WIRELESS COMMUNICATIONS AND MOBILE COMPUTINGWirel. Commun. Mob. Comput. 2009; 9:103–115Published online 7 February 2008 in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/wcm.599

An optimized forwarding protocol for lifetime extension ofwireless sensor networks

Mohammad Abdul Azim, M. Rubaiyat Kibria and Abbas Jamalipour∗,†

School of Electrical and Information Engineering, The University of Sydney, NSW 2006, Australia

Summary

Optimized routing (from source to sink) in wireless sensor networks (WSN) constitutes one of the key design issuesin prolonging the lifetime of battery-limited sensor nodes. In this paper, we explore this optimization problemby considering different cost functions such as distance, remaining battery power, and link usage in selectingthe next hop node among multiple candidates. Optimized selection is carried out through fuzzy inference system(FIS). Two differing algorithms are presented, namely optimized forwarding by fuzzy inference systems (OFFIS),and two-layer OFFIS (2L-OFFIS), that have been developed for flat and hierarchical networks, respectively. Theproposed algorithms are compared with popular routing protocols that are considered as the closest counterpartssuch as minimum transmit energy (MTE) and low energy adaptive clustering hierarchy (LEACH). Simulation resultsdemonstrate the superiority of the proposed algorithms in extending the WSN lifetime. Copyright © 2008 JohnWiley & Sons, Ltd.

KEY WORDS: wireless sensor networks (WSN); fuzzy inference system (FIS); energy-efficient routing

1. Introduction

Wireless sensor networks (WSN) are composedof spatially distributed low-end sensor nodes forcooperatively monitoring physical or environmentaldata such as temperature, pressure, humidity, andso on. Although primarily developed for militaryapplications, it has become an area of activeresearch in many diversified industries such as health,environmental, industrial, home automations, and soon [1]. Composed of a large number of nodes andconnected to the infrastructure via a single sink (ora very small number of sink nodes), WSN is largelyinvolved in collecting sensed data from different

*Correspondence to: Abbas Jamalipour, School of Electrical and Information Engineering, The University of Sydney, NSW2006, Australia.†E-mail: a.jamalipour@ieee.org

environments. Therefore, the dominant data flow isfrom sensor nodes to the sink although a small numberof transmissions do originate in the reverse direction(from the sink to the nodes) involving query andcontrol signaling. Because of the limited sensing,processing, and communicating potential of the nodes[2], the research into WSN has been dominated byinvestigations into routing protocols in order to extendthe collective WSN lifetime [2–4]. Note that thelifetime is significantly affected by the battery (orbattery power) of the sensors which is difficult toreplace in deployment scenarios such as battlefields,dense forests, etc. In addition, it is also affectedby the variation in the application-specific traffic

Copyright © 2008 John Wiley & Sons, Ltd.

104 M. A. AZIM, M. R. KIBRIA AND A. JAMALIPOUR

pattern, and different accessibility issues such asmobility, QoS, scalability, and so on. For example, thecontinuous traffic generated by a regular environmentalmonitoring application or the sudden burst of traffic bythe trigger-based military attack detection applicationwould impact the battery lifetime differently. Sincestraightforward incorporation of techniques from otherwireless networks such as ad hoc and cellularnetworks will not always be well suited due totheir inherent characteristics [2], new routing protocolsare warranted. Exploiting computations to reducethe amount of communication is one of the commonapproaches of achieving this goal.

Furthermore to maximize energy conservation, therouting is also designed as per the structure of theWSN layout (i.e., logical layout: flat and hierarchical).In flat networks where each node comprises ofhomogenous functionality, routing is straightforwardand is from source to sink over multiple hops. Incontrast, hierarchical networks, which form clusters (acollection of nodes in a virtual group (cluster formationmay change)), route data to the sink over multiplelogical planes. A number of high-end nodes (termedas cluster-heads (CHs)) carry out unique roles such asdata aggregation and cluster maintenance.

To address both these types of WSN layout, inthis paper we propose two differing routing protocolsthat prolong/extend the collective WSN lifetime byoptimizing the routing path. Here, collective lifetimeis defined as the time it takes for a node to diein WSN on account of cumulative data forwarding.Obviously, the node death is a direct result of batterydepletion from continuous data forwarding. Our maincontribution therefore lies in offering protocols thatare designed to prolong the WSN lifetime and canoutperform conventional algorithms. The first protocol,titled optimized forwarding by fuzzy inference systems(OFFIS), optimizes routing in flat networks. Fuzzyinference system (FIS) is utilized in optimizing thenext hop node selection in the forwarding pathbased on different routing metrics, mainly distance,battery power, and link usage. Although the utilizationof different metrics may incur longer delay andlonger path traversal, in comparison to existing routingprotocols the proposed algorithm offers promising per-formance, as demonstrated by the simulation results.Unfortunately, for large-scale networks flat routingdoes not scale well. The second protocol, titled two-layer OFFIS (2L-OFFIS), is proposed to address thisinherent problem. The hierarchy is organized into twodifferent logical planes where data forwarding traversesfrom individual nodes to a CH (in lower plane) and

among CHs to the sink (in upper plane). CH formationis assumed to be dynamic and scalable.

The remaining of the paper is organized as follows.Section 2 describes related works in literature forWSN routing, and optimization of particular metricsand data aggregation. Section 3 presents a detaileddescription of the OFFIS protocol. In Section 4, 2L-OFFIS algorithm is proposed. Section 5 discusses theadvantages and disadvantages of the protocols in detail.Simulation results and performance comparison withother protocols are presented in Section 6, followed bysome concluding remarks.

2. Related Works

There exists in literature a significant amount of workthat focuses on finding the optimal transmission pathfor the energy constrained sensor nodes in order toincrease the WSN lifetime. A review of some ofthe more closely related algorithms for both flat andhierarchical networks is therefore presented here tohighlight the significance of our work.

