IEEE SENSORS JOURNAL, VOL. 15, NO. 7, JULY 2015 3853 A ...A Virtual Coordinate-Based Bypassing Void Routing for Wireless Sensor Networks Dejing Zhang and Enqing Dong Abstract—To

IEEE SENSORS JOURNAL, VOL. 15, NO. 7, JULY 2015 3853

A Virtual Coordinate-Based Bypassing VoidRouting for Wireless Sensor Networks

Dejing Zhang and Enqing Dong

Abstract— To solve the routing void problem in geographicrouting, high control overhead and transmission delay are usuallytaken in wireless sensor networks. Inspired by the structurecomposed of edge nodes around which there is no routing void,an efficient bypassing void routing protocol based on virtualcoordinates is proposed in this paper. The basic idea of theprotocol is to transform a random structure composed of voidedges into a regular one by mapping edge nodes coordinates to avirtual circle. By utilizing the virtual circle, the greedy forwardingcan be prevented from failing, so that there is no routing voidin forwarding process from source to destination and controloverhead can be reduced. Furthermore, the virtual circle isbeneficial to reduce average length of routing paths and decreasetransmission delay. Simulations show the proposed protocol hashigher delivery ratio, shorter path length, less control packetoverhead, and energy consumption.

Index Terms— Wireless sensor networks, geographic routingprotocol, routing void, virtual coordinate.

I. INTRODUCTION

OVER the past decades, wireless sensor networks (WSNs)have been widely applied in many different

fields [1]–[3] in which routing protocol is one of thekey technologies. Since a sensor node exploits a pathdepending only on the location information of neighbornodes in geographic routing [4], routing protocol based ongeographic information is more efficient. Due to its highexpansibility and low influence by network size, geographicrouting has wide application prospects in large scaleWSNs [5], [6]. For example, plenty of nodes equipped withgeophones are distributed uniformly on the ground and havethe ability to get their own locations by global positioningsystem (GPS) or localization algorithms [7]–[10] in seismicexploration [11], [12], where geographic routing has potentialto serve as routing protocol. However, if a routing void,called local minimum [13], is encountered resulting from therandom distribution of sensor nodes, the greedy algorithm ingeographic routing will fail, and ultimately data transmissionalso fails in such situation.

Manuscript received November 3, 2014; revised December 21, 2014and January 13, 2015; accepted January 15, 2015. Date of publicationFebruary 3, 2015; date of current version May 13, 2015. This work wassupported in part by the National Natural Science Foundation of China underGrant 81371635, in part by the Research Fund for the Doctoral Program ofHigher Education of China under Grant 20120131110062, and in part by theScience and Technology Development Projects of Shandong province underGrant 2013GGX10104. The associate editor coordinating the review of thispaper and approving it for publication was Prof. Kiseon Kim.

The authors are with the Department of Mechanical, Electrical andInformation Engineering, Shandong University, Weihai 264209, China(e-mail: [email protected]; [email protected]).

Color versions of one or more of the figures in this paper are availableonline at http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JSEN.2015.2398852

To reduce the impact of the routing void, a strategy to isolatecertain region around a routing void is proposed in [14]. Nodeslocated in this region are banned from being selected as arelay node in order to prevent data packets from accessing tothe routing void. Ring-constraint forwarding (RCF) proposedin [15] establishes a multi-ring region around a routing void,in which relay nodes are selected to avoid routing void andbalance energy consumption. In [16], relay nodes are selectedaccording to the geographic location relationship between thedestination node and the routing void in order to preventfailing of greedy algorithm. These algorithms above have lowcomplexity, but high overhead of control packet and time delayresult in high energy consumption and inefficient transmission.Beyond that, routing void problem still exists around thoseestablished regions, and that no further scheme is proposed tosolve this problem. Greedy perimeter stateless routing (GPSR)composed of greedy forwarding and face mode is proposedin [17]. When routing void is encountered, GPSR worksunder face mode instead of greedy forwarding until findinga neighbor node closer to the destination. Boundary staterouting (BSR) proposed in [18] adopts the same strategy asGPSR to detour the void. In [19], network is divided intosome hexagon sub-nets, each of them is considered as a virtualnode. When void is encountered, face forwarding mode beginsto work among the virtual nodes. However, paths establishedby face forwarding are not optimized, a longer path may bechosen even if there exists a short one.

Recently, to solve void problem by using virtual locationinformation, some novel routing protocols have beenproposed [20]–[23]. The main strategy of these routing pro-tocols is to build sensor node’s virtual coordinate accordingto certain referenced nodes [20]–[22] or neighbor nodes [23].When the destination node is changed, virtual coordinates ofcorresponding nodes on the routing path have to be rebuilt,so current routing protocols based on virtual coordinates aremore suitable to the scenarios with fixed destination nodes.Furthermore, routing voids still exist in the network.

