784 ieee transactions on industrial … · 784 ieee transactions on industrial informatics, vol....

784 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 1, FEBRUARY 2014

R3E: Reliable Reactive Routing Enhancement forWireless Sensor Networks

Jianwei Niu, Member, IEEE, Long Cheng, Member, IEEE, Yu Gu, Member, IEEE, Lei Shu, Member, IEEE, andSajal K. Das, Senior Member, IEEE

Abstract—Providing reliable and efficient communication underfading channels is one of the major technical challenges in wirelesssensor networks (WSNs), especially in industrial WSNs (IWSNs)with dynamic and harsh environments. In this work, we present theReliable Reactive Routing Enhancement (R3E) to increase the re-silience to link dynamics for WSNs/IWSNs. R3E is designed to en-hance existing reactive routing protocols to provide reliable and en-ergy-efficient packet delivery against the unreliable wireless linksby utilizing the local path diversity. Specifically, we introduce a bi-ased backoff scheme during the route-discovery phase to find a ro-bust guide path, which can provide more cooperative forwardingopportunities. Along this guide path, data packets are greedily pro-gressed toward the destination through nodes’ cooperationwithoututilizing the location information. Through extensive simulations,we demonstrate that compared to other protocols, R3E remark-ably improves the packet delivery ratio, while maintaining highenergy efficiency and low delivery latency.

Index Terms—Industrial wireless sensor networks (IWSNs), op-portunistic routing, reliable forwarding, unreliable wireless links.

I. INTRODUCTION

W IRELESS sensor networks (WSNs) are replacing thetraditional wired industrial communication systems

since the industrial wireless sensor networks (IWSNs) offerseveral advantages including easy and fast installation andlow-cost maintenance [1]. IWSN applications, such as factoryautomation, industrial process monitoring and control, and

Manuscript received December 03, 2012; revised February 25, 2013;accepted April 01, 2013. Date of publication May 01, 2013; date of current ver-sion December 12, 2013. This work was supported in part by Research Fund ofthe State Key Laboratory of Software Development Environment under GrantBUAA SKLSDE-2012ZX-17, National Natural Science Foundation of Chinaunder Grant 61170296 and 61190120, Program for New Century Excellent Tal-ents in University under Grant NECT-09–0028, NRF2012EWTEIRP002-045,China Postdoctoral Science Foundation under Grant 2013M530511, SUTDSRG ISTD 2010 002, Singapore-MIT IDC IDD61000102a and IDG31100106aGrant. Paper TII-12-0821.J. Niu is with the State Key Lab of Software Development Environment, Bei-

hang University, Beijing 100191, China (e-mail: [email protected]).L. Cheng is with the State Key Lab of Software Development Environment,

Beihang University, Beijing 100191, China, and also with Singapore Universityof Technology and Design, 138682 Singapore (e-mail: [email protected]).Y. Gu is with Singapore University of Technology and Design, 138682 Sin-

gapore (e-mail: [email protected]).L. Shu is with Guangdong University of Petrochemical Technology, Beijing

525000, China (e-mail: [email protected]).S. K. Das is with University of Texas at Arlington, Arlingont, TX 76019USA

(e-mail: [email protected]).Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TII.2013.2261082

plant monitoring, require reliability and timeliness in for-warding messages among nodes [2]. However, the traditionalrouting protocols, such as AODV [3], AOMDV [4], and DSR[5], may find their limitations in industrial installations due tothe harsh environmental conditions, interference issues, andother constraints [6].In IWSNs, transmission failures can result in missing or de-

laying of process or control data, andmissing the process or con-trol deadline is normally intolerable for industrial applications,as it may cause chaos in industrial automation or possibly ter-minate the automation, ultimately resulting in economic losses[7]. The sensed data should be reliably and timely transmitted tothe sink node, and the programming or retasking data for sensornode operation, command, and query should be reliably deliv-ered to the target nodes [8]. It is also required that these networkscan operate for years without replacing the device batteries [9].Therefore, the reliability, timeliness, and energy efficiency ofdata forwarding are crucial to ensure proper functioning of anIWSN. However, one of the major technical challenges for re-alization of IWSNs is to provide reliable and efficient commu-nication in dynamic and harsh environments [1], [10]. This isbecause, in harsh industrial environments, sensor nodes may besubject to radio frequency (RF) interference, highly caustic orcorrosive environments, high humidity levels, vibrations, dirtand dust, or other conditions that challenge network perfor-mance [8].Since the varying wireless channel conditions and sensor