Flat networks: using distance as a cost metric,Reference [5] proposes a popular multi-hop routingprotocol termed minimum transmit energy (MTE) forflat sensor networks. MTE selects the next hop node inthe forwarding path depending on the required transmitenergy, which can be illustrated with an example.Suppose a source node A wants to send a packet to thedestination node B (sink) (as shown in Figure 1). NodesC and D are the intermediate nodes located within theradio transmission range of node A. According to MTEprotocol, node A calculates the costs of nodes C andD as costC = distance2

AC + distance2CB and costD =

distance2AD + distance2

DB, respectively. It thereforesends data to the node having the minimum cost. Un-fortunately, by considering only one metric (distance),the protocol fails to recognize other constraints, suchas battery power, link usage, etc., that impacts the loaddistribution among participating nodes.

A similar geographical distance-based approachhas also been adopted in the geographical multi-path

Fig. 1. MTE forwarding scheme.

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

AN OPTIMIZED FORWARDING PROTOCOL FOR WSN 105

routing (GMR) protocol proposed in Reference [6].The energy and mobility aware geographic routingprotocol (EM-GMR) in Reference [7] improves thisGMR concept by considering two additional metrics,that is, mobility and energy, while the forward pathselection is optimized by FIS [8]. The choice ofseveral metrics in place of simply distance (as isthe case in GMR) clearly enhances the performancemeasure. As such, similar multi-criteria principle hasbeen incorporated in the proposed OFFIS. Whilesome routing protocols proposed for ad hoc networkssuch as Reference [9] attempt to extend the networklifetime, the associated computational burden, and thea priori calculation of the entire path cost between thesource and the destination makes them undesirable forresource-limited WSN.

Hierarchical networks: for cluster-based hierarchi-cal networks, low energy adaptive clustering hierarchy(LEACH) [10] offers one of the most popular dataforwarding mechanisms. A distributed approach tocluster formation is adopted where the nodes selectthemselves as CHs. This is based on a predefinedpercentage of CHs or the CH density (P), set ordecided by the network provider prior to the clusterformation. CH selection is therefore carried out asfollows: (i) nodes generate a random number∼ [0,1],and (ii) if the number is less than a threshold valueT (n) then the node selects itself as a CH. T (n) isdefined by the expression, P

1−P ·(rmod (1/P))where r is

the current cluster formation round and rmod (1/P)denotes modulo operation. Cluster formation roundtrigger can be based on time online, number of packetsforwarded, and so on. A nodes does not participate inthe CH selection process if it has been a CH in thelast 1/P rounds. Upon broadcast of the CH selection,non-CH nodes join the cluster based on received signalstrength. Collected data from the nodes are aggregatedby the CH and forwarded to the sink. Figure 2 depictsan example of the direct communications experiencedin the LEACH sensor bed.

Instead of a distributed CH selection (as in LEACH),a more centralized method was proposed by LEACH-C in Reference [11] where the CH selection iscarried out by the sink. In this scheme, the sinkhas a priori information about node location andenergy, whereupon it calculates the average energyof the nodes. Nodes having energy level below theaverage energy are not considered as a prospective CH.The cluster is formed in such a way that the total amountof energy needed to transmit data from the non-CHs tothe CHs is minimized. Other than the cluster formation,the rest is the same as in LEACH.

Fig. 2. Transmissions in LEACH protocol.

Another variant of LEACH titled threshold-sensitiveenergy efficient sensor network protocol (TEEN) [12],improves performance as compared to LEACH bylimiting the number of generated packets through theintroduction of two thresholds, namely a hard threshold(Th) and a soft threshold (Ts). Data are sent to the sinkonly if the current value (number of packets) is greaterthan Th or if the difference between the current and theprevious value is greater or equal to Ts.

Unfortunately, in all the above-mentioned protocols(LEACH, LEACH-C, and TEEN), data are forwardedboth by nodes and CH through direct transmissionsinstead of hop-by-hop. Since these protocols aredesigned for homogenous sensor networks with lowpowered nodes, their direct link connectivity to the sinkmakes it difficult to offer coverage over a large area. Asolution to this problem is proposed in Reference [13]where the clusters are fixed and the CHs are high-end nodes with higher transmission and battery power.However, this approach will incur high monetaryinvestment by the network operator in deploying suchhigh performance nodes and may not be popular. Inaddition, it also suffers from higher transmission coston account of using longer hops.

An improvement over LEACH was proposed by thealgorithm titled power efficient gathering in sensorinformation systems (PEGASIS) [14]. It providesa near optimal chain-based mechanism where eachnode forwards data only to its nearest neighbor, dataaggregation is carried out, and a designated nodesends the fused data to the destination (i.e., sink).However, the improved performance comes at the costof increased data delivery time [15]. In addition, italso suffers from direct data forwarding between thedesignated node (in effect CHs) and the sink.

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

106 M. A. AZIM, M. R. KIBRIA AND A. JAMALIPOUR

Remaining battery power has also been consideredin deriving routing protocols for hierarchical networks.For example, in event driven clustering algorithm(EDC) [16] (similar in principle to hierarchical batteryaware routing (H-BAR) protocol in Reference [17]),authors propose selection of CHs depending on thehighest remaining energy where CHs aggregate androute data to the sink through mobile gateways. Asimilar method is adopted in the hop-by-hop dataforwarding scheme of Reference [18] where the nodeswith high remaining energy act as gateways. Thegateways provide services such as collecting the statusof the nodes, updating the routing tables, and so on,but most importantly gateways select the best path foreach node by using the centralized Dijkstra’s algorithm[19]. Other notable schemes include the two-phaseclustering (TPC) algorithm in Reference [20] whichsets up direct links from nodes to the CHs in phase-1and searches nodes closer than CHs for setting up smallrelay links in phase-2. Whereas the direct links are usedfor time critical data aggregation, relay links are usedto save energy during non-time critical data gathering.

