2007 IEEE International Conference on Signal Processing and Communications (ICSPC 2007), 24-27 November 2007, Dubai, United Arab Emirates

A GENERLAIZED ENERGY-AWARE DATA CENTRIC ROUTING FORWIRELESS SENSOR NETWORK

Tayseer A. AL-khdourCOE Dept., KFUPM, Dhahran, Saudi Arabia

khdourgccse. kfupm. edu. sa

Uthman BaroudiCOE Dept., KFUPM, Dhahran, Saudi Arabia

ubaroudig. kfupm. edu. sa

ABSTRACT

Recently, an Energy Aware Data centric routing protocol(EAD), was proposed and its performance outperforms otherexisting routing protocols such as LEACH. However, EAD hasan inherited problem that only closed nodes to the sink will bechosen as gateways which eventually leads into isolating therest of the network's nodes even though they still have enoughpower to communicate with the sink. In this paper, wegeneralized the EAD protocol (EADGeneral ) such that any nodein the network can be a gateway node. Hence, an additionalphase is added to dynamically select gateways. In this phase,gateway nodes will be selected among candidate gatewaysbased on their energy. The simulation results show significantimprovement of the proposed protocol in terms of longernetwork lifetime, higher throughput and lower energyconsumptions compared to original EAD.

Index Terms- Routing protocols; Wireless sensor networks;Energy Conservation; clustered networks

1. INTRODUCTION

Recent Advances in (MEMS) Micro Electro Mechanical Systemhas enabled the development of smart sensors. The overallarchitecture of sensor node consists of the sensor nodeprocessing subsystem, the sensor subsystem and thecommunication subsystem [1].

These advancements along with the advances in wirelesstechnology have enabled the creation of Wireless SensorNetworks (WSN) [2]. A WSN is composed of a large number ofsensor nodes that communicate using a wireless medium (air).The sensor nodes are deployed randomly in the environment tobe monitored. The sensor nodes distributed in ad hoc structure.In WSN there is no base station and not all nodes hear eachother. The WSN is a multi-hop network.

The WSN has many applications that are extended over awide range. Some ofthese applications are: physical security formilitary operations, indoor/outdoor environmental monitoring,seismic and structural monitoring, Industrial automation, bio-medical applications, and health and Wellness awareness [1].

Although the WSN is a wireless multi hop network, it hasdistinguished features over the traditional multi hop wirelessnetworks. These features are related to the ease of deploymentof sensor nodes, the system lifetime, the data latency, and thequality of the network [3]. These features must be taken intoaccount when designing different protocols that control theoperation of WSN such as MAC protocols and routingprotocols.

All the sensor nodes have a limited power supply(batteries). Since all the nodes are out of control, it is impossible

to replace or recharge these batteries. All the protocols must bedesigned taking into account the energy constraint. The sensornode not only loses energy while transmitting and receivingdata but it also loses energy while sensing the medium. Thepower consumed during the idle-listening state is about 50%-100% of the power consumed while receiving data It has beenshown in [4] that the idle:receive:send power consumptionratios are 1:1.05:1.4. Others [5] show that the idle:receive:sendpower consumption ratios are 1:2:2.5

Since all the nodes of WSN are distributed randomly,many nodes will be very close to each other. Therefore, there isa lot ofredundant in the data gathered from the environment. Toreduce the data traffic in WSN, data aggregation is needed. Dataaggregation is performed at each intermediate sensor node. Theaggregation function depends on the specific application.

In this work, we generalize the Energy Aware Data centricrouting Protocol (EAD) [6]. We call the modified EAD protocolEADGeneral. The proposed protocol intends to increase thelifetime of the network by increasing the number of candidategateway nodes. The building tree protocol is generalized suchthat not only the nodes that are close to the sink can beconnected directly to sink, but any node in the network can alsobe connected directly to the sink. A mechanism to selectgateways based on the energy of the nodes is proposed. Theselected node will act as a gateway as long as its energy isgreater than a threshold value Eth. New gateways will beselected each round and therefore a new tree will beconstructed.

The rest of this paper is organized as follows. In section 2,the related work will be described. The proposed protocol willbe introduced in section 3. In Section 4, the performanceevaluation ofthe EADGeneral and a comparison with EAD will bediscussed. Finally, section 5 will conclude the paper with thepossible directions for future work.

