analysis on Energy Cost for Wireless Sensor Networks

Yufu Jia Tianlin Dong JianShi Department of Electronics and Information Engineering Huazhong University of Science and Technology Email:jyf740620@yahoo.com.cn

Abstract
As one of data-gathering applications in information technologies, Wireless Sensor Network (WSN) has attracted more and more peoples attention. A key challenge is achieving energy consumption distribution of nodes for WSN with any scale in any general regions to the effect that node deployment density can be made to prolong systems lifetime. In this paper, we explore the optimal location of sink to prolong the network lifetime, give the definition of Information Barycenter and set up a model for WSN with nodes deployed ideally in monitored region. EC curve is proposed to reveal that energy consumption of nodes on different locations takes on curve shift along with its distance to sink. Simulations are carried out to pick up energy consumption of nodes on different positions on the preconditions that all the nodes are deployed in the interested region with uniform distribution or stochastic distribution to validate the availability of the EC curve.

1. Introduction
Driven by advances in MEMS micro-sensors, wireless networking, and embedded processing, Wireless Sensor Network have become increasingly available for commercial and military applications such as environmental monitoring , industrial sensing and diagnostics, monitoring critical infrastructures and information collecting for battlefield awareness. Sensor nodes in such a network usually have limited onboard processing and wireless communication capabilities, and are equipped with batteries with limited power. Moreover, it is impractical or

infeasible to replenish energy via replacing batteries on these sensors in most applications. To prolong the network lifetime, many literatures explored all aspects of WSN which involves the problems such as Hardware-design, Access control, Orientation, Data fusion, Cluster Routing, Power control, Density control, etc. Simulation environments with nodes deployed in test area with stochastic distribution or uniform distribution are set up to validate the proposed algorithms and protocols. But it is well perceived that sensor network should be deployed with reasonable density based on energy consumption. WSN has ad hoc property. Each node has two functions: information gathering and information forwarding. The one near sink should forward much more data than the further-side does. The latter will be disconnected with its upstream nodes and become a faultage, or have no contacts with its neighbors and become information islands. Existences of faultage and information islands reduce network lifetime significantly. Ideal working-node density can be obtained by controlling the monitored areas coverage with density control algorithm such as SPAN, OGDC etc, We can also adjust the amount of initial energy of nodes properly on some given application circumstance to alleviate the influence imposed by difference of energy consumption. So probing into energy consumption of nodes on different positions is very useful. Node energy behavior model is necessary to resolve the question about energy consumption distribution. Models for energy dissipated in computation and communication are discussed specifically in [8]. It assumes that the simple model includes the radio energy consumption which dissipates in the transmitting or receiving circuitry and the energy loss

Proceedings of the Second International Conference on Embedded Software and Systems (ICESS05) 0-7695-2512-1/05 $20.00 2005

IEEE

for the transmit amplifier to achieve an acceptable SNR. An r2 energy loss is adopted due to channel transmission is relatively short distance transmissions. Sensing energy dissipation to sense one bit is assumed to be a constant in [6]. The energy dissipated for sensing a bit is approximately equal to the energy dissipated in receiving a bit. Some other papers propose and account for a fundamental question concerning the limits of energy efficiency of sensor networks--what is the upper bound on the lifetime of a sensor network that collects data from a specified region using a certain number of energy-constrained nodes? Bhardwaj et al.[4] derive the fundamental upper bounds for a variety of scenarios assuming that node energy models is based on l/dn path loss behavior and all the nodes have optimal mobility, collaboration and ideal power control ability of transmitting. They do not consider network topology or the effect of data aggregation of data streams. The work is extended by taking into account of these factors in [9]. They derive tight or near-tight but topology insensitive bounds for non-aggregating sensor networks where the source resided in a region according to a predefined location. These bounds are derived by employing the formalism of feasible role assignments. Their model only considers the power consumption of sensor nodes that sense active events, process, transmit and/or receive data, not sensor nodes that are monitoring but sensing no active events. Wendi et al. [5] divide system energy consumption into two parts in the analysis of optimum number of clusters. The energy consumption by non-cluster head follows the Friss free-space model and the energy dissipation by cluster head follows the multi-path model. But it didnt consider energy consumption in the relay between cluster heads. Arnon et al. [7] derive an upper bound on the average network lifetime for a code division multiple access (CDMA) communication system. The communication procedure assumes that most of the energy consumed is transmission. They proposes to shape the sink antenna gain in such a way that maximizes the network lifetime, by taking advantage of the spatial non-uniformity distribution of the sensors on the ground to increase the sinks partial coverage to the network. Coleri et al. [10] investigate the lifetime of networked sensor nodes where sensors are organized