node failures may cause network topology and connectivitychanges over time, to forward a packet reliably at each hop, itmay need multiple retransmissions. This results in undesirabledelay as well as additional energy consumption. Opportunisticrouting (OR) [11]–[16] has been proposed as an effectivecross-layering technique to combat fading channels, thusimproving the robustness and energy efficiency in wirelessnetworks. The idea of opportunistic routing is to take advantageof the broadcast nature of wireless communication, involvingmultiple neighbors of the sender into local forwarding. Sincethe wireless medium is shared, each node can overhear datapackets sent by its neighbors. In the network layer, a set offorwarding candidates are specified in the data packet and thesenodes will follow the assigned priorities to relay the packet.Essentially, only one node is chosen as the actual forwarder atthe MAC layer in an a posteriori manner.Reactive routing protocols [3], [4], [17] are designed to re-

duce the bandwidth and storage cost consumed in table drivenprotocols. These protocols apply the on-demand procedures todynamically build the route between a source and a destination.Routes are generally created and maintained by two different

1551-3203 © 2013 IEEE

NIU et al.: R3E: RELIABLE REACTIVE ROUTING ENHANCEMENT FOR WIRELESS SENSOR NETWORKS 785

phases, namely: route discovery and route maintenance. Routediscovery usually occurs on-demand by flooding an RREQ(RouteRequest) through the network, i.e., when a node hasdata to send, it broadcasts an RREQ. When a route is found,the destination returns an RREP (RouteReply), which containsthe route information (either the hop-by-hop information [3]or complete addresses from the source to the destination [5])traversed by the RREQ.In this work, we propose a Reliable Reactive Routing En-

hancement (R3E) to increase the resilience to link dynamicsfor WSNs/IWSNs. Our design inherits the advantages of op-portunistic routing, thus achieving shorter end-to-end deliverydelay, higher energy efficiency, and reliability. R3E is designedto augment existing reactive routing protocols to combat thechannel variation by utilizing the local path diversity in the linklayer. As a new addition to the cooperative forwarding designspace in WSNs/IWSNs, our major contributions are as follows.• We consider the effect of route discovery on the cooper-ative forwarding performance and combine the solutionsto reliable route discovery and efficient cooperative for-warding problems. A robust virtual path that can providemore cooperative forwarding opportunities is found with alow overhead in the route discovery phase, which not onlyimplements the forward path setup in reactive routing, butalso facilitates cooperative forwarding along the discov-ered path.

• We propose a simple yet effective cooperative forwardingscheme. Along the discovered virtual path, data packetscan be greedily forwarded toward the destination throughnodes’ cooperation without utilizing location information.

• Through comprehensive performance comparisons, wedemonstrate the effectiveness and feasibility of R3Edesign, which is compatible with most existing reactiverouting protocols in WSNs/IWSNs.

The remainder of this work is organized as follows. Section IIdescribes the network model and motivation. Section III elab-orates the design of R3E. Section IV provides the simulationresults followed by the related work in Section V. Finally,Section VI concludes the paper.

II. NETWORK MODEL AND MOTIVATION

A. Network Model

We consider a dense multihop static WSN deployed in thesensing field. Since opportunistic routing is normally effectivefor wireless networks with higher node densities (e.g., more thanten neighbors per node) [18], we assume that each node hasplenty of neighbors. When a node has packets to send to thedestination, it launches the on-demand route discovery to find aroute if there is not a recent route to a destination.We assume that the MAC layer provides the link quality esti-

mation service. There has been a lot of existing work on howto measure wireless link quality in an efficient and accuratemanner. We can use a representative link estimation method,such as those in [19] and [20]. In [21], through a set of real ex-periments, the authors reported that the size of the packets hasa direct relationship with the packet reception ratio (PRR) in

Fig. 1. Example of the guide path.

wireless networks. Short control messages, such as RREQ andRREP, have higher PRRs than data packets.Each node periodically sends HELLOmessages to keep track

of its neighborhood information. The HELLO message con-tains the IDs (addresses) of a node’s one-hop neighbors and thePRRs of the corresponding links. After the HELLO messageexchange, essentially, each node maintains the two-hop neigh-borhood information.

B. Motivation

Before providing the detailed design, we first illustratethe motivation behind R3E design. The idea of opportunisticrouting is to utilize the path diversity for cooperative caching,that is, in each hop, neighboring nodes that hold the copies ofa data packet serve as caches, thus the downstream node couldretrieve the packet from any of them [22]. The rationale is that,the path with higher spatial diversity (more potential helpernodes) may possibly provide more reliable and efficient packetdelivery against the unreliable links. With this observation, weaim to find such a reliable virtual path to guide the packets tobe progressed toward the destination. We call this virtual patha guide path, in which the nodes are named as guide nodes. Asshown in Fig. 1, is a guide path, andnodes C and G are the guide nodes. The guide path points outthe general direction toward the destination, and the routingdecision is made a posteriori, i.e., the actual forwarders arechosen based on the packet reception results at each hop.