2. RELATED WORK

Many Routing and MAC protocols for WSN are proposed todecrease the energy consumption in the sensor nodes. All theexisting WSN MAC protocols try to increase the time where thenode is in sleep state. Some of these MAC protocols arecontention based protocols such as: Sensor MAC (S-MAC),and T-MAC, [7], [8]. Some of the MAC protocols arescheduled based protocols such as: (ALPR MAC) and BMA[9],[10]. The routing protocols for WSN can be classified asdata centric, hierarchical, or location based [11]. Examples ofrouting protocols for WSN are presented in [11].

Boukerche et al proposes an Energy-Aware Data-CentricRouting Algorithm (EAD) [6]. In EAD protocol, each round isdivided into two phases: Building the Tree (BT), and DataTransmission (DT). An energy aware tree rooted at the sink isconstructed in phase-1. The tree consists of leaf and non-leafnodes. A leaf node is a node that has no child. On the other

1-4244-1236-6/07/$25.00 © 2007 IEEE 117

hand, a non-leaf node is a node that has at least one child. Leafnodes sense the monitored area and transmit the correspondingdata to its parent. On the other hand, the non-leaf nodes alsosense the monitored area and they act as intermediate nodes totransmit data from lower level to upper level of tree. In the datatransmission phase, data is transmitted from sensor nodes to thesink. Data transmission period may be repeated many times in asingle round. All the nodes use CSMA/CA for transmitting thedata.

All the leafnodes ofthe tree will turn their radio OFF mostof the time. On the other hand, all the non-leaf nodes will turntheir radio ON all the time. When an event occurs, the leafnodes will collect the related data and turn its radio ON totransmits the data to its parent. When a non-leaf node receivesdata from all its children, it will aggregate the data and send it toits parent. Since the radio of the non-leaf sensor nodes will bealways ON, they will lose much power than the leaf nodes.

In EAD protocol, all the nodes are connected to the sinkthrough few nodes that are close to the sink. These nodes areconsidered as gateways. These nodes will be non-leaf nodes formost of time; they will consume a lot of energy. Therefore, theywill die early. When they die, the rest of network will beisolated. However, those isolated nodes still have non-consumed energy. Therefore, energy utilization is not soefficient in EAD protocol.

3. A GENERALIZED EAD PROTOCOL DESCRIPTION

To generalize EAD protocol, we assume that each node has theability to transmit its data for long distance, i.e. its transmissioncan reach the sink. Each node has power control capability suchthat the transmission energy depends on the distance to thedestination node. When a node sends data to its nearestneighbor, the transmission energy will be small compared withthe transmission energy required to transmit data to the sink.

In our proposed protocol, a new phase; Selecting Gateways(SG), is added. In this extra phase, gateways nodes, nodes thatwill communicate directly with the sink, will be selectedautonomously. The nodes that have energy greater than athreshold value Eth will communicate with the sink directly andwill act as a gateway in the current round. The time frame forproposed protocol is shown in Figure 1.

Round -1 Round -2

SG|BT|DT|DT|DT |SG|BT|DT|DT|DT|

SG: Selecting GatewayBT: Building Tree PhaseDT: Data Transmission phase

Figure 1. Time frame for the EADGeneral Protocol.

3.1. Selecting the Gateways

In this phase, gateway nodes are selected. It is assumed that thenetwork is virtually divided into tiers. Each tier includes nodesthat are located at specific distance from the sink. For example,tierO includes all nodes that are located at distance less than do,tier, includes all nodes that are located at distance greater thando and less than d1 and so on. Figure 2 shows an example of anetwork and its associated tiers.

Sink

60 ~~~~---------

-- -----2 -- ----- ---- ----

Figure 2. A Sample Network with its tiers

Initially, the nodes of tierO will be considered as potentialcandidate gateways. Based on their energy, some ofthese nodeswill advertise themselves as gateways. They will act as

gateways until their energy drops bellow a threshold value E1thThen new gateways will be selected from the nodes of tier]. Theselected nodes will act as gateways until their energy dropsbelow Eth and so on. When all tiers are considered and no more

nodes can be selected as gateways based on the current Eth. Anew cycle will start, in this cycle new gateways will be selectedfrom tierO using smaller value of Eth and so on.