in a tree-based multi-hop networks. They analyze the lifetime of nodes in four different groups based on their distances to the data sink using the finite automata technique. However, their analysis is primarily on the lifetime of individual nodes instead of that of the network. To the best of our knowledge, there are many papers interested in the upper bound lifetime of wireless sensor networks, but the special topic on investigating the energy consumption distribution in deployed region has not been discussed. This paper focuses on estimating node energy consumption in the interested area base on node energy behavior model and ideal network deployment model. The remainder of this paper is organized as follows: In the next section, we set up a nodes energy consumption behavior model. In section III, we build up an ideal network deployment model. EC curve is proposed to figure out the variety of nodes energy consumption on different positions. We also explored the optimal location of sink for achieving longest network lifetime based on node energy consumption behavior model. Section IV contains a wide array of experimental results followed by a brief discussion of the conclusion.

2. Nodes energy consumption behavior model


A sensor node is made up of four basic components: a sensing unit, a processing unit, a transceiver unit, and a power unit. Sensing units are usually composed of two subunits: sensors and analog-to-digital converters (ADCs). The analog signals produced by sensors based on the observed phenomenon are converted to digital signals by ADC, and then fed into processing unit[3]. With that the generated data is integrated and encrypted by processing unit and transmitted to the next hop by transceiver unit. The following summarizes the energy-consumption models for each sensor component (use for references [4] [5] [6]).
Etx = (11 + 2 d ) r

------ (1) ;

Erx = 12 r ------------------ (2) ; Esense = 3 r ----------------- (3)


Ecpu = Esec urity + Eroute + E DA

; ;

------ (4) ;

EDA = eDA r --------------------- (5)

Proceedings of the Second International Conference on Embedded Software and Systems (ICESS05) 0-7695-2512-1/05 $20.00 2005

IEEE

Where a11 is the energy/bit consumed by the transmitter electronics, a2 accounts for energy dissipated in the transmit op-amp (including op-amp inefficiencies), a12 is the energy/bit consumed by the receiver electronics and a3 is the energy consumption of sensing a bit. This paper assumes that each node senses data with same bit rate r0. For the distance between two nodes is small, energy dissipation follows the Friss free-space model, that is n=2. The processing units energy consumption is the sum of three parts: security computation energy consumption, routing computation energy consumption and data aggregating energy consumption. The first two is computed with the data packets that are to be transmitted as the unit, while the latter is computed with bit as the unit. eDA is the energy/bit consumption of data aggregating by the processing unit.

3. EC curve and optimal location of sink


In order to find out the energy consumption variety rule of the nodes in the sensor networks, we try to put the nodes deployment into ideal state of the sensor network, and seek the energy consumption distribution of the sensing field under such state.

3.1. Ideal network deployment model

node aggregation, then the clusters do the data aggregation process and decrease the number of the messages transmitted to the sinks to economize the energy and raise the network expanding characteristic. This paper adopts the network clustering strategy, which supposes that the network nodes are divided into two categories: clusters headers (CH) and none cluster headers (NCH). Every cluster has n ' sensor nodes; all nodes in the cluster are distributed evenly and cover the cluster section completely. Each NCH nodes data rate is r0 bit/s. Data generated by each NCH sensor will be forwarded to its nearest CH node first. The CH node receives data with the total size of n ' r from each NCH node, aggregates it into a packet with the size of r1 and forwards it to its upstream CH nodes. In this paper, we assume that nodes in the sensor network are deployed under the ideal condition. From figure 1 we can see that the network deployment model carves up the monitored region into m ( m 3 ) pieces of same sectors, sink perches at the center of the region. Each sector has n layers (The one nearest to the sink is the first layer). The location of each CH node is fixed, and lies on the cross point of each line, as figure 1 shows. Given that Ecluster is the total energy consumption by all the sensor nodes in each cluster to process the data with the size of n ' r . It includes the energy consumption of sensing and transmitting data with the size of r0 to CH node by each NCH node, and the energy consumption of CH nodes receiving data with the size of n ' r0 ,
0 0