III. MAIN DESIGN

Here, we first present the R3E functional architectureoverview, followed by a detailed description of R3E design,which is compatible with most existing reactive routing proto-cols in WSNs/IWSNs.

A. Architecture Overview

Fig. 2 illustrates an overview of the functional architecture ofR3E, which is a middle-ware design across the MAC and thenetwork layers to increase the resilience to link dynamics forWSNs/IWSNs. The R3E enhancement layer consists of threemain modules, the reliable route discovery module, the poten-tial forwarder selection and prioritization module, and the for-warding decision module. The helper node and potential for-warder are interchangeable in this work.The reliable route discovery module finds and maintains the

route information for each node. During the route discoveryphase, each node involved in the cooperative forwarding


Fig. 2. Functional architecture overview of R3E.

process stores the downstream neighborhood information, thatis to say, when a node serves as a forwarder, it already knowsthe next-hop forwarding candidates along the discoveredpath. The other two modules are responsible for the runtimeforwarding phase. When a node successfully receives a datapacket, the forwarding decision module checks whether it isone of the intended receivers. If yes, this node will cache theincoming packet and start a backoff timer to return an ACKmessage, where the timer value is related with its ranking inthe intended receiver list (called forwarding candidate list).If there is no other forwarder candidate with higher prioritytransmitting an ACK before its backoff timer expires, it willbroadcast an ACK and deliver the packet to the upper layer,i.e., trigger a receiving event in the network layer. Then, thepotential forwarder selection and prioritization module attachesthe ordered forwarder list in the data packet header for the nexthop. Finally, the outgoing packet will be submitted to the MAClayer and forwarded towards the destination.

B. Reliable Guide Path Discovery

1) RouteRequest (RREQ) Propagation: If a node has datapackets to send to a destination, it initiates a route discovery byflooding an RREQ message. When a node receives a non-dupli-cate RREQ, it stores the upstream node id and RREQ’s sequencenumber for reverse route learning. Instead of rebroadcasting theRREQ immediately in existing reactive routing protocols, weintroduce a biased backoff scheme at the current RREQ for-warding node. The aim of this operation is to intentionally am-plify the differences of RREQ’s traversing delays along dif-ferent paths. This operation enables the RREQ to travel fasteralong the preferred path according to a certain defined metric.Let and denote the last-hop node and current forwarding

node of an RREQ, respectively. Let denote the set of ’sone-hop neighbors, and denote the common neighborset between and . We define a helper between andas the common neighbor of and , satisfying

and , where is the PRR between and . Forcooperative routing, there exists an implicit constraint, that is,the nodes in the helper set should be able to hear from eachother with a reasonably high probability. Let denote theset of helpers between and . In other words, is thecommon neighbor set between and on the premise that

Fig. 3. Example illustrating the biased backoff scheme for RREQ propagationduring the route discovery phase. The RREQ that travels along the path

arrives at the first.

any two nodes in can overhear each other, and, . .

Let denote the backoff delay at the current forwardingnode , which receives an RREQ from . is calculated asdefined in

(1)

where is a time slot unit; the HopCount is the RREQ’s hopdistance from the source node thus far. The rationale is that, theneighbor with more forwarding candidates, better link qualities,as well as shorter hop-count will have a shorter backoff delay torebroadcast the RREQ.Fig. 3 illustrates the biased backoff scheme. Any node

that forwards the RREQ will calculate the backoff delay byassuming itself as a guide node, and considering the last-hopnode as its upstream guide node. For example, nodes A, B, andC receive an RREQ from the source S. When node C calculatesits backoff delay, it considers itself as a guide node and S as theupstream guide node. From the local neighbor table, C knowsthat A and B are helper nodes. Then, it can calculate the valueof backoff delay. In Fig. 3, the label beside the helpernode A means that and . At node C, thebackoff delay is about according to (1). Compared withA and B, C has a shorter backoff delay. When C’s backoff timerfirst expires, the RREQ is rebroadcasted. Consequently, nodeC has a higher priority to forward the RREQ. Similarly, nodeF forwards the RREQ before D and E. Thus, the RREQ thattravels along the path arrives at the first.From (1), we can see that the higher priority is possibly

given to the path with more potential helpers. Upon receivingan RREQ, a destination replies by sending an RREP messageback to the source along the reverse route. In case of receivingthe same RREQ multiple times, the destination shall only replyto the first received RREQ and neglect others. Algorithm 1describes how a node handles a received RREQ.