To select the gateways, the sink broadcast an ADVmessage. The ADV message contains a field for E5th InitiallyADV message is broadcasted with minimum energy such that itreaches the nodes of tierO only. When a node receives the ADVmessage, it compares its energy with Eth, and then it respondswith a JOIN message. A JOIN message contains a confirmfield. Confirm is set to 1, ifthe node's energy is greater than Eth,i.e. the node can be a gateway and it selects the sink as itsparent, otherwise confirm is set to 0. After the node sends itsJOIN message, it will act as gateway in the current round.Assuming reliable channel, it does not need a confirmation fromthe sink to be a gateway. When the sink receives JOINmessages from all nodes in the target tier and the confirmfield=0 in all the received JOIN messages, then no node fromthe target tier can be a gateway, the sink will broadcast a new

ADV message with higher transmission energy using the same

Eth to select a gateway from the next tier. The nodes of the nexttier will respond with JOIN messages according to their energy.The process will continue until all tiers are considered and no

node has energy greater than Eth; no node can be a gateway. Anew cycle will start from tierO with new Eth,Eth(new) =eEth(current), where O<e< 1. And following the same

procedure as above, new gateway nodes will be selected fromtierO. For each cycle, a fixed Eth will be used, and at thebeginning of each new cycle, Eth will be reduced by the factor e.

The sink and nodes will exchange messages using the CSMAmechanism. The node has to be ON until it receives the ADVmessage from the sink and then it sends the JOIN message.Since the node does not need confirmation from the sink, it willgo to sleep immediately after sending the JOIN message.

After selecting the gateways, the next phase will be startedto build the tree. The gateway nodes will initiate the process ofbuilding the tree.

3.2. Building the Tree:

To build a tree rooted at the sink, we use the same protocolproposed in [6] with two modifications; the gateway nodes notthe sink will initiates the process of building the tree and theinitial status of gateways will be not 0. Building the tree isperformed by broadcasting control messages. The controlmessages are broadcasted using minimum energy transmission

118

such that the transmitted signal will reach the nearest neighborsonly.

4. PERFORMANCE EVALUATION

To evaluate the performance of the proposed protocol, a

simulator using C-language was built. We assume a gridnetwork with 100 sensors distributed in an area of 50X50 M2.Where nodes distributed between (x=0,y=0) and (x=45, y=45),

and the distance between nodes are either 5 or 5 m. Eachnode will have eight neighbors. The Sink is located at location(x=25, y=50).

We assume that the control message length is 48 byteswhile the data message length is 100 bytes. We assume perfectaggregation in which a set of data packets are aggregated intoone packet. We use a power control model in which the energy

consumed during transmission depends on the transmissiondistance [3]. The energy consumed during transmission of k bitsfor a distance d meters (ETX) and the energy consumed toreceive k bits (ER,) are calculated by:

E = kEeiec + kE d2TX elec ~amp (2)ERx = kEelec

Where Eelec represents the electronics energy and itdepends on factors such as the digital coding, modulation,filtering, and spreading of the signal. On the other hand, Eamprepresents the amplifier energy. In our experiments we assume

Eejec=50nJ/bit and Eamp==00pJ/bit/m2, which are the same

values used in [3]. The initial energy stored in each node is 2 J.we assume that the initial Eth is 50% of the initial energy (i.e. 1

J). Every cycle Eth will be reduced be a factor of 0.5 (e=0.5). Asummary of simulation parameters is shown in Table 1.

Table 1. Simulation ParametersPARAMETER VALUEMonitored area dimension |50 X50 m2

Maximum distance between two | 2 *

nodes

Electronics Energy (Eelec) 50nJ/bitAmplifier Energy (Eam,j) 100pJybitIm2Initial Energy in Each Node 2JControl Packet size 48 byteData Packet size 100 byteEthrshold (Single Data Transmission 4*10-4 j

eri od)

Initial Eth J

For each configuration, we measure the network lifetime(number of live nodes per time), throughput (number of datapackets delivered to the sink), and total consumed energy, withdifferent number of periods for Data Transmission phase (1, 5,and 10). The node will die, and will not participate in thecoming round, if its energy becomes less than a threshold value(Ethreshold). In the worst case, a node will be a non-leaf nodeand it will have eight children in the coming round. Theminimum energy needed for this node to be able to participatein the coming round is Ethresholdmin. Ethresholdmin is calculatedusing (2). In a single transmission period, the node will receiveeight data packets and transmit one data packet. Ethresholdminwill be:

Ethresholdmil k* Eeiec + kEampd2 + 8kEeiec (3)

Using the assumed values for k, Eelec, Eamp and d,Ethresholdmin for a single data transmission period will be 376pJ. Taking into account the energy needed in selecting

gateways, sensing channel, and building tree, we assume

Ethreshold to be 400 iJ for a single data transmission period.For five and ten Data Transmission periods, Ethreshold willapproximately be 2000, and 4000 iJ respectively. On the otherhand, we assume the isolated nodes as died nodes. Isolatednodes are the nodes that did not receive a broadcast message tojoin the tree.