Based on the above discussion, we have identifiedseveral key areas of routing protocol design that can ineffect extend the WSN lifetime (as highlighted below):

� Small multiple hops: the energy consumed forthe transmission is proportional to the square of thedistance from the sender to the receiver. Therefore,instead of a single large hop multiple short hopsare preferable to reduce the transmit energy cost.

� Shortest path: an imaginary line connecting senderand receiver node is considered the shortest path ofcommunication. Forwarding packets along this lineis more proficient than a detour.

� Highest remaining energy: nodes having greaterremaining energy can participate more in commu-nications than the nodes having smaller amount ofpower thereby extending the WSN lifetime.

� Load distribution: in the event that the concentrationof traffic in some particular areas is more than that ofother areas, using shortest path will cause implosionalong the selected path. Uniform distribution oftraffic can improve the situation to a greater extent.

We have developed OFFIS [21] as a solutionfor flat sensor networks that optimizes the routingpath according to all the above-mentioned criteriathrough a single distributed algorithm. In a latersection, we will show that this optimization algorithmenhances the lifetime of the network. For large-scaledeployment where WSN contains a large number of

nodes distributed along a field, scalability becomesan issue. In such cases, highly correlated data canbe aggregated for compression purposes therebyleading to further reduction of the transmit energy. Ashierarchical routing techniques are considered superiorin addressing this problem, we have developed 2L-OFFIS [22] as an extension of the OFFIS protocolthat supports non-delay critical applications such asenvironmental data collection and so on. Here, 2L-OFFIS operates in a two-level hierarchical fashion incluster-based WSN. A categorical description of thesetwo protocols are presented in the following sections.

3. Optimized Forwarding by FuzzyInference Systems (OFFIS)

Assumptions:

� Available routing metrics : shortest path, minimumdistance to next hop node, battery power, and linkusage (number of packets transmitted previously bythe same link).

� Accessible battery power (trivial to access the batterylevel from the sensor node).

� A priori location information (available from, forexample, global positioning system (GPS), assisted-GPS (A-GPS), and so on). Work is underway tooffer more efficient localization methods withoutwhich the sensed data will be rendered useless.Each application will need the location information,at least in a coarse form, to monitor an interestedarea. Note that in the proposed routing algorithmthe location information exchanged between theparticipating nodes are fairly reliable. This helpsin avoiding the formation of routing loops in theforwarding path.

� Nodes are deployed densely enough such that nodehas at least one node in the forwarding path (i.e.,guaranteed delivery). A dense deployment of nodesin WSN is a common approach of deployment andadds robustness to the system.

3.1. Protocol Operations

As per the protocol, sensor nodes (in proactiveor reactive mode) gather the local metrics forthe surrounding area through inter-node messageexchange. In reactive implementation, nodes onlydisseminate queries for their single hop neighbor’sinformation whenever data are needed to be transferred.Alternatively, in proactive implementations neighbors

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

AN OPTIMIZED FORWARDING PROTOCOL FOR WSN 107

Fig. 3. Metrics for next hop node selection in WSN.

are updated by the periodic beaconing/broadcastingof the routing table. Note that this informationexchange (either proactive or reactive) relates to onlyneighboring nodes (generally limited number) withinthe transmission range, and for battery usage andlocation information only. As such, it eliminates thenecessity to build a mirror of the entire network andthus limits the signaling overhead. While sending data,protocol selects the optimal next hop node through theFIS from multiple candidates, followed by adjustmentsof transmit power in accordance to the distance from thecandidate node, and subsequent forwarding of the datapackets. Since the correlation between different metricsis difficult to model mathematically, FIS facilitates jointoptimization.

Let us assume that a source node needs to senda data packet to the destination sink as shown inFigure 3. The magnified portion of the figure (within thetransmission range denoted by the dotted lines) showsone of the candidate nodes (n1), the shortest distance(ds1) to the imaginary source–sink line, and inter-nodedistance (d1). To limit the load on FIS, source discardsthe nodes that are not present in the forwarding path(denoted by lightly shaded nodes) thereby reducingthe number of potential candidates to three, that is, n1,n2, and n3. Table I shows the routing metrics used byFIS in selecting the optimal next hop node where di

denotes the distance of the candidate nodes from thesource, dsi denotes the distance of the candidate nodesfrom the shortest path while pi and li denote the power(battery usage) of a candidate node and the link usage,

Table I. Metric routing table.

Node Distance Distance Power Link usageni di dsi pi li

n1 d1 ds1 p1 l1n2 d2 ds2 p2 l2n3 d3 ds3 p3 l3

respectively: i = 1, 2, 3, · · · , n. While di, dsi, li can belocally estimated (based on location information andnumber of packet transmission over the correspondinglink), information regarding pi needs to be periodicallyupdated from the neighbors. As such, all four metricsneed to be considered in the selection process. Notethat the metrics are normalized in order to implementthe fuzzy rules.

3.2. Membership Functions

The input–output relationships of the metrics is definedby their corresponding membership functions (MFs).MF can be defined as a linear relationship betweenthe metrics and their reported cost. A brief discussionof the MF construction rules for each cost metricis therefore given below. Upon deriving the MF,projections are drawn from the input–output curvewhich forms trapezoids (as shown in Figure 4). Tomake a decision, all different trapezoids are added

Fig. 4. FIS-based node selection: weighted averagecalculation from individual MFs.

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

108 M. A. AZIM, M. R. KIBRIA AND A. JAMALIPOUR

up, their weighted averages are calculated, and thedesired node is selected based on the minimumaverage.

Rule–1: Distance from the node—for small hops,transmission cost follows the first order radio model(approximately), that is, the power is proportionalto the square of the distance. Consequently, the MFof the distance (from the source node) is representedby the curve shown in Figure 4. Here, distances d1,d2, and d3 are the inputs of the MF and the projectedtrapezoids, that is, the weights of the correspondingnodes are the outputs from the particular MF. Theheight of the trapezoids (denoted as hd) for the ithnode is defined as hdi = kdist · d2

i , where kdist is aconstant for all values of di = [0, 1] (normalized):i = 1, 2, 3, · · · , n.