Routing protocols based on virtual coordinate have variousforms, which make them flexible to implement accordingto practical network conditions without constraint from thephysical locations. Though greedy algorithm is simple inprinciple and low in complexity, it cannot be applied to allsensor nodes when some routings based on virtual coordinateare adopted in the network. To solve previous problems, anefficient bypassing void routing protocol based on virtualcoordinate mapping (BVR-VCM) is proposed in this paper.The basic idea of BVR-VCM is to build virtual coordinatesof the whole void edge nodes by mapping their geographic

1530-437X © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

3854 IEEE SENSORS JOURNAL, VOL. 15, NO. 7, JULY 2015

Fig. 1. Routing void in greedy forwarding.

coordinates to a virtual circle that covers the void, and thenestablish a path by using these virtual coordinates. The virtualcircle composed of edge nodes can solve routing void prob-lem and make greedy algorithm work in entire forwardingprocess, in this way overhead of control packets are reduced.Furthermore, the establishing and maintaining for virtual coor-dinates are not affected by alternation of destination node, thusenergy consumption can be reduced. Due to the establishingprocess of virtual coordinates, the proposed routing protocol ismore suitable for stationary sensor networks, such as seismicexploration, in which nodes are stationary during their workingperiods.

II. PROBLEM DEFINITIONS

We consider the following situation: sensor nodes aremodeled by a unit graph. All nodes within communicationrange Rc of a node n are considered as neighbors of n andbidirectional links exist between n and its neighbors.

A. Routing Void in Geographic Routing

In geographic routing, when greedy forwarding is adopted,it can be easily interrupted due to the terrain or radio coverage,for example, pools, hills or buildings which locate in thesensor area. The finite distance of communication range canalso cause greedy forwarding failing. When a sensor nodetries to forward the packet to one neighbor node that isgeographically closer to the destination node than itself, butsuch node doesn’t exist, then a routing void is encountered.Greedy forwarding fails in this situation. As shown in Fig. 1,a node n1 tries to forward a packet to the destination node d1by greedy forwarding in multi hops. First, node n1 sendsthe packet to n2 by greedy forwarding. Since the neighbornodes set of n2 is {n1, n3, n4}, none of which is closer to thenode destination d1, and then a routing void is encounteredand greedy forwarding fails to deliver the packet. Similarly,a routing void is encountered at node n5 when it tries toforward a packet to the destination node d2.

Around the obstacle area in Fig. 1, greedy forwarding failsat node n5 as described above. But for different destinations,greedy forwarding may not fail at the same node. For example,if n5 tries to forward a packet to the destination node d1, packetcan arrive at d1 along with the path n5→n6→n7→d1 withoutrouting problem.

Fig. 2. Edge structure without routing void.

B. Structure Without Routing Void

Assuming the number of edge nodes around an obstaclein WSNs is Nb , the set of edge nodes is {bk | k = 1, · · · Nb},both of the following conditions should be satisfied:{

d(bk, bk+1) < Tc, k = 1, · · · Nb − 1

d(b1, bNb ) < Tc(1)

and⎧⎪⎨⎪⎩

{bk+i | d(bk, bk+1) < Tc, k = 2, . . . Nb − 2,

2 ≤ i ≤ Nb − k} = ∅

{bk | d(b1, bk) < Tc, k = 3, . . . Nb − 1 = ∅

(2)

where d(x, y) represents the Euclidean distance betweennode x and y, Tc represents the communication distance ofnodes, i is an integer. According to formula (1), (2), every edgenode can only communicate with its two neighbors belongingto the set {bk|k = 1, · · · Nb}.

If all the edge nodes around the obstacle have the samedistance to a point O as following:

d(bk, O) = R, k = 1, . . . Nb (3)

where R is a constant. In this situation, all the edge nodeslocate on a circle with center point O and radius R as shownin Fig. 2. In this type of structure composed of the edge nodes,every edge node has two and only two neighbors locating onthe circle. According to the geometrical features of circle, thereis no routing void around this obstacle area for any destinationnode in the network.

In Fig. 2, all nodes in the network adopt greedy algorithm toselect relay nodes; node s, d represents source and destinationrespectively, packet delivering process is used as an example todescribe the structure without routing void. First, edge node b1receives a packet from source s, it has two relay candidatesin neighbors, b2 and b5. Since the two candidates locate onthe same circle, there is at least one node that can be selectedas relay node by greedy algorithm in this condition, so b2 isselected and no routing void is encountered. Similarly, thepacket reaches edge node b4, and then b4 selects n5 by greedyalgorithm. Finally, the packet reaches destination node d allby greedy algorithm without routing void problem in thedelivering process.

ZHANG AND DONG: VIRTUAL COORDINATE-BASED BYPASSING VOID ROUTING FOR WSNs 3855

III. BVR-VCM ROUTING PROTOCOL

The proposed routing protocol BVR-VCM consists ofgreedy mode and void processing mode. In BVR-VCM,greedy algorithm is adopted to select relay node in greedymode. If greedy mode fails when a routing void is encountered,void processing mode is activated. Void processing mode iscomposed of three phases, according to processing in the order,respectively void detecting, virtual coordinate mapping andvoid region dividing. After the implement of void processingmode, the virtual coordinates of edge nodes are established.Then greedy mode is reactivated, these edge nodes that havethe virtual coordinates can be selected as the relay node bygreedy algorithm. In the following section, three main phasesin void processing mode and the main steps of entire processin BVR-VCM are described.