Algorithm 1 How a node handles the RREQ received fromnode

1 Procedure: void RecvRREQ (Packet *p)2 if Non-duplicate RREQ then3 if is the destination node then4 Send out RREP;5 else


6 ;7 //get common neighbor set ;8 Sort descendingly ordered by ;9 , ;10 // is always the first item of ;11 While do12 if then13 // is within the transmission range of any

node in ;14 ;15 end16 ;17 end18 Calculate and call ;19 //schedule a timer whose value is , then call

when the timer expires;20 end21 else22 Drop ;23 end

2) RouteReply (RREP) Propagation: When a node receivesan RREP, it checks if it is the selected next-hop (the upstreamguide node) of the RREP. If that is the case, the node realizesthat it is on the guide path to the source, thus it marks itself asa guide node. Then, the node records its upstream guide nodeID for this RREP and forwards it. In this way, the RREP ispropagated by each guide node until it reaches the source via thereverse route of the corresponding RREQ. Finally, this processfinds a guide path from the source to the destination.

Algorithm 2: How a node handles the RREP received fromits downstream guide node

1 Procedure:void RecvRREP (Packet *p)2 if Non-duplicate RREP then3 if then4 // is the selected next-hop & guide node ;5 Mark myself as a guide node;6 Record and ;7 Get RREP’s next-hop node id ;8 Attach , , and

to RREP;9 /* is ’s upstream guide node; the helper set

is ordered descendingly by the PRR toward thedownstream guide node;*/

10 Call forwardRREP p;11 else if then12 // is a helper in ;13 Record , , and ;14 Drop ;15 else16 Drop ;17 end18 else19 Drop ;20 end

Fig. 4. RREP packet structure. Suppose guide node sends out an RREP tothe upstream guide node , and node overhears thismessage.

In our design, the RREP message has twofold functions. Itnot only implements the forward path setup, i.e., marking guidenodes along the reverse route, but also notifies the potentialhelpers to facilitate cooperative forwarding. Specifically, twosets of helpers and their relay priority assignments are includedin the RREP. Suppose , , and are three adjacentguide nodes, the upstream link helper set and down-stream link helper set together with their PRRstoward the corresponding downstream guide nodes are piggy-backed to the RREP when node forwards it. Due to the broad-cast nature of wireless communication, all of the helper nodes in

are expected to overhear this RREP. When the guidenode receives the RREP from , it records its downstreamguide node and . When the upstream link helpersin receive the RREP, they record , ,, and , which will be useful in the data forwarding

phase. Algorithm 2 describes how a node handles the RREP re-ceived from its downstream guide node.Since R3E is an enhancement layer over existing reactive

routing protocols, a bit (the R3E flag) is used to indicate that theR3E function is enabled. There is not much protocol overheadfor the RREQ message, where only the hopcount is included.R3E incurs a certain protocol overhead in RREP, i.e., a sequenceof node IDs are piggybacked to the RREP, as shown in Fig. 4.However, the overhead will be eventually compensated by per-formance gain during the data transmission phase. Suppose theguide node sends out an RREP to the upstream guide node

, and node overhears this message. Wedefine the downstream node set of , denoted by , asthe sequential nodes that rank ahead of in the piggybackednode list of RREP. As seen in Fig. 4, is thepotential downstream helper set of .Fig. 5 is an example illustrating the RREP propagation

corresponding to Fig. 3. Suppose F is the current RREPforwarding guide node, C is F’s upstream guide node, theupstream link helper set and downstream link helperset will be attached in the RREP. When the guidenode F forwards the RREP to its upstream guide node C,for example, helper D overhears this RREP and records thepiggybacked information . D’s one-hopneighbor set is . Therefore, its potentialforwarding candidates when forwarding data packets towardthe would be .Since the wireless links are unreliable, especially in harsh

IWSNs, we also consider the possibilities of RREQ and RREPtransmission failures. R3E is fault-tolerant to the failure ofRREQ, since there exist multiple paths between the source andthe destination. However, the transmission reliability of RREP


Fig. 5. Example illustrating the RREP propagation to implement cooperativeforwarding, in which , , and are adjacent guide nodes in se-quence.

is desirable to be guaranteed, since the RREP returned by thedestination node may collide with RREQs in the network,which will be shown in Section IV-C. In addition, if an RREPis lost, the source node probably needs to launch another routediscovery process again, which will result in a long routingdiscovery delay.