Rule–2: Distance from the shortest path—the MF,in this case, is the same as rule–1 because the metrichere is also a distance. Inputs ds1, ds2, and ds3 alongwith the outputs, that is, the corresponding trapezoidsare shown in Figure 4. Here, the heights of thetrapezoids are expressed as hdsi = kdist · d2

s for di =[0, 1] : i = 1, 2, 3, · · · , n, kdist being the same as inrule–1.

Rule–3: Battery usage—it can be considered thatdata forwarding will impose a bigger burden uponnodes with lesser battery power than those with largerenergy. MF is therefore set in such a way that for up to30 per cent of the battery usage there exists little effecton the cost of the decision algorithm. As the usage goeshigher (30–70 per cent), it shows moderate resistanceto forwarding and when it is at 70–100 per cent itshows the highest resistance to forwarding a packet.Here, the power MF curve is simply made up of threeline segments corresponding to the percentage powercost. We take the simple line equation y = mx + c inrepresenting the segments where m represents the slopeand c represents the y-intercept of the line. We setthe slopes of the line segments as 0.333, 0.75, and 2which are selected to reflect the higher resistance ofbattery usage to data forwarding (e.g., slope 2 for 70–100 per cent usage whereas slope 0.333 for 30 per centof usage). This results in y-intercept values of 0,−0.125, and −1, respectively. The arbitrary selectionof m and current power usage enables the networkprovider or operator to customize the routing protocolas per the service requirements. Obviously with a largerm, the nodes will become more sensitive to depletedbattery level. On the other hand, if enough batterypower is available then the constraints on the valueof m can be relaxed to a degree. This entails theincorporation of predetermined separate MF functions

by the operators for different levels of networkloads.

Hence, the heights of the trapezoids (denoted as hp)can be expressed as

hpi =

0.333 · kpower · pi, if 0 < pi ≤ 0.3

0.75 · kpower · pi − 0.125, if 0.3 < pi ≤ 0.7

2 · kpower · pi − 1, if 0.7 < pi ≤ 1

where kpower is a constant for all values ofpi, i = 1, 2, 3, · · · , n. Figure 4 shows the corres-ponding MF.

Rule–4: Link Usage—according to this rule, withinthe proposed algorithm the more a path is used (causingbattery energy depletion) the more it becomes reluctantto forward a packet. Hence, the link usage MF becomesa linear function as shown in Figure 4 and the heightsbecome hli = klink · li where klink is a constant for all li,i = 1, 2, · · · , n. Due to normalization (considering themaximum value of X–Y axis in the corresponding MFbeing one), the value of kdist, kpower, and klink is one.Note that the slope of the linear function is governedby the frequency of link usage which may vary over acertain range. Hence, similar to the battery usage, thisslope can be defined a priori by the operator in relationto the traffic load.

3.3. Weighted Average

The area of the trapezoids drawn earlier from differentMFs can be expressed as follows:

Ai,j = (1 − (1 − hi,j)2)

2(1)

where A denotes the area of the trapezoid and j is the jthMF. The decision (selection of the best node to forwarddata) is then made based on the costs derived by theMF rules and their weighted average. For the differenttypes of MF output, the weighted average can bewritten as

Avi =∑g

j=1 wj · Ai,j∑gj=1 wj

(2)

where g is the number of different metrics: i =1, 2, · · · , n, j = 1, 2, · · · , g, and wj is the weight ofthe jth metric. The operator may define wj so asto prioritize one metric over another thereby facili-tating the inception of service-oriented applications.However, to simplify the calculations, in this paper

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

AN OPTIMIZED FORWARDING PROTOCOL FOR WSN 109

we consider all metrics to have the same priority,that is, we take the relative weights of the individualmetrics (wj) as one. Such an assumption enables therouting approach to consider all metrics in a fairand impartial manner. Furthermore, Av1, Av2, andAv3 denote the corresponding weighted average ofthe nodes n1, n2, and n3, respectively. Consideringd1 < d3 < d2, ds3 < ds2 < ds1, p3 < p2 < p1, andl2 < l1 < l3, we get, Av1 > Av2 > Av3, that is, n3becomes the optimal next hop node for the source.Figure 4 shows the FIS-based weighted averagecalculation (denoted by different sized black circles)for the candidate nodes. It uses the heights of thetrapezoids that are derived from the individual MFs(corresponding to different metrics). The flowchartin Figure 5 summarizes the entire operation ofOFFIS.

Fig. 5. Flowchart for OFFIS operation.

4. Two-Layer Optimized Forwarding byFuzzy Inference Systems (2L-OFFIS)

Two-layer OFFIS (an extension of OFFIS) has beendeveloped to accommodate the additional necessity ofthe aggregation support in a large-scale deployment.The sensor bed is divided into two logical planeswhere the lower plane (termed as forwarding layer one(FL1)) forwards data packets from individual nodesto the CHs (functions as a bridge between the layers),while the upper plane (i.e., forwarding layer two (FL2))routes these packets from CHs to the sink. The logicalseparation of the routing mechanism simplifies therouting, distributes the load sharing among nodes andCHs, and as demonstrated later prolongs the collectiveWSN lifetime. The principal functionalities of 2L-OFFIS can be summarized by the following threecategories:

� Formation of clusters—the predefined CH percent-age (P) on which the cluster density will be based onis set or decided prior to the formation of the cluster.Based on this percentage, nodes choose themselvesrandomly as CHs. Although in 2L-OFFIS the clusterformation follows the method proposed in LEACH,schemes such as LEACH-C and others can alsobe used. Upon selection of CH, nodes broadcastthe decision to the neighborhood (i.e., neighboringnodes within transmission range). The neighboringnodes join the cluster based on the signal strength ofthe received broadcasts.