Uo

z

z

10U S F -F__ 1;,-l -1 - -__9101.

90I I

80 - ---EAD(1 Period) - I -l EAD(5 Periods) . I i

70 Bg++-EAD (10Periods) -

60 - GEADGeneral (1 Period) t- - - - --

50 + EAD,General(5 Periods) _ I _EAD,General(1O Periods) E I I

40 - - -

30 - .--

20 __- -T- _--_II- _I_ __

20

0 10 20 30 40 50 60 70 80 90 100Time (sec.)

Figure 3. Number of Live Nodes vs. Time, (EAD, EADGeneral)

20

18

n 16

14

a)Wu 12(

10

o 8

6'--

4'

2'

0 10 20 30 40 50 60 70 80 90 100Time (sec.)

Figure 4. Total Consumed Energy vs. Time (EAD, EADGeneral)

Figure 3 to Figure 5 show a comparison between theresults obtained when implementing the original protocol(EAD) and the proposed general EAD protocol (EADGeneral).Figure 3 shows that EADGeneral protocol outperforms EAD interms of the network lifetime for all data periods. For example,in 10 data periods, the network lifetime when implementingEADGeneral protocol is about 97 seconds, while it is only 67seconds when implementing EAD protocol. An improvementby 44.8% is achieved. The improvement is due to utilizing theisolated nodes by increasing the candidate gateways.

In both protocols, the number of live nodes is stable around100 then goes down. It goes down very fast in the EADprotocol, while it goes down gracefully in EADGeneral protocol.Further, in the EAD protocol, only the nodes close to the sinkwill act as gateways for the entire network lifetime. Since thegateway nodes will wait until it receives all the data packetsfrom all leaf nodes, they will transmit their data packets to thesink at the end of the data transmission period, so they have toturn their radio ON for the whole data transmission period. Thiswill consume a lot of energy; therefore, they will die very early.When the gateway nodes die, the rest of the nodes will beisolated since they will not be able to communicate with thesink although they may still have unutilized energy. This can beobserved in Figure 4, which shows a comparison between the

119

10 - 4--t - - X4 - - -1 -1- - - - -1-- - - - - - - - -.

o_-- - D(0 . id

° ~~ - - XSv _ -1- - - 1- - - E- D, eea(1 Period) __.

O - --4 <-X - - - _ __ _ _ _ i_ D,General(5 Periods) __.

I I-

^ / R~~~~~~~~~~EAD,General(l 0 Periods)

1-

total consumed energy when implementing the EAD protocol,and the total consumed energy when implementing EADGeneralprotocol. Since we have 100 nodes in the network, and initiallyeach node has 2J, then the total initial energy in the network is200J. From Figure 4, we observed that the total consumedenergy in the EAD protocol with different data transmissionperiod does not exceed 180J. There is still more than 20J as

non-consumed energy. This non-consumed energy is stored inthe isolated nodes. It is considered a wasted energy. In EADprotocol, the gateway nodes will die approximately at the same

time so the number of live nodes goes down very fast. On theother hand, in the EADGeneral protocol, any node in the networkcan be a gateway. When the nodes close to sink die, any othernode in the network can be a gateway, therefore the number ofisolated nodes will be decreased compared with EAD protocol.Consequently, the number of live nodes will decrease more

smoothly in the EADGeneral protocol compared with EADprotocol. Since any node can be a gateway in EADGeneral, mostof the nodes will be able to communicate with the sink and willstay active in the network until all its energy is utilized. Thiscan be illustrated by Figure 4 where the total consumed energy

in EADGeneral protocol is about 200 J, which is very close to thetotal initial energy in the network. The energy utilization ismore efficient in EADGeneral protocol compared with EADprotocol.

.:

a1)

-o

a1)a)

n

-0

0 10 20 30 40 50 60Time (sec.)