C. Cooperative Forwarding

The cooperative forwarding procedure in R3E is describedas follows. The source node broadcasts a data packet, which in-cludes the list of forwarding candidates (helper nodes and thedownstream guide node) and their priorities. Those candidatesfollow the assigned priorities to relay the packet. Each candi-date, if having received the data packet correctly, will start atimer whose value depends on its priority. The higher the pri-ority, the shorter is the timer value. The candidate whose timerexpires will reply with an ACK to notify the sender, as wellas to suppress other contenders. Then, it rebroadcasts the datapacket toward its downstream link. If no forwarding candidatehas successfully received the packet, the sender will retransmitthe packet if the retransmission mechanism is enabled. We de-note as the backoff timer value of the th candidate. Sincethe lower priority forwarding candidate needs to wait and con-firm that no higher priority candidate has relayed the packet be-fore it takes the forwarding task, is an increasing functionof . Suppose the link layer protocol is based on CSMA/CAMAC, can be defined as

(2)

where is the value of Short Inter Frame Space, isthe transmission delay for sending an ACK.We denote as the one-hop medium delay [23] of theth candidate, which is the time interval between the senderbroadcasting a data packet and the th candidate claiming thatit has received the packet. We define this as

(3)

where is the value of Distributed Interframe Space(DIFS), is the transmission delay of data packet, andthe signal propagation delay is ignored. For sender , given

sequential forwarding candidates , wehave the expected one-hop media delay as

(4)

where and .In order to forward data toward the destination with minimum

number of transmissions, we can apply the basic greedy for-warding rule in geographic routing [24] as the relay priority rulein R3E, i.e., data packets are greedily forwarded to the neighborgeographically closest to the destination. In this way, R3E ob-viates the necessity of utilizing location information, while en-abling data packets to be greedily forwarded toward the des-tination with the help of the robust guide path. Note that therelay priority rule can also adopt other variant metrics, e.g.,the one-hop throughput metric [23] to achieve the best paththroughput.From the realistic link conditions in wireless networks, a

potential forwarder with a higher PRR toward the downstreamguide node possibly has a shorter distance from that guidenode, as longer distances normally result in lower receivedsignal strength and thus increased probability of packet loss[25]. Therefore, the relay priority rule is as follows. When aguide node transmits the data packet, the downstreamguide node has the highest priority; and the helper nodesin are ordered descendingly according to theirPRRs toward . Suppose the downstream guide nodefails to receive the packet, while a helper inreceives the packet and takes the forwarding task. The for-warding candidates of are given by . Morespecifically, the forwarding candidate set of is composedof three parts: 1) the helper nodes who have higher prioritiesthan in ; 2) the downstream guide node ; and3) . To achieve the minimumnumber of transmissions, the relay priorities are ordered as:Priority of 3) Priority of 2) Priority of 1).Revisiting the examples in Figs. 3 and 5, we show the helper

nodes and their priorities at each hop in the data forwardingphase in Fig. 6(a). The helper nodes and their priorities at thefirst hop are . Suppose guide node C fails to receivethe packet correctly, while helper B successfully receives thepacket, as shown in Fig. 6(b). B takes the forwarding task in-stead of C. Then B updates its helper set as and forwardsthe data packet to its downstream potential forwarders. It can beseen that R3E is resilient to the wireless link dynamics.

IV. PERFORMANCE EVALUATION

Here, we present performance evaluation results. We imple-ment R3E as an extension to AODV [3], named AODV-R3E,using the ns-2 simulator [26], and compare the performancewith three baseline routing protocols. All the results have beenaveraged over 100 runs, and the related standard deviations areprovided as error bars.


Fig. 6. (a) Example illustrating the forwarding candidates and their priorities ateach hop. (b) Corresponding cooperative forwarding procedure at the first hopin the MAC layer.

Fig. 7. Example illustrating the primary path in REPF [27]. Only the nodeswhich can connect the two-hop away primary forwarding nodes are consideredas helper nodes. Thus, it restricts the cooperative forwarding to a very limitedscope.

A. Comparison Baselines

We first introduce the comparison baselines. We choose threerouting protocols, described here.• AODV-ETX[3]: The route discovery phase finds aleast-ETX (expected transmission count) path from asource to a destination. In AODV-ETX, the link layer re-transmission is enabled, i.e., at most three retransmissionsat each hop are sanctioned. Note that the other routingprotocols do not adopt the retransmission mechanism.

• REPF [27]: REPF (Reliable and Efficient Packet For-warding) protocol is designed to improve the AODVrouting performance by utilizing local path diversity.The route discovery phase finds an efficient primary path(composed of a set of primary forwarding nodes) in termsof the accumulated path ETX, and alternative paths whichhave similar cost. However, REPF restricts the helpernodes to a very limited scope, i.e., only the nodes whichcan connect the two-hop away primary forwarding nodesare considered as helper nodes, as shown in Fig. 7. As aresult, it does not fully utilize the forwarding opportuni-ties provided by available neighboring nodes in evenlydistributed networks.