A. Void Detecting Phase

The main function of the void detecting phase is to collectedge node information around the routing void after the void isencountered. When routing void emerges in the transmissionprocess, the node at which the greedy mode fails is definedas the discovery node. After the discovery node discovers avoid, it stores data packets temporarily at first, then generatesa void detecting packet for starting a void detecting process.During the process, the void detecting packet records the timewhen the void is encountered, edge node’s label and geograph-ical coordinate. The detecting process can be performed byleft-hand (right-hand) rule. Eventually the detecting packagereturns to the discovery node. The information of edge nodescan be represented as set {bk|k = j, j + 1, · · · imax}.

In the process of void detecting, there may be multiple dis-covery nodes in the same void region, so there may be multipledetecting packets around current void at the same time. In thiscondition, in order to avoid the repetition that different detect-ing packets detect and forward around the same void, the edgenodes record the time when void is encountered after receivinga detecting packet. Based on the sequence of discovery time,a node discards the detecting packet if the time recorded inthe current packets is later than their records, otherwise thenode forwards the detecting packet. Finally, in the current voidregion, only the detecting packet send by the earliest discoverynode can complete the entire void detecting process.

B. Virtual Coordinate Mapping Phase

The virtual coordinate mapping phase is responsible formapping the edge node coordinates stored in the detectingpacked to a virtual circle, i.e., converting a structure com-posed of edge nodes to the structure without routing void asdescribed in section II.B.

The detecting packet that returns to the discovery nodestores all the information of the current void, including node’slabel and geographic coordinates. (x j , y j ), (x j+1, y j+1), · · ·(ximax , yimax ) represent the nodes coordinates, thus the centercoordinate of the void is

(xo, yo) = (1

(imax − j + 1)

imax∑k= j

xk,1

(imax − j + 1)

imax∑k= j

yk)

(4)

Algorithm 1 The Virtual Coordinate Mapping Algorithm1: initialize i = j, m = ∅

2: while i ≤ imax do3: if only one interaction x ′, y ′ between ri and U then4: (x ′

i , y ′i ) = (x ′, y ′)

5: if M = ∅ then6: xtemp, ytemp = (x ′, y ′)7: else8: while c! = 0 do9: x ′

i−c = x ′ + xc

10: y ′i−c = y ′ + yc

11: Remove bi−c from M12: end while13: end if14: else15: Add bi to M16: end if17: i + +18: end while

The maximum distance between edge nodes and the voidcenter is

do = max{dk|dk =√

(xk − xo)2 + (yk − yo)2,

k = j, j + 1, · · · imax} (5)

We define a circle with center point O and radius R as thevirtual mapping circle of the void. The process of virtualcoordinate mapping is implemented after the virtual mappingcircle is determined. The virtual coordinate mapping algorithmis shown in the algorithm 1.

In Algorithm 1, U represents a set of line segments thatlink two neighbor nodes of the void, rk represents a ray thatstarts from the center O of virtual mapping circle through theedge node bk , M represents a set that stores multiple nodesinformation, c represents the cardinal number of the set M ,(xc, yc) represents the coordinate offset of the cth node,the virtual coordinates to be determined are (x ′

j , y ′j ),

(x ′j+1, y ′

j+1), · · · (x ′imax

, y ′imax

) , (x ′, y ′) is the intersection ofnode bk and virtual mapping circle.

Fig. 3 shows a sample of the virtual coordinate mappingprocess. O is the center of void, and the arc is a part ofthe virtual mapping circle. The virtual coordinates of edgenode b1, b5 are b′

1 and b′5 respectively, and the virtual

coordinates of edge node b2, b3, b4 are b′2, b′

3, b′4 respectively,

locating between the virtual node b′1 and b′

5.After virtual coordinate mapping, the discovery node

generates a distributing packet which contains the virtual coor-dinates and center of virtual mapping circle. The distributingpacket is forwarded along the way (the routing void edge) thatthe detecting packet establishes. Every edge node broadcastsa message containing both virtual coordinate and center ofvirtual mapping circle to its neighbor nodes after receivingthe distributing packet. In this way, both the edge nodes andthe neighbor nodes learn the void information.


Fig. 3. Sample of virtual coordinate mapping.

Fig. 4. Void region dividing.

C. Void Region Dividing Phase

The main function of the void region dividing phase is todivide the surrounding area of the void into three differentregions, in which different routing strategies are applied.According to the void position and the location of destinationnode of the packets, the surrounding area of a void is dividedinto approaching region, departing region and free region, asshown in Fig. 4. O is the center of mapping virtual circle, d isthe destination node, the circle shown in dotted line is thevirtual mapping circle. Two tangents to the mapping virtualcircle through the destination node have the intersections atpoints m and n respectively. The quadrilateral region formedby O, m, d , and n is defined as departing region of the virtualmapping circle, as the region B shown in Fig. 4. The areaformed by two tangents of the virtual mapping circle, exceptthe departing region, is defined as approaching region of thevirtual mapping circle as the region A shown in Fig. 4. Thearea outside of the two tangents is defined as free region ofthe virtual mapping circle as region C shown in Fig. 4.

It should be noted that three regions are divided accordingto the current routing void; and, the division of three regionsis different for different destination nodes.