70 80 90 100

Figure 5. Throughput vs. Time, (EAD, EADGeneral)

On the other hand, due to the increase in the networklifetime a significant improvement in the throughput isachieved. For example, the total data packets delivered to thesink when implementing EADGeneral protocol for 10 data periodsis about 62000 packets, while it is about 10000 packets in EAD.An improvement by 520% is achieved. The significantimprovement in throughput can be attributed to increasing thenumber of gateways. In the EADGeneral protocol, there will bemore gateways than in the EAD protocol. Increasing the numberof gateways will increase the data packets delivered to the sink.Although number of data packets delivered to the sink isimproved in the EADGeneral protocol, we must take into accountthat these data packets are more correlated compared with thedata packets in the EAD protocol.

5. CONCLUSION

In this paper, we generalize the EAD protocol such that anynode in the network can be a gateway. An energy aware

mechanism to select gateways is proposed. The selectedgateways will initiate the building tree protocol. The proposedprotocol, EADGenereal is examined against the EAD usingsimulation. It shows significant improvements in terms of

larger network lifetime, less consumption energy and higherthroughput.

As future work, network size and the number of dataperiods must be studied to find the optimal network size andoptimal number of data paeriods for longer network lifetime,reduced energy consumption, and higher throughput.

6. ACKNOWLEDGMENT

This work is supported by King Fahd University of Petroleumand Minerals, Dhahran, Saudi Arabia.

7. REFERENCES

[1] K. M. Slivalingam, Tutorial: Wireless Sensor Networks,http://www.cs.umb.esub/-krishna

[2] I. F. Akylidiz, W. Su, Y. Sankarasubramaniam, and E.Cayirci, "A survey on sensor networks," EEE PersonalCommunications Magazine, vol. 40, no 8, pp. 102-114,August 2002.

[3] W. R. Heinzelman, A. P. Chandrakasan, and H.Balakrishnan, "An Application-Specific ProtocolArchitecture for Wireless Microsensor Networks," WEEETrans On Wireless Communications., vol. 1, no. 4, pp.

660-670, Oct. 2002.

[4] M. Stemm, R. H. Katz, "Measuring and reducing energyconsumption of network interfaces in hand-held devices,"EICE Transactions on Communications, vol. E80-B, no.

8, pp. 1125-1131, Aug. 1997.

[5] 0. Kasten, Energy Consumption, EldgenossischeTechnische Hochschule Zurich, [online]. Available:http://www.inf. ethz.ch/-kasten/research/bathtub/energy_consumption.html.

[6] A. Boukerche, X. Cheng, J. Linus, "A PerformanceEvaluation of a Novel Energy-Aware Data-CentricRouting Algorithm in Wireless Sensor Networks,"Wireless Networks, vol.11, pp. 619-635, 2005.

[7] W. Ye, J. Heidemann, and D. Estrin, Fellow, "MediumAccess Control With Coordinated Adaptive Sleeping forWireless Sensor Networks," EEE/ACM Transactions on

Networking, vol. 12, no. 3, pp. 493-506, June 2004.

[8] T. V. Dam and K. Langendoen, "An adaptive Energy-Efficient MAC Protocol for Wireless Sensor Networks,"in Proc. ACM Sensys'03, Los Angeles, CA, Nov. 2003,pp. 171-180.

[9] S. Mishar and A. Nasipuri, "An Adaptive Low PowerReservation Based MAC protocol for wireless sensor

networks," in Proc. IEEE International Conference on

Performance, Computing, and Communications, Phoenix,Arizona, April 2004, pp. 731 - 736.

[10] J. Li, G. Y. Lazarou, "A Bit-Map-Assisted Energy-Efficient MAC Scheme for Wireless Senor Networks," inProc. Third International Symposium on InformationProcessing in Sensor Networks 2004 (IPSN'04),,Berkeley, California, USA, April 26-27, pp. 55-60.

[11] K. Akkaya and M. Younis, "A Survey of RoutingProtocols in Wireless Sensor Networks, " in the ElsevierAd Hoc Network Journal, vol. 3/3 pp. 325-349, 2005.

120

x 104I__ I__I__ I_I__I__I_I

~ EAD (1 Period) I I I6 _ _EAD (5 Periods) _ _ __

EAD(10Periods) I I I

5 _ _ T -E EADGeneral (1 Period) _ _ iEAD,General(5 Periods) I II i

4 - - t ---------------;- EAD,General(l0Periods)

I I /. i

3 -~~~~~~~~~~~~~~~~~~~~~~-

II--t -l--_ t - ---t-I

,I

__N1-