� Sleep strategy—CHs assign the broadcast schedulefor every node within that particular cluster wherethe schedule assignment depends on the distance ofthe nodes from the CHs. Distant nodes obtain theearlier slots whereas the closer nodes get the laterones. Nodes maintain their transceiver in sleep modeexcept during their assigned time slots since they areonly allowed to broadcast during their designatedslots. Nodes estimate the set of neighboring nodes forwhich they are in the forwarding path and maintaintheir receiver on only for the corresponding time slotsof the calculated set.

� Data forwarding—in the two layer system, forward-ing of data begins with sending the sensed datafrom the sensor nodes to the corresponding CHsover FL1. After receiving packets from every nodein the cluster, CH aggregates the data and forwardsthe amalgamated information to the sink. Note thatboth FL1 and FL2 use OFFIS protocol for nexthop node selection. Figure 6 presents a simplifiedlayout of 2L-OFFIS-based WSN where the two

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

110 M. A. AZIM, M. R. KIBRIA AND A. JAMALIPOUR

Fig. 6. Data forwarding in layered WSN network.

layers (FL1 and FL2) are clearly marked out. Eventhough both the layers adopt similar mechanism inforwarding data packets, FL2 density is significantlylesser than FL1 thereby causing larger distancetraversal between nodes (i.e., CHs). A flowchartof the 2L-OFFIS protocol operation is shown inFigure 7.

5. Impact of the Proposed Protocols

Although 2L-OFFIS provides optimized forwarding aswell as data aggregation in hierarchical network ascompared to OFFIS (limited to flat networks), both

Fig. 7. Flowchart for 2L-OFFIS operation.

the protocols can be regarded as efficient and scalablein terms of their closest counterparts. Below, weprovide a comprehensive examination of the proposedprotocols in terms of the size of the networks, IDmaintenance, routing table maintenance, aggregation,deaths in the network, optimized decision for dataforwarding, number of hops, and processing power.

5.1. Network Size

As WSN size grows beyond limit, it becomes infeasibleto run flat routing scheme [23], necessitating theadoption of hierarchical routing protocols whichare considered as more scalable and efficient incomparison. Two-layer OFFIS addresses this limitationof OFFIS. Moreover, since the size of the network isnot restricted to the transmission range of the sensors(as in LEACH), hop-by-hop data forwarding (as in 2L-OFFIS) can cover a large geographical area.

5.2. ID Maintenance

It is not desirable that a large sensor network willhave a global addressing scheme because of the hugemaintenance overhead. Both the OFFIS and 2L-OFFISdo not need to maintain the ID, thereby saving thecost of global addressing mechanism and making thenetwork scalable.

5.3. Routing Metrics and Tables

Selection of the metrics within the routing table(distance, battery power, and link usage) is prudent andjustifiable. This implies, as stated earlier, that the metricinformation is available to the nodes. The availabilityis in the form of small table (containing metrics of onlysingle hop neighbors) transfer between neighboringnodes. The formulation of the metric table for singlehop neighbors instead of multi-hop saves the storagecost as well as the communication cost (e.g., transmit-receive energy and bandwidth).

5.4. Aggregation

Instead of the actual sensed data, often an aggregatedvalue is of more importance to the operator such as theaverage, maximum, minimum, etc. By incorporatingsuch aggregation in its forwarding scheme, 2L-OFFISeliminates redundancy, minimizes the number oftransmissions, and subsequently saves a significantamount of energy.

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

AN OPTIMIZED FORWARDING PROTOCOL FOR WSN 111

5.5. Deaths in the Networks

It is justifiable to transmit data over small multi-hops,as the energy required to transmit data is proportionalto the square of the distance. The downside of thesmall multi-hop scheme for static sensor networksis that the traffic increases exponentially near thesink. Consequently, nodes in the surrounding regiondie first. Too many deaths in that particular areacauses network disintegration even though a largepercentage of the nodes are still working. The situationis a lot better if a mobile sink is deployed inthe networks. In Reference [24], authors show thatthe lifetime can be extended if the sink movesalong the periphery of the sensor field. But in real-life scenario, we may not always encounter such aworking environment. A much improved performanceis offered by LEACH where the distant nodes die first.However, since LEACH forwards data by transmittingdirectly through long-hops from CHs to sink, itutilizes more energy and experiences higher numberof node deaths compared to 2L-OFFIS. For boththe cases (LEACH and 2L-OFFIS), the death of aCH disrupts the sending of the data to the sinkfrom each and every node inside that particularcluster. This situation will persist until the next clusterformation.

5.6. Optimized Decision for Data Forwarding

FIS (well known and easy to implement optimizeddecision making algorithm), used by the proposedprotocols and based on several cost metrics,facilitates an optimal selection process for both flatand hierarchical networks. The metrics are identifiedfollowing an extensive literature review (as describedin Section 2). By preferring short hops and shortestpath over long-hops and detours, the algorithms savesignificant transmission energy (alternatively batteryenergy). Higher remaining powered sensors contributemore in forwarding data, thereby equitably distributingthe energy throughout the sensor bed. In addition,using link usage as a contributing metric allows fairdistribution of the workload among the participatingnodes and makes the protocol more efficient inextending the WSN lifetime.

5.7. Number of Hops

Though multiple hops have many favorable criteria(e.g., less distance, less transmission power, and so on),it can result in delayed data collection which makes the

proposed protocols more desirable for non-time criticalapplications.

5.8. Processing Cost

Due to the algorithmic complexity, the nodes requiresome amount of battery power while running the FIS.However, the energy required to run the algorithm (or inother words the processor) is significantly less than thetransmit energy. For example, as is widely recognized,the amount of power required to transmit 1 bit at adistance of 100 m is equivalent to the amount of powerthat is required to run millions of CPU instructions [2].The proposed protocol exploits this relation and uses itfavorably to calculate the optimized transmission. Thetechnique here is therefore to spend small amount inorder to save more.

6. Performance Comparisons

To evaluate the performance measure, the proposedprotocols are compared with their nearest counterparts(i.e., widely referenced MTE and LEACH) in theassociated deployment zone, that is, in flat andhierarchical networks. Comparisons with MTE andLEACH offer straightforward performance measurethat can be extended to other protocols that haveundergone similar comparisons.