D. Routing Based on Virtual Coordinate

After implementing the three phases above, the edgenodes of void contain two types of location informationwhich are geographic coordinates and virtual coordinates,and that surrounding area of routing void has been dividedinto three different regions according to the destination node.Discovery node restarts greedy mode by utilizing the virtual

coordinates, the packet stored in the first phase is forwardedto relay nodes. In the greedy mode, the ways to select relaynode in three regions are different, but they all use the greedyalgorithm. The main steps of BVR-VCM are as follows:

Step 1: Node receives data packet;Step 2: Determine if virtual coordinate is used by itself or

its neighbors, if any, go to step 3, or go to step 4;Step 3: If node locates in the approaching region, it uses

virtual coordinate to select the relay node; if locating indeparting region, it uses geographic coordinate to select therelay node in priority; if locating in free region, it usesgeographic coordinate to select the relay node;

Step 4: If there is no routing void encountered in greedyalgorithm during the process of selecting relay node, go tostep 6, or go to step 5;

Step 5: Void process mode is activated, and virtual coordi-nates of edge nodes around routing void are established. Thengo to step 3;

Step 6: Node sends packet to the relay node.All the data packets arriving at destination mainly

experience three different routing processes in BVR-VCM.The first is arriving at destination without void problem:step 1→step 2→step 4→step 6. In this process, no void isencountered, the selection of relay nodes is depending ongreedy forwarding only. If a void is encountered for the firsttime during the packet delivery, the process in BVR-VCMis: step 1→step 2→step 4→step 5→step 3→step 6.In this process, virtual coordinates of edge nodesaround current void are established. The third processis: step 1→step 2→step 3→step 6. In this process, packetsreach the surrounding area of void, relay nodes are selected byutilizing virtual coordinates of edge nodes. Since the virtualmapping circle of void constitutes the structure withoutrouting void, packets can bypass the void, and that thelocation of source node or destination node has no influenceon the greedy algorithm.

IV. PROTOCOL ANALYSIS

A. Routing Path Analysis

According to the description of section III.B, after thevirtual coordinate mapping phase, the neighbors of the edgenodes around the routing void receive the broadcastingmessages. These neighbor nodes have the opportunity toestablish a short path by utilizing the routing void informationbefore void is encountered. As shown in Fig. 5, the node bs

tries to send a packet to a destination node. After bs sends thepacket to bv by greedy algorithm, a routing void is encounteredat node bv .

If face forwarding is adopted to solve routing void problem,it is always activated at the edge node bv . Assuming thedistribution of nodes is uniform, the nodes density in thenetwork is ρ. The probability that at least one edge node existsin the communication coverage of both bs and bv is

P = 1 − e−ρσ (6)

where σ is the intersection area within communicationcoverage of both bs and bv (the area of obstacle excluded).After establishment of virtual coordinates around routing void


Fig. 5. Routing discovery in BVR-VCM.

Fig. 6. Path exploring around void.

in BVR-VCM, the node bd is selected as relay node as shownin Fig. 5, thus the number of hops decreases. As the nodedensity ρ increases, shorter paths can be established in mostof cases.

The BVR-VCM can utilize the information of the edgenodes which are obtained during the void detecting phase toexplore a shorter bypassing path. As shown in Fig. 6, thedotted circle with center O and radius R is the virtual circle,b′

1, b′2 and b′

3 represent virtual nodes of three correspondingedge nodes, and node d represents the destination. Assumingrouting void is encountered at node b′

1, according toBVR-VCM, b′

1 will choose relay node in itsneighbor, b′

2 and b′3. Because the three virtual nodes

are located at the same virtual circle, b′1 can obtain the result

of � d Ob′3 < � d Ob′

1 < � d Ob′2 by the virtual coordinates

of virtual circle center O and neighbor nodes it learns fromdistributing packet. In this situation, b′

1 chooses b′3 as relay

node instead of b′2. However, in face forwarding, b′

1 chooseb′

2 as relay node if left-hand rule is adopted.Fig. 7 shows an example of the routing exploring process in

seismic exploration [11]. A wireless seismic node tries to senddata to the date center, but there are two obstacles betweenthe node and the date center. P1 is the optimal path whenthere is no routing void in the network, P2 represents the pathselected by algorithm based face forwarding, and P3 representsthe path explored by BVR-VCM. Because algorithms basedface forwarding can only explore path along the same directionwhen routing void is encountered, as P2 shown in Fig. 7, thefinal path is much longer than P1 and P3.

B. Control Overhead Analysis

Assuming the length of control packet header in the greedyalgorithm is Lg , the lengths of detecting packet and distrib-uting packet in BVR-VCM are Lde and Ldi respectively, theedge node number of routing void is Ns , the average number of

hops from source to destination is hBVR-VCM . When n packetsare sent, the length of the control packet is

(Lde + Ldi ) × Ns + Lg × hBVR-VCM × n (7)

The average overhead of the control packet for eachpacket is

LBVR-VCM = (Lde + Ldi ) × Ns/n + Lg × hBVR-VCM (8)

As shown in (8), when the number of packets increases, theaverage overhead in BVR-VCM is approximately equal thatin greedy algorithm:

LBVR-VCM ≈ Lg × hBVR-VCM (9)

Comparing with protocol based on face forwarding, assum-ing the length of control packet header in the greedy algorithmis L f , the proportion of the greedy algorithm is g, and theaverage number of hops from source to destination is hg .When n packets are sent, the length of control packet is

(Lg × g + L f (1 − g)) × hg × n (10)

The average overhead of the control packet for eachpacket is

LFACE = (Lg × g + L f × (1 − g)) × hg (11)

Since 0 < g < 1, L f > Lg , and hBVR-VCM ≤ hg learntfrom section IV.A, then

Lg × hBVR-VCM ≤ Lg × hg < LFACE < L f × hg (12)

For the same fixed wireless sensor network, g is constant, asshown in (11). In this condition, no matter how many packetsare send, LFACE is constant, the value range of LFACE is shownin (12). While LBVR-VCM decreases as the increasing of thepackets sent. So there exists a value named nc, if n > nc,there is LBVR-VCM < LFACE , at this point the control packetin BVR-VCM is always smaller than that in the faceforwarding protocol.