• GOR[24]: In order to show that R3E enables data packetsto be greedily progressed toward the destination, we alsoreport the evaluation results of the Geographic Oppor-tunistic Routing (GOR). In our simulation, both R3E and

GOR follow the same relay priority rule, i.e., minimizingthe number of end-to-end data transmissions. We im-plement GOR as follows: all of the one-hop neighborsthat are nearer from the destination than the current for-warding node and can hear from each other are selected ashelper nodes, and the nodes closer to the destination aregiven higher relay priorities. Since the network is denselydeployed, the routing recovery mechanism bypassing“holes” is not considered in the simulations.

B. Simulation Setup

We define the node density as the number of nodes deployedin a 200 m 200 m square area. One hundred randomly con-nected topologies for each node density setting are generatedusing the setdest tool in ns-2, ranging from 100 to 200. The nodetransmission range is set to 50 m. The destination node is po-sitioned at bottom left (0 m, 0 m), and the source node is posi-tioned at top right (200 m, 200 m). In AODV-R3E, the systemparameter is set to 0.005s. In order to highlight the poten-tial collision problem between the returning RREP and RREQs,RREP is not acknowledged by the guide nodes at each hop inour simulation.We use the Nakagami distribution defined as

(5)

to describe the power of a received signal, where is theGamma function, denotes the Nakagami fading parameter,and is the average received power. We set in our sim-ulation. Assuming TwoRayGround signal propagation, can beexpressed in

(6)

as a function of , the distance between the sender and receiver.In (6), where is the transmission power, and are the an-tenna gains, and are the antenna heights, and is the lossfactor, and is the path-loss exponent. We set ,

, , and in our simulation. We as-sume that a packet is received successfully if the received signalpower is greater than the receiving power threshold. Then, byusing (5) and (6), we can derive the PRR at a certain distance[28].We choose four main evaluation metrics.• Packet delivery ratio: the ratio of the number of packetsreceived by the destination to the total number of packetssent by the source.

• End-to-end delay: the time taken for a packet to be trans-mitted from the source node to the destination node.

• Data transmission cost: it is measured as the total numberof data transmissions for an end-to-end delivery per packet.

• Control message cost: it is defined as the total numberof control message transmissions (such as RTS, CTS andACK) for sending a single packet to the destination.

C. Performance Overview

We evaluate the impact of network density on the perfor-mance of the different protocols. The evaluation results are


Fig. 8. Performance comparison, only the successful end-to-end transmissions are counted. (a) Packet delivery ratio. (b) End-to-end delay. (c) Data transmissioncost. (d) Control message cost.

shown in Fig. 8, where the node density is varied from 100 to200.Fig. 8(a) illustrates the end-to-end packet delivery ratio under

different node densities. GOR achieves very high packet de-livery ratio, since the node density is always high, and it takesfull advantage of the local forwarding opportunities, involvingmaximum possible number of neighbors which are closer to thedestination. For AODV-ETX, it already finds a more efficientpath using the ETX metric than using the hop-count metric [3].In addition, packets are forwarded using unicast transmission,and three retransmissions at each hop are allowed in the linklayer in AODV-ETX. That is why it also achieves a high packetdelivery ratio. We observe that AODV-R3E shows above 95%packet delivery ratio on average, which is comparable to thatof AODV-ETX and GOR, without enabling the MAC layer re-transmission mechanism. However, REPF only provides around20% packet delivery ratio on average. The main reason is thatREPF restricts the node cooperation to a very limited scope.With the increase of node density, the intuition is that the

packet delivery ratio should also increase. However, it is sur-prising that the delivery ratio drops a little as the node densityincreases in AODV-R3E, as shown in Fig. 8(a). An examinationof the simulation traces in MAC layer reveals that, as the net-work becomes more congested, the RREP returned by the desti-nation node may collide with RREQs in the etwork. Therefore,it is better that each guide node acknowledges the reception ofthe RREP.Fig. 8(b) depicts the performance comparison on the

end-to-end packet delivery delay. We know that the delayincurred by retransmission is much larger than the coordinationdelay introduced by the cooperative forwarding. Therefore,AODV-ETX requires longer time to finish the end-to-endtransmission, because it requires several retransmissions toforward a packet to the destination. AODV-R3E shows veryclose performance to the GOR. Note that only the successfulend-to-end transmissions are counted in Fig. 8. From Fig. 8(b),it seems that REPF yields the best performance. Actually, thisis not true. Because it only shows the results for around 20%successful transmissions. REPF should yield worse results ifthe failed end-to-end transmissions are counted, which we willdiscuss in Section IV-D. Overall, AODV-R3E can provide 50%improvement over REPF in terms of the delivery delay.Fig. 8(c) and (d) reports the changes of the data transmission

cost and control message cost, respectively, under different nodedensities. From Fig. 8(c), it clearly shows that REPF requires

two or three retransmissions to finish the transmission task. Asexpected, AODV-R3E only incurs approximately 5% highertransmission cost than the ideal baseline scheme GOR. How-ever, GOR requires the location information. Since the traffic istransmitted using unicast in AODV-ETX, it incurs a higher con-trol message overhead compared with other protocols, as shownin Fig. 8(d).Another interesting observation about AODV-R3E and