aggregating, encrypting, routing and forwarding the gained packet to the upstream places. We only consider point-to-point encryption in the ideal state, so the security computation energy consumption during forwarding between CH nodes is out of consideration. The total energy consumption
EC j

of

cluster j consists of the energy consumption of processing n ' r0 bits data, and the energy consumption Echtx of receiving data that are sent by
m

Figure 1. Network deployment model The clustering algorithm lets nodes join the multi-hop telecommunication within the particular

the downstream CH nodes and forwarding these data to its upstream CH nodes. The systems total energy consumption Etotal is the sum of all the clusters. For the distance d between the NCH node i and its CH
i

node being short, Friss free-space transmission model ( =2) is adopted, the distance d between CH

Proceedings of the Second International Conference on Embedded Software and Systems (ICESS05) 0-7695-2512-1/05 $20.00 2005

IEEE

nodes being long, the multipath transmission model ( =4) [5] is adopted. According to the system model above, we can get the formula as follows: (I)
Etotal =
mn (n +1) 2

ECi
i =1

; ;

(II) EC = Echtx + Ecluster


m m

(III)
Ecluster = Esec urity + Eroute + eDA n ' r0 + (
n' i =1

(11 + 12 + 3 + 2 d i 2 ) r0 ) + (11 + 2 d 4 ) r1

3.2. Node energy consumption analysis

To make the systems energy consumption lowest, the deployment model of ideal states requires that the data generated by each NCH node should be sent to the sink along the shortest route. As shown in figure 1, there are m special lines carving up the monitored region into m sectors. We call these lines as thread stems. CH nodes on thread stem forward data to sink along with this thread stem. Each CH node on non-thread stem has 6 neighbors and forwards data toward left father and right father respectively. For convenient analysis, we assume that the cluster header which is not on thread stem forwards data to the father nodes along the same direction at the same time, and along the forwarding direction the process alternately takes place along with time going. For the Table 1 Total energy consumption of each cluster
Cluster 1
Cluster 2

energy consumption of each node on the same layer being uniform in our ideal network deployment model, we assume that all the CH nodes on non-thread stem forward data toward left father at the time t0, toward right father at t1, back to left at t2, right at t3 At the initial time, each CH node only receives data from NCH nodes belonging to itself and forwards data to the upper CH node. Henceforth CH nodes forward not only the data from NCH nodes at this moment but also the data transmitted by the downriver CH nodes at former moment (CH nodes on the last layer only forward the data from NCH nodes at this moment). Data forwarded by CH nodes will increase along with time going until system goes into steady phase after n moments. At present, data forwarded by all CH nodes lying on layer k includes data generated by all NCH nodes belonging to layer n at the former nth moment, data generated by all NCH nodes belonging to layer n-1 at the former n-1th moment, , data generated by all NCH nodes belonging to layer n-k+1 at the former n-k+1th moment, , data generated by all NCH nodes belonging to layer k at this moment. Assumes that the sum of the former 4 items in formula (III) is P1, the fifth item is P2. Table 1 sum up the analysis above and list the total energy consumption of each cluster when system goes into steady state.