Simulations are carried out in MATLAB for a 200 ×200 m2 sensor bed where 200 nodes are randomlydeployed across the field. Radio transmission ismodeled with the first order radio model that has beenquite extensively used in routing protocol evaluation[10]. It enables the system to derive the transmitand receive energy cost associated with data transferover the wireless interface. For example, using thismodel the transmit and receive cost of sending a k bitpacket over a distance d is written as ET (k) = Eb ·k + Em · k · d2 andER(k) = Eb · k, respectively. Here,Eb denotes per bit energy dissipation in transmit orreceive circuitry and has a value of 50 nJ/bit while Em

refers to per bit energy dissipation per meter square forthe amplifier to achieve acceptable signal-to-noise ratio(SNR) and has a value of 100 pJ/bit/m2.

A categorical evaluation of the protocols for thissystem model is therefore presented below.

OFFIS—for OFFIS, the focus of the simulationis: (a) to observe the behavior of the protocol withdifferent combinations of the routing metrics, (b) toevaluate the performance of the protocol with variousnode densities, and (c) to compare its performance

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

112 M. A. AZIM, M. R. KIBRIA AND A. JAMALIPOUR

with MTE. Two different types of sink deployments areconsidered, that is, static sink (in the central location)and mobile sink (randomly moving in the sensor field).The relative position of the sink directly affects theWSN lifetime. For example, a mobile sink along theperiphery of the field induces less resource usage fornodes near the sink compared to a static configuration[24]. In OFFIS, sink mobility is simulated in a field withnormalized X–Y axis and considers constant velocityduring successive measurements, random time interval,and random movement over [0, 2π].

As was stated in Section 2, performance enhance-ment of EM-GMR over GMR is influenced by thenumber of metrics (three as opposed to one) utilizedin deriving the optimal path. Since OFFIS adopts asimilar principle, the effect of different combination ofmetrics on protocol performance provides a measureof performance enhancement, as shown in Figure 8.The figure reflects this performance measure in termsof number of transmissions against number of deadnodes as the sink changes its location from mobileto static. Although time related to varying levels(percentage) of node deaths is a good measure forillustrating the lifetime extension of a WSN, theproposed routing algorithm provides performancecomparison of actual node deaths with respect to datapacket transmission. Obviously, the more deaths theWSN encounters, the more it is likely to experiencenetwork partitioning. Alternatively, the fewer deathsthe WSN encounters, the longer the network canprovide data delivery capability (as demonstrated byincreased packet transmission). As such, the results

Fig. 8. Performance comparison between combinations of theoptimized routing metrics with a mobile sink and a static sink

in the central location.

Fig. 9. Performance comparison between different sensordensities in the network.

can be considered as a fair representation of theperformance improvement of the proposed algorithm.Note that in the figure, number of transmissions denotethe total packet transmissions accumulated throughoutthe network for successive node deaths. Differentmetric combinations are considered which includes dds

(considers metrics d and ds), ddsl (considers metrics d,ds, and l), ddsp (considers metrics d, ds and p), andlastly ddspl (all the metrics). It can be seen from thefigures that for both the sink conditions the protocolperforms best when all the metrics are considered.This is primarily due to the fair load distribution andenergy dissipation among all the nodes in a distributedmanner.

To formulate a bound on sensor field density, theprotocol evaluation is carried out with different valuesof node density, as shown in Figure 9. Same asbefore, performance measure in this case is definedin terms of number of transmissions against numberof dead nodes. It can be seen from the figure thatprotocol performance increases with increased nodedensity (from 200 to 400 nodes). Intuitively, this canbe attributed to the fact that when node density inWSN increases, more nodes become available aroundthe sink to perform the forwarding task. In such acase, because of the load distribution among increasednumber of nodes near the sink, nodes take longer todie thereby increasing the collective WSN lifetime.However, after a certain limit (beyond 400 nodes),further increase in the density results in performancedegradation (as is the case when density is 500 nodes).Note that the nodes around/near the sink are morevulnerable to energy depletion due to their participationin the forwarding process. An increase in density

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

AN OPTIMIZED FORWARDING PROTOCOL FOR WSN 113

beyond a certain limit therefore makes the data loadacross the field high enough to counter the positiveeffects of increased presence of nodes in the centrallocation. Also note that due to the random sensor nodedeployment and the random nature of data generationspeed, the curves do not necessarily show similarbehavior pattern. For example, in Figure 9, the curvedepicting node density of 400 clearly demonstratessomewhat different behavior than the other densitiesprior to 3 per cent node deaths. However, as moredata are generated in the network, the randomness inthe behavior pattern diminishes and exhibits similarnature as others (as shown in the figure beyond3 per cent node deaths).

Having evaluated the necessity and impact ofdifferent cost metrics combinations on WSN lifetimeextension, the OFFIS protocol is then compared withMTE for both static and mobile sink conditions.Figure 10 exhibits the performance comparison ofthe two protocols. As can be seen from the figure,OFFIS enjoys a distinctive superiority over MTE. Thisis mainly because of the lack of proper distributionof energy dissipation among the participating nodesin MTE. While OFFIS considers a range of metricfunctions in selecting the optimal next-hop node,MTE considers only distance as a cost functionthereby offering sub-optimal selection process. Whilethe performance of OFFIS far outperforms MTE,intuitively we can surmise that with better MFs thisperformance can be further enhanced.

Two-layer OFFIS—for 2L-OFFIS, the simulationsare carried out with identical sensor field layout and

Fig. 10. Performance comparison between OFFIS and MTEprotocols with a mobile sink and a static sink in the central

location.

Fig. 11. Comparison between 2L-OFFIS and LEACH forenergy usage in successive rounds.

number of nodes, as in OFFIS. Cluster formation(termed as rounds) takes place after every 1000transmissions from each of the 200 nodes, mainly bysequential packet transmission. The percentage (P) ofa sensor node becoming a CH is set to 20 per cent. Thetransmission range of FL1 and FL2 are chosen as 40and 80 m, respectively. Both 2L-OFFIS and LEACHare compared with the above configuration.