V. SIMULATION RESULTS

A. Routing Path Analysis

Simulations are performed in NS2 to demonstrate andevaluate the performance of BVR-VCM. In the simulationsthree scenarios are provided, the first scenario is to evaluatethe affection of different void sizes, the second is to evaluatethe affection of multiple routing voids, and the third is toevaluate the performance in random distribution network.

In order to evaluate the affection of different void sizesin the first scenario, 900 nodes are deployed uniformly ina square region, a cross-shaped void is assumed to locate atthe center area of the square region where there is no nodedeployed. In order to compare, a destination node is randomlyselected below the center area, while source node is randomlyselected above. In every simulation, one node is assumedto be destination node and four nodes are source nodes, theradius of routing void is changed from 90m to 270m. Inthe second scenario, different numbers of the cross-shapedvoids are deployed randomly in the simulation area, onedestination node and four source nodes are randomly selected


Fig. 7. Routing discovery in seismic exploration.

TABLE I

SIMULATION PARAMENTERS

Fig. 8. Void size versus delivery ratio.

in the network. In the third scenario, nodes are deployedrandomly in a square region, one destination node andfour source nodes are randomly selected in the network aswell. AODV [24], GPSR and Greedy-Vip(n) (n = 1, 2)are implemented and compared in terms of delivery ratio,end-to-end delay, average number of hops, control overheads,and energy consumption. Table I summarizes the simulationenvironmental parameters.

B. Impact of Routing Void Size

Fig. 8 shows the delivery ratio at different routingvoid sizes. As a whole, delivery ratios in Greedy-ViP(n),

Fig. 9. Void size versus average delay.

BVR-VCM and GPSR decrease as routing void size increases,but ratio in AODV keeps around 95%. Because AODV isbased on flooding, it has high probability to establish apath between source and destination. Therefore delivery ratioin AODV is rarely affected by void size. And ratios inGreedy-ViP(n) decrease quickly, that means greedy algorithmfails frequently as void size increases. When void size is biggerthan 210m, delivery ratio in BVR-VCM becomes lower thanin AODV, but it’s always higher than these in other protocols.

The average delay is defined as the average transmissiondelay of successful delivery packets. As shown in Fig. 9,the average delay in AODV is nearly two times more thanthat in BVR-VCM, Greedy-ViP(n) or GPSR. The reason isthat it takes time to establish the routing path in AODVbefore packet is delivered. The average delay in BVR-VCMis slightly lower than that in GPSR but higher than thesein Greedy-ViP(n). As the routing void increases, the averagedelays in both BVR-VCM, Greedy-ViP(n) and GPSR increase,while delay in GPSR grows rapidly than that in BVR-VCMand Greedy-ViP(n).

Fig. 10 shows the impact of routing void size on the averagenumber of hops. There are 7 fewer hops in BVR-VCM andGreedy-ViP(n) than that in AODV, and average hops in GPSR


Fig. 10. Void size versus average hops.

TABLE II

ENERGY CONSUMPTION VERSUS VOID SIZE(0∼100s)

increases rapidly as the void size grows. Due to the featurethat a shorter path can be found in BVR-VCM according tothe virtual circle of routing void and position of destinationnode, as shown in Fig. 9 and Fig. 10 respectively, the averagehopss and transmission delay in BVR-VCM increase slightlyas void size grows. Although Greedy-ViP(n) outperforms otherprotocols in both average delay and hop, plenty of packets aredropped during the delivery process as shown in Fig. 8.

TABLE II and TABLE III depict the average energyconsumption per packet respectively during time rangesfrom 0s to 100s and 0s to 200s. Due to involving lots ofnodes for the establishing and maintaining of routing paths,the average energy consumption in AODV is much more thanthese in BVR-VCM, Greedy-ViP(n) or GPSR, and there isgreat distinction between time ranges from 0s to 100s and0s to 200s in AODV. While average energy consumptionin BVR-VCM, Greedy-ViP(n) and GPSR are lower thanthat in AODV, this is because establishing of routing pathjust depends on the neighbor nodes’ location information.After the virtual coordinate mapping in BVR-VCM, greedyforwarding is adopted in the entire routing process. Therefore,average energy consumption in BVR-VCM is lower than thatin GPSR or AODV during time ranges from 0s to 200s.But in the first 100s, energy consumption in BVR-VCM ismore than that in Greedy-ViP(n). Besides, the two tablesshow that the average energy consumption in GPSR rarelychanges in two time ranges, which is because path establishingin GPSR is independent from the beginning to the end insimulation. In Greedy-ViP(n), establishing of virtual positonscosts certain energy and all the nodes in network are involved.