Cluster 3
P1 P2 P1 2P2 P1 3P2 P1 n 1 k P2

Cluster k
P1 P2 P1 2P2 P1 3P2) P1 n 1 k

Cluster n-2
P1 P2 P1 2P2 P1 3P2

Cluster n-1

Cluster n
P1 P2 P1 3P2

Layer n Layer n-1 Layer n-2 Layer k


P +( 1

P1 P2 P1 3P2 P1 6P2

P1 P2 P1 2P2 P1 3P2 P1 n 1 k P2

P1 P2 P1 2P2 P1 6P2

(n k + 3)(n k ) + 1) P2 2

P +( 1
P2

(n k + 3)(n k ) + 1) P2 2

According to table 1, the total energy consumption of nodes at layer k can be counted as follows:
[kP + 1 n2 k 2 + n + k P2 ]m --- (6) 2

The deployment model in steady phase assumes that all nodes are fixed. In the practical deployed sensor network, the function of all CH nodes is undertaken by each sensor node in the cluster section in turn. Therefore ones on the same layer cost same size of energy approximately. Because of the influence of energy consumption and dynamic routing, the

forwarding of data does not strictly follow the route and step analyzed above. Instead the energy consumption of the clusters on thread stems of the same layer tend to be well-balanced with energy consumption of the clusters which are not on thread stems. Therefore we get the energy consumption per second of each node inside layer k is:
P n2 k 2 + n + k 1 + P2 --- (7) n' 2n ' k

From (7) we know that average energy consumption of nodes decreases as level of layer k

Proceedings of the Second International Conference on Embedded Software and Systems (ICESS05) 0-7695-2512-1/05 $20.00 2005

IEEE

increases. What depicted by (7) are series of discrete points. In a general way, the proposed clustering algorithms select CH nodes base on local decision but not on the distance. Nodes in two neighboring layers can also form some clusters together. So the variety of energy consumption of sensors should be consecutive. In order to obtain the consecutive energy consumption curve derived from (7), we try to replace the level of layer with distance. Define that D=nrt is the range of investigated region and x is the distance between the midpoint of layer k and sink, replace k with x , n with D, the average energy consumption of ones on layer k can be reckoned roughly as follows :
P 1 + n' D cos ( ) x ' cos( ) + Dd cos ( ) + x ' d cos( ) m m m m P
2 2 2 2

adjust deployment density or nodes initial energy to acclimatize system to the variety of energy consumption on different positions.

3.3. Optimal location of sink

Figure 2. Sketch map about WSN Since all the data generated from different places of monitored area flows to the sinks, we can see from the above analysis that the location of the sink determine the energy consumptions size and uniformity degree of nodes in different places, and have important influence on the network lifetime. Base on the node energy behavior model, we try to obtain the optimal location of sink for the longest lifetime of system. Given that N nodes are placed in the monitored area , sink locates at position (xb, yb) (see figure 2). Each node has initial size of energy Eini(i) and generates data with the rate r((x, y), t). Energy consumption to sense and forward one bit from (x, y) to sink is e(x, y). System will last for Llifetime, such equation can be got:
N i =1

2n ' x ' d cos( ) m

----- (8)

Define x as the distance between some given NCH node and sink. This model assumes that CH node is the one whose distance to sink is smallest among all the nodes in the same cluster and CH nodes locating at the same layer are further to sink than the midpoint. In order to alleviate the influence of the two factors, a coefficient ( 1 ) is introduced to amend the distance, let x = x ' . When the distance of node to sink is further, the influence of the two factors becomes much less. So should diminish when x increases but could not be less than 1, then let
1
x'

( is a

Einit (i) =

e( x , y ) [

Llifetime 0

r (( x, y), t ) dt ]dxdy

------- (10)

constant). Base on (8), we can get the estimation of energy consumption of node whose distance to sink is x:
P 1 + n' D 2 cos 2 ( ) ( x ) 2 cos( ) + Dd cos2 ( ) + ( x )d cos( ) m m m m P 2n '( x ) d cos( ) m

- (9)

The wireless sensor network adopts the way of multi-hop to transmit data, and all the nodes have optimal mobility, collaboration and ideal power control ability of transmitting. Since the computation energy consumption of different nodes is relatively evenly, we assume here that E
cpu

Formula (9) is the curve to reveal variety of energy consumption of nodes on different positions. We call it as EC curve. When node approaches sink, the derived result from (9) becomes too large. In real networks, no nodes are deployed very near sink because of sinks sense capacity, so we confine the independent variable x in (9) with x r (0 1) . As a
t