GOR is that, in Fig. 8(b), as the node density increases, theend-to-end delay also slowly increases, whereas the numberof data transmissions slightly decreases, as can be seen inFig. 8(c). This is because of the relay priority rule applied inAODV-R3E and GOR. As the node density increases, moreavailable forwarding candidates can be involved into the coop-erative forwarding. However, the higher forwarding prioritieswill be assigned to neighbors with better advancements towardsthe destination, but with relatively lower PRRs. Consequently,it increases the one-hop medium time [23], which is defined asthe interval between the sender broadcasting a data packet andthe forwarding candidate’s first claim of receiving the packet.

D. Insights

Here, we investigate the internal state of each routing pro-tocol and understand the underlying reasons why AODV-R3Eprovides better performance than other alternative solutions.Fig. 9(a) and (b) illustrates the cumulative distribution func-

tion (CDF) curves of end-to-end delivery delay in 100-node net-works and 200-node networks, respectively. Since REPF onlyachieves around 20% successful transmissions, for a fair com-parison, we compare the top 20% simulation rounds with lowerend-to-end delivery delay of AODV-R3E and GOR with all theresults of REPF. For example, the average delivery delays forGOR, AODV-R3E and REPF in 100-node networks are 0.0513,0.0528, and 0.0538 s. From Fig. 9, we observe that GOR andAODV-R3E experience shorter delivery delays than REPF.Fig. 10(a) and (b) reports the CDF of forwarding candidates

at each hop for GOR, AODV-R3E and REPF. It is obvious thatAODV-R3E explores a larger cooperation opportunity thanthe other two schemes. The reason is that, AODV-R3E finds arouting path with more cooperative forwarding opportunitiesin the route discovery phase, whereas GOR applies the greedyforwarding scheme with only local information. Therefore,choosing the optimal routing direction is not guaranteed inGOR. REPF has the least number of forwarding candidates ateach hop, which confirms our argument that REPF restricts the


Fig. 9. CDF of end-to-end delay: (a) 100-node networks and (b) 200-nodenetworks.

Fig. 10. CDF of forwarding candidates at each hop: (a) 100-node networks and(b) 200-node networks.

Fig. 11. One-hop delivery ratio.

cooperative forwarding to a very limited area. For example,when the node density is 100, REPF involves three forwardingcandidates at each hop, whereas GOR and AODV-R3E, onaverage, have 3.6 and 4.3 forwarding candidates, respectively.In 200-node networks, REPF has 5.5 forwarding candidates,whereas GOR and AODV-R3E on average involve 6.7 and 7.5forwarding candidates at each hop, respectively.Although having more cooperative forwarding opportunities

per hop than GOR, AODV-R3E still generates a little larger de-livery delay and transmission cost than that of GOR. The reasonis that AODV-R3E prioritizes forwarding candidates based ononly the PRRs toward the downstream guide node. In somecases, the potential forwarder with the highest PRR toward thedownstream guide node is not necessarily the closest node to thedestination node. Therefore, AODV-R3E does not always findthe optimal order of forwarding candidates without utilizing thelocation information.Fig. 11 plots the one-hop delivery ratio for AODV-R3E

and REPF, which is defined as the probability that at leastone forwarding candidate successfully receives the packet ateach hop. From the figure, we can see that, as compared with

Fig. 12. Route discovery cost.

Fig. 13. Unreliable link models in the simulations.

REPF, AODV-R3E yields a remarkable increase of 20% inone-hop delivery ratio. It reveals the reason why REPF onlyachieves a very low packet delivery ratio on average when theretransmission mechanism is disabled.Fig. 12 compares the route discovery overhead between

AODV-R3E and REPF, which measures the total number ofrouting discovery control message transmissions, e.g., RREQand RREP. REPF incurs much higher communication overheadand increases the network load during the route discoveryphase, since it allows duplicate propagating RREP packets inorder to find a better virtual path. We observe that the routediscovery overhead in both AODV-R3E and AODV-ETX is

, where is the number of nodes. However, the routediscovery overhead in REPF increases exponentially as thenetwork size increases.