TABLE III

ENERGY CONSUMPTION VERSUS VOID SIZE(0∼200s)

Fig. 11. Void size versus control overhead.

In BVR-VCM, only the edge nodes of the voids involved, andextra control packets are generated when void processing modeis activated. However, greedy algorithm mode rarely fails aftervoid processing mode, which reduces the average consumptionas the simulation goes on. That means the longer time networkruns, the lower average energy consumption it takes.

Less routing control packets overhead can improve theenergy efficiency in wireless sensor networks and extend thelife time of the network. Fig. 11 shows the simulation resultof control packet overheads at different void radiuses.

BVR-VCM can learn void information by activating voidprocessing mode only once, which can make packets from anysource nodes bypass routing void. Greedy algorithm is the onlymethod in process of finding a path, therefore, not only is thecontrol packet overhead small, but the influence from routingvoid sizes on total packet overhead increases slightly. Althoughrouting void size has little affection on control packet overheadin AODV, its overhead is the biggest because of the flooding-based mechanism. Every time the routing void is encountered,face forwarding that costs more control packets is activatedin GPSR. The number of nodes involving into face forwardingincreases as the size of routing void grows, which leads torapid growth of control packet overhead. In Greedy-ViP(n),as the routing void size increases, control packet overheaddecreases. According to the simulation result, the bigger thevoid size is, the less delivery ratio will be in Greedy-ViP(n),that means less packets are delivered, so the energy controlpacket overhead becomes less and less as void size increases.


Fig. 12. Packet size versus delivery ratio.

Fig. 13. Packet size versus average delay.

C. Impact of Packet Size

When the length of data packet transmitted in the networkgrows, the time spent on sending and receiving increases.Because of reason above, the collision probability of theradio transmission increases, which leads to the decreasingof delivery ratio and the increasing of transmission delay.Fig. 12 and Fig. 13 show the impact of different size packetson the delivery ratio and the transmission delay when voidradius maintains 150m.

As shown in Fig. 12, delivery ratio in BVR-VCMdecreases as the packet size increases from 128 bytes to 512,falling by 8 percent. In the same condition, delivery ratioin Greedy-ViP(n), GPSR and AODV reduce by about 12%,15% and 30% respectively. Because more control packetsare used in AODV than others, the high probability ofcollision causes more packets loss when packets size increases.In Greedy-ViP(n), packets are forwarded by greedy algorithm,congestion may emerge at some nodes which closes to thedestination nodes. In GPSR, the selecting of relay node is notoptimized, so the packets congest on the same side of the void,which aggravates transmission collision. However, establishingof routing path is optimized in BVR-VCM, so the congestionis dispersed and alleviated around the routing void.

Fig. 13 shows that the average transmission delay in AODVis sensitive to packet size, increasing nearly 10 times whenpack sizes grow from 128 byte to 512. While in other protocolsbased geographic information or virtual position, the affectionon average delay suffering from packet size is smaller.

Fig. 14. Multi-void versus delivery ratio.

Fig. 15. Multi-void versus average hops.

D. Impact of Multiple Routing Voids

In the second scenario, multiple routing voids are deployedrandomly in the simulation region instead of one single void.The radius of every void is 45m. Other parameters are equalto those in the first scenario.

Fig. 14 shows the impact of multiple routing voids ondelivery ratio. Because of flood-based, AODV keeps steadydelivery ratios when the number of voids changes. As morerouting voids emerge in the network, greedy algorithm failsmore frequently. Therefore, as shown in Fig. 14, the deliv-ery ratio in Greedy-ViP(n) decreases rapidly as the numberof voids increases. However, the delivery ratios in bothBVR-VCM and GPSR decrease slightly due to differentrouting strategies. The BVR-VCM gains better advantageover GPSR.

Fig. 15 shows the impact of multiple routing voids on theaverage number of hops. Average hops in AODV are biggerthan other protocols but more stable as the number of routingvoids increases. Because there is no optimal strategy in GPSR,longer paths might be established especially when multiplerouting voids are deployed in the network. In BVR-VCM, theaverage hops have slower rate of descent, because the pathshave been optimized as described in Sec. IV. A. As a whole,Greedy-ViP(n) outperforms other protocols in average hops,but plenty of packets are dropped when the number of routingvoids increases as shown in Fig. 14.

TABLE IV and TABLE V depict the average energyconsumption per packet respectively during time rangesfrom 0s to 100s and 0s to 200s when different numbers


TABLE IV

ENERGY CONSUMPTION VERSUS MULTI-VOID(0∼100s)

TABLE V

ENERGY CONSUMPTION VERSUS MULTI-VOID(0∼200s)

Fig. 16. Impact of node density on delivery ratio.

of routing voids are deployed in the simulation region.As shown in the two tables, the simulation results havethe similar tendencies as those in the first scenario.Geography-based protocols have less energy consumptionthan that in AODV, and energy consumptions inGreedy-ViP(n) and GPSR rarely change during the simulationperiods. During period from 0s to 100s, consumption inBVR-VCM has no advantage than Greedy-ViP(n) or GPSR.However, as simulations goes on, energy consumption inBVR-VCM decreases significantly.