=0

. According to

Jensens inequality, only when all the hop distances are made to be equal, is e(x, y) minimal. At this time, the best distance of nodes is d = n
1 , so the 2 ( 1)

minimum of e(x, y) can be deduced as follows:


e( x , y ) = (11 + 12 )
( x xb )2 + ( y yb ) 2 12 + 3 1 11 + 12 n 2 ( 1)

rule, adopts 0.25. Different nodes energy consumption reflects deployment density magnitude of the corresponding location on the precondition that all the nodes have same initial size of energy and have ability to control its working states to minimize the degree of coverage. Base on EC curve, we can

--- (11)

Integrate (10) with (11),


N i =1

Einit (i ) =

[(11 + 12 )

( x xb )2 + ( y yb ) 2 12 + 3 ][ 1 11 + 12 n 2 ( 1)

Llifetime 0

r (( x, y ), t )dt ]dxdy

------- (12)

Proceedings of the Second International Conference on Embedded Software and Systems (ICESS05) 0-7695-2512-1/05 $20.00 2005

IEEE

Generally speaking, a11= a12= a3= a,[9]


N i =1

Einit (i ) = 2 2 2

( x xb ) 2 + ( y yb ) 2 [

Llifetime 0

r (( x, y ), t )dt ]dxdy

------- (13) If all the nodes sense with rate r(x, y) which dont vary with time, the lifetime can be achieved as follows:
N

Einit (i) Llifetime =


i =1

2 2 2

r ( x, y ) ( x xb ) 2 + ( y yb )2 dxdy

---- (14)

simulation, energy consumption per bit is transformed into the one consumed per second. It assumes that node cost same size of energy when it is in active state with the one in idle state, but it cost none when in sleep mode. Values of key parameters of model are taken as follows: a11=a12=a3=50nJ/bit, when =2, 2 a2=10pJ/bit/m , and when =4, a2=0.013pJ/bit/m2, eDA=5nJ/bit/signal, Esec = E = 0 , r0=r1=1kbps,
urity route

According to equation (12) and (14), such conclusion can be draw that optimal location of sink is the point that demands the sum of product of sensed data by each node and distances to sink reduces to minimum. This point is called as Information Barycenter. In order to make the gross of energy consumption minimal, we should place sink on Information Barycenter of interested region as possible as we can.

d=200m, =10.

4.1. Optimal location of sink

Target Nodes
900 800 700 600 500 400 300

Sensor Nodes

Sink

4. Simulation result
The simulation scenario consists of one sink node, some target nodes and sensor nodes. Each target node generates a stimulus every two seconds. Each sensor node will generate 500 bits to forward when it senses a stimulus. The sensing range of sensor node is 100 m and the transmitting range is 200 m. Seismic propagation model is used on the sensor channel. In wireless protocol stack, AODV and 802.11 is used as the underlying ad-hoc routing protocol and MAC protocol. We dont take into account the energy consumed in forwarding some controlling packets generated by MAC and network layers in this simulation. Energy is consumed not only by active communications, but also by wireless devices in the idle and/or sensing state. As a matter of fact, the energy consumed by wireless devices in the idle and/or monitoring state is only a little less than that in the transmitting or receiving states. Thus it makes more sense to derive the network lifetime under the scenario that only a minimum set of sensors are turned on, while the other sensors operate in the low-power mode (or sleep mode). In our simulations, all the nodes geographical position information has been known precisely. The radio range of nodes is twice of the sensing range, i.e., coverage implies connectivity. So we adopt the OGDC density control algorithm to work between MAC and network layers. In the

200 100

100

300

500 X

700

900

Figure 3. Nodes deployment scheme (1)


Target Nodes 900 800 700 600 500 400 300 200 100 100 200 300 400 500 X 600 700 800 900 Y Sensor Nodes Sink

Figure 4. Nodes deployment scheme (2) In order to validate that Information Barycenter is the optimal location of sink for the longest longevity and lowest energy consumption of network system, we take two nodes deployment schemes to explore. The first one is uniform distribution for the representation of regular regions. 81 target nodes and 64 sensor nodes are evenly distributed over a (800800) m2 rectangle region (see figure 3). The second one assumes that nodes are distributed in the region arbitrarily (see figure 4) for the representation of irregular regions. The Information Barycenter of the interested region is (500, 500) in figure 3 and (395, 416) in figure 4. In order to illuminate the influence of sinks position,