Figure 11 shows the comparison between theaccumulated (or cumulative) energy used in the wholesensor bed by the protocols. The slope of the curvesat a particular round represents the rate at whichenergy usage changes. It is clearly evident from thefigure that 2L-OFFIS outperforms LEACH in energyconsumption over different timescales (construed asrounds). Note that only a portion of the curves (upto round 10) reflect comparative power usage by theprotocols. This is because after round 10, on accountof dead nodes, the slope of the LEACH curve reducesreflecting less traffic generation by fewer active nodes.This makes it difficult to offer a comprehensive analysisof the system performance.

Simulations also reveal that the first node in 2L-OFFIS dies (in effect causes sensor field partitioning)after the 25th cluster formation (or rounds). Byrunning LEACH for equal number of transmissions andsubsequently measuring the energy usage, we find that2L-OFFIS outperforms LEACH in terms of remainingnodes and their battery power. For the same amount ofdata transmissions, 2L-OFFIS requires only one-sixthof the battery power of LEACH. Sixty per cent of thenodes are dead in LEACH whereas only one node diesin 2L-OFFIS at the same time reinforcing its claim ofsuperiority over LEACH.

A later variant of LEACH, that is, LEACH-C isshown to perform better than the original due to its

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

114 M. A. AZIM, M. R. KIBRIA AND A. JAMALIPOUR

superior cluster formation mechanism [11] (as wasstated in Section 2). In a centralized manner, the sinkderives the k number of optimal CHs through simulatedannealing method. This method attempts to minimizethe amount of energy needed for data forwardingby non-CH nodes to CH by minimizing the totalsum of the squared distances between them and theirnearest CHs. All the rest being the same, intuitivelywe can expect that adopting LEACH-C’s clusterformation in 2L-OFFIS (since any viable method isapplicable) would enable the protocol to provide betterperformance. Strictly speaking, the multi-hop dataforwarding mechanism throughout the 2L-OFFIS isdesirable than the direct transmission from CH to sink(as is the case in LEACH-C). Not only does it obviatethe constraints of higher transmission energy for longerdistance, it also facilitates large-scale deployment.

7. Conclusions

Limited battery energy of sensor nodes necessitates theformulation of a routing mechanism that can prolongthe WSN lifetime. To address this problem, in thispaper we proposed two differing algorithms for flat andhierarchical networks, respectively. For comparisonpurposes, the protocols were compared with popularalgorithms such as MTE and LEACH. Despite datatraversal over increased number of hops, the resultsclearly demonstrated the enhanced performance of theproposed protocols in extending the WSN lifetime.This can be largely attributed to limiting the energy costand distributing traffic load evenly among participatingnodes (through the adoption of multiple cost metrics).Although FIS has been incorporated to optimize thenext-hop node selection, other optimization algorithmssuch as analytical hierarchy process (AHP)—greyrelational analysis (GRA)-based methods can also offeralternative solutions. Our future work will includesuch an AHP–GRA-based mechanism that addressesdata transfer for both time and non-time criticalapplications. Here, AHP will provide the relativeweights of different cost metrics through a multi-criteria-based decision algorithm, while GRA willrank the candidate nodes and select the optimal nexthop node by comparing the similarity between eachoption and the ideal reference vector. Hence, themore similarity there is, the more preferable willthe subsequent selection. Since the derivation of therelative weights of the cost metrics can be operator-defined, the proposed framework will enable the WSNoperators to offer service-centric applications that are

governed by varying levels of end-to-end delay of datadelivery.

References

1. Romer K, Friedemann M. The design space of wirelesssensor networks. IEEE Wireless Communications 2004; 11(6):54–61.

2. Al-Karaki JN, Kamal AE. Routing techniques in wireless sensornetworks: a survey. IEEE Wireless Communications 2005; 11(6):6–28.

3. Ibriq J, Mahgoub I. Cluster-based routing in wireless sensornetworks: issues and challenges. In International Symposium onPerformance Evaluation of Computer and TelecommunicationSystems (SPECTS) 2004; pp. 759–766.

4. Akyildiz IF, Su W, Sankarasubramaniam Y, Cayirci E. A surveyon sensor networks. IEEE Communications Magazine 2002;40(8): 102–114.

5. Meng T, Volkan R. Distributed network protocols for wirelesscommunication. In IEEE International Symposium on Circuitsand Systems (ISCAS), Vol. 4, May 1998; pp. 600–603.

6. Jain R, Puri A, Sengupta R. Geographical routing using partialinformation for wireless ad hoc networks. IEEE PersonalCommunications 2001; 8(1): 48–57.

7. Liang Q, Ren Q. Energy and mobility aware geographicmultipath routing for wireless sensor networks. In IEEE WirelessCommunications and Networking Conference (WCNC), Vol. 3,New Orleans, LA, 13–17 March 2005; pp. 1867–1871.

8. Mendel JM. Fuzzy logic systems for engineering: a tutorial.Proceedings of the IEEE 995; 83(3): 345–377.

9. Toh C-K. Maximum battery life routing to support ubiquitousmobile computing in wireless ad hoc networks. IEEECommunications Magazine 2001; 39(6): 138–147.

10. Heinzelman W, Chandrakasan A, Balakrishanan H. Energy-efficient routing protocols for wireless micorsensor networks.In 33rd Hawaii International Conference on System Sciences(HICSS), Maui, January 2000; pp. 3005–3014.

11. Heinzelman W, Chandrakasan A, Balakrishanan H. Anapplication-specific protocol architecture for wireless microsen-sor networks. IEEE Transactions on Wireless Communications2002; 1(4): 660–670.

12. Manjeshwar A, Agarwal DP. TEEN: a routing protocol forenhanced efficiency in wireless sensor networks. In IEEEParallel and Distributed Processing Symposium (IPDPS), 23–27April 2001; pp. 2009–2015.