E. Impact of Node Density

In the third scenario, nodes are randomly deployed in asquare region, one destination node and four source nodes arerandomly selected in the network. Different numbers of nodesare simulated in the simulation area. In this scenario, routingvoids are encountered randomly, the less node density is, themore probability of encountering routing void will be.

Fig. 16 shows the impact of node density on delivery ratio.AODV keeps steady delivery ratio when the node

Fig. 17. Impact of node density on average hops.

density changes, because it has high probability to establisha path between source and destination due to the floodingstrategy. When the number of nodes is small, nodes have fewneighbors to be their relay candidates. Protocols based geo-graphic information are affected seriously by the node density,as shown in Fig. 16, protocols except AODV have low deliveryratio at low node density. The delivery ratio in Greedy-ViP(n)shows the greedy forwarding fails frequently. In BVR-VCMand GPSR, they all try their best to find paths along the routingvoid, therefore, the delivery ratio is improved. When the num-ber of nodes is large, nodes have more neighbors to be chosenas relay candidates. Therefore, geographic protocols have moreopportunity to prevent greedy algorithm from failing. TheBVR-VCM gains better advantage over others. The reason is ithas the ability to choose relay nodes along two different sidesof routing void, which can reduces the number of droppedpackets caused by the collision of sending and receiving.

Fig. 17 shows the impact of node density on the averagenumber of hops. In randomly deployed network, a path canbe established in most of the case if there exists a connectionin multi-hop when link-based protocols are adopted, but thepath may not be optimized in hop count. Therefore, thereare eight more hops in AODV than others and hop countchanges slightly when node density increases. By utilizing thevirtual coordinates of the edge nodes and their neighbors inBVR-VCM, data packets from different sources will detourthe voids along shorter paths. As the node density increases,this advantage will be weakened. The hop count is slightlylower than that in GPSR. Although Greedy-ViP(n) have thebest performance in the simulation, the cost of dropped packetsis heavy as we analyze above.

VI. CONCLUSION

To solve routing void problem in geographic routing,BVR-VCM is proposed by utilizing the edge structure withoutrouting void. BVR-VCM uses void detecting, virtualcoordinate mapping and void region dividing to solve voidproblem, and then establishes the path around void accordingto the virtual coordinates of edge nodes. Because voidprocessing mode is performed only once for a routingvoid, the complexity of routing protocol can be reduced.Simulations show that the proposed BVR-VCM routingprotocol has advantages in terms of average delivery ratio,


transmission delay, et al. Besides, lower control overhead inBVR-VCM also reduces the energy consumption.

Due to the hardware resource, the application range ofthe proposed protocol may be confined to special fields inwhich sensor nodes are equipped with enough redundantresources, such as seismic exploration. Future work will be tomake proposed protocol generalized to common applications.To eliminate the possibility that the discovery packet couldoverload when detecting large voids, the alternative methodof void detecting will be taken into consideration.

REFERENCES

[1] M. Chen, J. Wan, S. Gonzalez, X. Liao, and V. C. M. Leung, “A surveyof recent developments in home M2M networks,” IEEE Commun.Surveys Tuts., vol. 16, no. 1, pp. 98–114, Feb. 2014.

[2] M. Li, Z. Li, and A. V. Vasilakos, “A survey on topology controlin wireless sensor networks: Taxonomy, comparative study, and openissues,” Proc. IEEE, vol. 101, no. 12, pp. 2538–2557, Dec. 2013.

[3] S. Zhang, D. Li, and J. Chen, “A link-state based adaptive feedbackrouting for underwater acoustic sensor networks,” IEEE Sensors J.,vol. 13, no. 11, pp. 4402–4412, Nov. 2013.

[4] F. Cadger, K. Curran, J. Santos, and S. Moffett, “A survey of geographi-cal routing in wireless ad-hoc networks,” IEEE Commun. Surveys Tuts.,vol. 15, no. 2, pp. 621–653, May 2013.

[5] B. Tang and L. Zhang, “Optimization of energy multi-path routingprotocol in wireless sensor networks,” J. Syst. Eng. Electron., vol. 35,no. 12, pp. 2607–2612, Dec. 2013.

[6] X. Wang, J. Wang, K. Lu, and Y. Xu, “GKAR: A novel geographicK-anycast routing for wireless sensor networks,” IEEE Trans. ParallelDistrib. Syst., vol. 24, no. 5, pp. 916–925, May 2013.

[7] S. Lee, E. Kim, C. Kim, and K. Kim, “Localization with a mobile beaconbased on geometric constraints in wireless sensor networks,” IEEETrans. Wireless Commun., vol. 8, no. 12, pp. 5801–5805, Dec. 2009.

[8] W. Liu, E. Dong, Y. Song, and D. Zhang, “An improved flip ambiguitydetection algorithm in wireless sensor networks node localization,”in Proc. 21st Int. Conf. Telecommun., Lisbon, Portugal, May 2014,pp. 206–212.

[9] J. Wang, E. Dong, F. Qiao, and Z. Zou, “Wireless sensor networks nodelocalization via leader intelligent selection optimization algorithm,” inProc. 19th Asia-Pacific Conf. Commun., Bali, Indonesia, Aug. 2013,pp. 666–671.