Proceedings of the Second International Conference on Embedded Software and Systems (ICESS05) 0-7695-2512-1/05 $20.00 2005

IEEE

sink is placed at 1: (100, 100), 2: (150, 150), 3: (200, 200), 4: (250, 250), 5: (300, 300), 6: (350, 350), 7: (395, 416), 8: (400, 400), 9: (450, 450), 10: (500, 500) respectively in different running. Figure 5 shows the total energy consumption of all the nodes in each simulation running which last for 100 seconds. The curve derived from deployment scheme 1 is degressive. As sink approaches the Information Barycenter, the total energy consumption declines. The curve from deployment scheme 2 is irregular, but the total energy consumption is comparatively small when sink approaches the neighborhood of the Information Barycenter. Packet drop ratio is defined as the ratio of the number of the stimulus generated by all the target nodes to the number of packets received by sink. It is important to reflect the performance of network system. From figure 6 we can see that packet drop ratio is minimal when simulate with sink at Information Barycenter both by deployment scheme 1 and scheme 2. Deploying nodes arbitrarily makes some target nodes uncovered with. A small proportion of stimulus can not be sensed by sensor nodes, so packet drop ratio obtained by scheme 2 is much larger than that by scheme 1. In figure 5, the total energy consumption reaches minimum when sink is at Information Barycenter. So we can say that putting sink at Information Barycenter can achieve best availability of data- gathering with lowest energy consumption.
nodes deployment scheme 1 total energy cost (joules)
8 7. 5 7 6. 5 6 5. 5 5 0 2 4 6 8 10

4.2. Comparison between EC curve and network simulation result


Running the simulation with sink at Information Barycenter by scheme 1 for 100 seconds ten times. We divide the interested region into 64 sub-regions with each node in its center by the area unit 100100 m2. Encircling sink, these sub-regions are organized into 4 groups. Given that Eave is the average energy consumption of nodes in each group. Figure 7 presents the values of Eave derived from simulation and calculating base on formula (8) respectively. It reflects the approaching degree of EC curve to the simulation result approximately.
EC curve energy cost (joulses) 0.14 0.12 0.1 0.08 0.06 1 1.5 2 2.5 3 ID of Group 3.5 4 4.5 5 simulation result

Figure 7. Aproaching degree of simulation result to EC curve


energy cost (joule) 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 ID of node 40 50 60 base on EC curve simulation result

nodes deployment scheme 2

Figure 8. Comparison of energy cost in monitored region According to formula (9), is an important factor to estimate the energy consumption distribution. Experiments reveal that when adopts 1% of the transmitting range, it works better. Figure 8 lists the average energy consumption of each node derived in the 10 simulations and the calculated result base on formula (12) with =20. The ID of node in figure 8 is
100 x confirmed by the sequence which equals y 100 9 + 100 .

position of sink

Figure 5. Comparison of total energy cost

nodes deployment scheme 1 4 3.5 3 2.5 2 1.5 1 0.5 0 0 2 4 packet drop ratio

nodes deployment scheme 2

6 position of sink

10

The distinction of two derived energy consumption is about 8%. Trend of variety of energy consumption is consistent approximately.

Figure 6. Comparison of packet drop ratio

5 Conclusions
Power management is one of important

Proceedings of the Second International Conference on Embedded Software and Systems (ICESS05) 0-7695-2512-1/05 $20.00 2005

IEEE

propositions in the research area of wireless sensor networks. Deploying nodes according to EC curve roughly can balance the difference of energy consumption of nodes on different positions. General applications of wireless sensor network demands too many sensor nodes deployed in the given region and each node supplied with same initial amount of energy. It is unfeasible to deploy sensors node by node. What we recommend is that only in the very large monitored region, adjusting deployment density properly can be useful. Grasping the EC curve of monitored area of the sensor network and finding the upper bound of network lifetime in ideal state provide an absolute standard for evaluating the effectiveness of route protocols energy.