13. Hempel M, Sharif H, Raviraj P. HEAR-SN: a new hierarchicalenergy-aware routing protocol for sensor networks. In 38thAnnual Hawaii International Conference on System Sciences(HICSS), Hawaii, 03–06 January 2005.

14. Lindsay S, Raghavendra C. PEGASIS: power efficient gatheringin sensor information systems. In IEEE Aerospace Conference,Vol. 3, Big Sky, Montana, March 2002; pp. 1125–1130.

15. Tabassum N, Mamun QEK, Urano Y. COSEN: a chainoriented sensor network for efficient data collection. In IEEEInternational Conference on Information Technology (ITCC),April 2006; pp. 262–267.

16. Zeng-wei Z, Zhou-hui W, Huai-zhong L. An event-drivenclustering routing algorithm for wireless sensor networks.In IEEE Intelligent Robots and Systems, Vol. 2, Japan,28 September–2 October 2004; pp. 1802–1806.

17. Musunuri R, Cobb JA. Hierarchical-battery aware routingin wireless sensor networks. In IEEE Vehicular TechnologyConference (VTC), Vol. 4, Dallas, TX, 25–28 September 2005;pp. 2311–2315.

18. Younis M, Youssef M, Arisha K. Energy-aware routing incluster-based sensor networks. In Modeling, Analysis and

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm

AN OPTIMIZED FORWARDING PROTOCOL FOR WSN 115

Simulation of Computer and Telecommunication Systems, FortWorth, Texas, 11–16 October 2002; pp. 129–136.

19. Cormen TH, Leiserson CE, Rivest RL, Stein C. Introduction toAlgorithms (2nd edn). MIT Press and McGraw-Hill: Cambridge,MA, 2001.

20. Choi W, Shah P, Das SK. A framework for energy-saving data gathering using two-phase clustering in wirelesssensor networks. In International Conference on Mobile andUbiquitous Systems: Networking and Services, Boston, MA, 22–26 August 2004; pp. 203–212.

21. Azim MA, Jamalipour A. Optimized forwarding for wirelesssensor networks by fuzzy inference system. In IEEEInternational Conference on Wireless Broadband and UltraWideband Communications (AusWireless), Sydney, 13–16March 2006.

22. Jamalipour A, Azim MA. Two-layer optimized forwarding forcluster-based sensor networks. In IEEE International Sympo-sium on Personal, Indoor and Mobile Radio Communications(PIMRC), Helsinki, 11–14 September 2006.

23. Hong X, Xu K, Gerla M. Scalable routing protocols for mobilead hoc networks. IEEE Network 2002; 16(4): 11–21.

24. Luo J, Hubaux J. Joint mobility and routing for lifetimeelongation in wireless sensor networks. In IEEE Conference onComputer Communications (INFOCOM), Vol. 3, Miami, 13–17March 2005; pp. 1735–1746.

Authors’ Biographies

Mohammad Abdul Azim [S’05] ob-tained his B.Sc. Engineering (Electricaland Electronics) and M. Sc. Engineering(Computer and Telecommunications)from Chittagong University ofEngineering and Technology (CUET),Bangladesh, and the University ofWollongong, Australia, in 1996 and2000, respectively. He is currently

working toward a Ph.D. in Telecommunications. His mainarea of research includes wireless sensor networks, energyefficient routing algorithms, wireless LAN security, etc.

M. Rubaiyat Kibria [S’00] receivedhis Bachelors in Engineering (Electricaland Electronics), Master of Engineering(Research), and Ph.D. in 1997, 2002,and 2007, respectively. He was awardedthe Prime Minister’s Gold Medal foracademic excellence in the bachelorsprogram. He has been a reviewer formany IEEE conferences and journals,

and has authored many publications. Currently, he isworking as a Research Fellow at the School of Electricaland Information Engineering, the University of Sydney,Australia. His main area of research includes wirelessnetwork design and architecture, sensor networks, wirelessmesh networks, and network security.

Abbas Jamalipour holds a Ph.D. fromNagoya University, Japan. He is theauthor of the first book on wirelessIP and two other books, and hasco-authored six books and over 180technical papers, all in the field ofmobile communications network. He isa Fellow of IEEE (for contributionsto next generation networks for traffic

control), a Fellow of Institute of Engineers Australia; anIEEE Distinguished Lecturer; the Editor-in-Chief of the IEEEWireless Communications; and a Technical Editor of severalscholarly journals including IEEE Communications, WileyInternational Journal of Communication Systems, Journalof Communication Network, etc. His areas of research arewireless data communication networks, wireless IP networks,next generation mobile networks, traffic control, networksecurity and management, and satellite systems. He wasone of the first researchers to disseminate the fundamentalconcepts of the next generation mobile networks andbroadband convergence networks as well as the integrationof wireless LAN and cellular networks; some of which arebeing gradually deployed by industry and included in theITU-T standards. Dr Jamalipour has authored several invitedpapers and been a keynote speaker in many prestigiousconferences. He served as the Chair of the Satellite and SpaceCommunications Technical Committee (2004–2006); andis currently the Vice Chair of Communications Switching andRouting TC; and Chair of Chapters Coordinating Committee,Asia-Pacific Board, all from the IEEE CommunicationsSociety. He is a voting member of the IEEE GITCand IEEE WCNC Steering Committee. He has been aVice Chair of IEEE WCNC 2003 to 2006, ProgramChair of SPECTS 2004, Chair of symposiums at IEEEGLOBECOM 2005 to 2007 and IEEE ICC 2005 to2008, among many conference leadership roles. He hasreceived several prestigious awards, such as 2006 IEEEDistinguished Contribution to Satellite CommunicationsAward, 2006 IEEE Communications Society Best TutorialPaper Award, and 2005 Telstra Award for Excellence inTeaching.

Copyright © 2008 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2009; 9:103–115

DOI: 10.1002/wcm