[10] W. Liu, E. Dong, and Y. Song, “Robustness analysis for node multilatera-tion localization in wireless sensor networks,” Wireless Netw., Nov. 2014.[Online]. Available: http://dx.doi.org/10.1007/s11276-014-0865-0

[11] INOVA, Houston, TX, USA. (Jun. 2012). FireFly RecordingSystem. [Online]. Available: http://www.inovageo.com/demo/images/stories/resources/FireFly_Brochure_100525.pdf

[12] D. Mougenot. (2012). Land acquisition systems: From central-ized architecture to autonomous sources and receivers. Sercel,Houston, TX, USA. [Online]. Available: http://www.sercel.com/products/Lists/ProductPublication/Sercel-Land-Acquisition-Systems.pdf

[13] N. Ahmed, S. S. Kanhere, and S. Jha, “The holes problem in wirelesssensor networks: A survey,” in Proc. ACM SIGMOBILE MC2R, 2005,pp. 4–18.

[14] F. Yu, S. Park, Y. Tian, M. Jin, and S. Kim, “Efficient hole detourscheme for geographic routing in wireless sensor networks,” in Proc.Veh. Technol. Conf., Singapore, May 2008, pp. 153–157.

[15] G. Trajcevski, F. Zhou, R. Tamassia, B. Avci, P. Scheuermann, andA. Khokhar, “Bypassing holes in sensor networks: Load-balance vs.latency,” in Proc. IEEE Global Telecommun. Conf., Houston, TX, USA,Dec. 2011, pp. 1–5.

[16] C. Chang, C. Chang, Y. Chen and S. Lee, “Active route-guiding protocolsfor resisting obstacles in wireless sensor networks,” IEEE Trans. Veh.Technol., vol. 59, no. 9, pp. 4425–4442, Nov. 2010.

[17] B. Karp and H. T. Kung, “GPSR: Greedy perimeter stateless routing forwireless networks,” in Proc. ACM 6th Annu. Int. Conf. MobiCom, 2000,pp. 243–254.

[18] C. J. Lemmon and P. Musumeci, “Boundary mapping and boundary-state routing (BSR) in ad hoc networks,” IEEE Trans. Mobile Comput.,vol. 7, no. 1, pp. 127–139, Jan. 2008.

[19] X. Fang, H. Gao and S. Xiong, “RPR: High-reliable low-cost geograph-ical routing protocol in wireless sensor networks,” J. Commun., vol. 33,no. 5, pp. 29–37, May 2012.

[20] P. Huang, C. Wang, and L. Xiao, “Improving end-to-end routingperformance of greedy forwarding in sensor networks,” IEEE Trans.Parallel Distrib. Syst., vol. 23, no. 3, pp. 556–563, Mar. 2012.

[21] Y. Noh, U. Lee, P. Wang, B. S. C. Choi, and M. Gerla, “VAPR: Void-aware pressure routing for underwater sensor networks,” IEEE Trans.Mobile Comput., vol. 12, no. 5, pp. 895–908, May 2013.

[22] S. Chen, G. Fan, and J.-H. Cui, “Avoid ‘void’ in geographic routingfor data aggregation in sensor networks,” Int. J. Ad Hoc UbiquitousComput., vol. 1, no. 4, pp. 169–178, 2006.

[23] J. You, Q. Han, D. Lieckfeldt, J. Salzmann, and D. Timmermann,“Virtual position based geographic routing for wireless sensor networks,”Comput. Commun., vol. 33, no. 11, pp. 1255–1265, 2010.

[24] C. E. Perkins and E. M. Royer, “Ad-hoc on-demand distance vectorrouting,” in Proc. 2nd IEEE Int. Workshop Mobile Comput. Syst. Appl.,Feb. 1999, pp. 90–100.

Dejing Zhang received the B.S. degree incommunication engineering and the M.S. degreein communication and information systemsfrom Shandong University, Weihai, China,in 2005 and 2009, respectively, where he iscurrently pursuing the Ph.D. degree. His mainresearch interests are in routing protocols andembedded system in wireless sensor networks.

Enqing Dong received the B.S. degree in geologyfrom the China University of Mining and Technol-ogy, Xuzhou, China, in 1987, the M.S. degree inapplied geophysics from Changan University, Xi’an,China, in 1993, and the Ph.D. degree in electricalengineering from Xi’an Jiaotong University, Xi’an,in 2002. From 1993 to 1998, he was an Instructorwith the Department of Electrical Engineering, Xi’anPetroleum University, Xi’an. From 2002 to 2005, hewas a Full Professor with the School of Electricsand Information Engineering, Soochow University,

Suzhou, China. From 2006 to 2007, he was a Visiting Professor with theBroadband Communications Laboratory, Harvard University, Cambridge, MA,USA. Since 2008, he has been with the School of Mechatronics and Infor-mation Engineering, Shandong University, Weihai, China, as a Full Professor.His current research interests are in signal processing with applications inwireless sensor networks and MRI image processing.

Documents

IEEE SENSORS JOURNAL, VOL. 15, NO. 7, JULY 2015 3853 A ...A Virtual Coordinate-Based Bypassing Void Routing for Wireless Sensor Networks Dejing Zhang and Enqing Dong Abstract—To