Networks Modeling, Analysis and Simulation of Computer and Telecommunications Systems, 2002. MASCOTS 2002. Proceedings. 10th IEEE International Symposium on 11-16 Oct. 2002 Page(s):129 136 [7]. Arnon, S.; Deriving an upper bound on the average operation time of a wireless sensor network Communications Letters, IEEE Volume 9, Issue 2, Feb. 2005 Page(s):154 156 [8]. W. Heinzelman, et. al, "Energy-Scalable Algorithms and Protocols for Wireless Microsensor Networks," Proc. International Conference on Acoustics, Speech and Signal Processing (ICASSP '00), June 2000. [9]. Bhardwaj, M.; Chandrakasan, A.P.; Bounding the lifetime of sensor networks via optimal role assignments INFOCOM 2002. Twenty-First Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE Volume 3, 23-27 June 2002 Page(s):1587 - 1596 vol.3 [10]. S. Coleri, M. Ergen, and T. J. Koo. Lifetime analysis of a sensor network with hybrid automata modelling. In First ACM International Workshop on Wireless Sensor Network and Applications (WSNA02), Sep. 2002. [11]. Bo Hong,Viktor K Prasanna. Optimizing System Life time for Data Gathering in Networked Sensor Systems. Algorithms for Wireless in and Ad-hoc with Networks(A-SWAN)(Held MobiQuitous 2004), 2004-08 [12]. Wen Hu,Chun-tung Chou,Sanjay Jha et al. Deploying long-lived and cost-effective hybrid sensor networks[C].In:The First Workshop on Broadband Advanced Sensor Networks ,BaseNets 2004,San Jose,CA, 2004-10 [13]. D Tian,N Georganas.A Coverage-Preserving Node Scheduling Scheme for Large Wireless Sensor Networks[C].In:WSNA02,Atlanta,Georgia,2002-09 [14]. J-H Chang,L Tassiulas. Routing for maximum system lifetime inwireless ad hoc networks[C].In :the 37th Annu Allerton Conf Communication,Control and Computing,Monticello,IL, 1999-09 [15]. Sinha and A. Chandrakasan, Dynamic Power Management in Wireless Sensor Networks, IEEE Design & Test of Computers, Apr. 2001, vol. 18, Page(s):6274. conjunction

Acknowledgments
The authors wish to acknowledge enlightening discussions with Manni Luo and Jiang Li and thank Bingxin Shi and Jicheng Liu for reading the draft.

References
[1]. B. Chen, K. Jamieson, H. Balakrishnan, and R. MOrris. Span: An energy-efficient operation in multihop wireless ad hoc networks. In Proc. of ACM MobiCom01, 2001. [2]. Honghai Zhang and Jennifer Hou, Maintaining coverage and connectivity in large sensor networks, in International Workshop on Theoretical and Algorithmic Aspects of Sensor, Ad hocWireless and Peer-to-Peer Networks, Feb 2004. [3]. Ian on F. Sensor Akyildiz, Networks. Weilian IEEE Su, Yogesh Sankarasubramaniam, and Erdal Cayirci. A Survey Communications Magazine August 2002 [4]. Manish Bhardwaj, Timothy Garnett, Anantha P. Chandrakasan. Upper Bounds on the Lifetime of Sensor Networks. Communications, 2001. ICC 2001. IEEE International Conference on Volume 3, 11-14 June 2001 Page(s):785 - 790 vol.3 [5]. Wendi B. Heinzelman, Anantha P. Chandrakasan, and Hari Balakrishnan. An Application-Specific Protocol Architecture for Wireless Microsensor Networks Wireless Communications, IEEE Transactions on Volume 1, Issue 4, Oct. 2002 Page(s):660 - 670 [6]. Mohamed Younis, Moustafa Youssef, Khaled Arisha. Energy-Aware Routing in Cluster-Based Sensor

Proceedings of the Second International Conference on Embedded Software and Systems (ICESS05) 0-7695-2512-1/05 $20.00 2005

IEEE

Vous aimerez peut-être aussi