E. Results With Increased Packet Losses

In an IWSN, the harsh industrial environments may result invery unreliable wireless links due to many factors [1], [8]. Toevaluate the effect of increased packet losses, here, we use asimple linear link lossy model based on the distance betweentransmitter and receiver. In these simulations, a link with a dis-tance of 0 m has 0% probability of additional packet loss, anda link with distance of 50 m has 50% probability of additionalloss, which is the same as the setting in [25]. The PRRs de-rived from the distance between transmitter and receiver withthis modified channel model are shown in Fig. 13.Fig. 14(a) shows the end-to-end packet delivery ratio as

the node density increases. Compared with the results usingthe channel model without increased packet losses shown inFig. 8(a), the packet delivery ratio for all protocols decreases.From this figure, we can see that AODV-R3E even outperforms


Fig. 14. Performance comparison with increased packet losses, only the successful end-to-end transmissions are counted. (a) Packet delivery ratio. (b) End-to-enddelay. (c) Data transmission cost. (d) Control message cost.

GOR with increased packet losses and shows the highest av-erage packet delivery ratio among these protocols. It indicatesthat finding a reliable guide path is very important if the wire-less links are highly unreliable. We also observe that generally,as the node density increases, the packet delivery ratio for bothAODV-R3E and GOR also increases. While the AODV-ETXonly provides less than 50% delivery ratio, REPF shows lessthan 10% delivery ratio.Fig. 14(b) plots the end-to-end delay among the four

compared schemes. For the same reason discussed earlier,AODV-R3E has a little larger delay than GOR. When com-paring with the results shown in Fig. 8(b), the delivery delaysof both AODV-R3E and GOR increase due to the poor linkqualities.Fig. 14(c) and (d) shows the data transmission cost and

control message cost, respectively, under the modified channelmodel. It is interesting that, for AODV-R3E and GOR, theperformance on these two metrics does not change muchcompared with the results in Fig. 8(c) and (d). The reason forthis is that these results only count the successful end-to-endtransmissions, and there is no retransmission in AODV-R3Eand GOR. Thus, the end-to-end hop-count does not increasefor those successful transmissions. However, in the modifiedchannel model with increased packet losses, the end-to-enddelivery delay increases since the one-hop medium time isincreased. This is because, from (4), the one-hop medium timeincreases as the link qualities become poor.

F. Performance Summary

In the previous sections, we see that AODV-R3E without uti-lizing the location information achieves a comparable perfor-mance to GOR. Compared with REPF, it significantly improvesthe packet delivery ratio, and, compared with AODV-ETX, itprovides much lower data transmission cost and control mes-sage cost. Remarkably, in the modified channel model with in-creased packet losses, AODV-R3E even performs better thanGOR, since it forwards data packets along the path which pro-vides more cooperative forwarding opportunities.

V. RELATED WORK

Opportunistic routing to combat fading channels in wirelessnetworks has been an active research area. A number of oppor-tunistic routing protocols have been proposed for this purpose[15], [16]. ExOR [11] is a representative opportunistic routing

protocol. However, ExOR, as well as SOAR [29], requires everynode to measure and maintain the global topology of the net-work. In geographic random forwarding (GeRaF) [24], the for-warding node broadcasts data packets to a collection of neigh-bors which are prioritized according to their advancements to-ward the destination. The node that is closest to the destinationwill be chosen as next-hop forwarder in a distributed manner.In [30], the authors proposed a contention-based geographicrouting protocol which incorporates the cooperative relayingand leapfrogging. RRP [22] was proposed to enhance the ro-bustness of routing against path breakage, e.g., due to channelinterference or node mobility. The issue addressed in RRP ishow nearby nodes cooperatively forward packets for unreliablemobile wireless sensor networks. In cluster-based forwarding(CBF) [31], each node forms a cluster such that any node inthe next-hop’s cluster can take the forwarding task. However,these existing works assume that a path has already been estab-lished between a source and a destination, leaving the robustend-to-end path discovery unaddressed.In recent years, IWSNs have been designed and developed

for many industrial applications, e.g., industrial process moni-toring and control, environment and equipment monitoring, anddistributed control systems, etc. [32]–[35]. They are expected totake the place of conventional industrial wired communicationsystems [1], [2], [36], making it possible to wirelessly mon-itor and control the environment in real time. Although thereexists a large research effort in the area of WSNs, many re-search issues have not been well addressed such that WSNs canbe adopted properly for industrial automation. In [37], Aker-berg et al. discuss major requirements for typical WSN appli-cations in process automation and outline the research direc-tion for IWSNs. EARQ [38] presents a routing protocol forIWSNs to achieve real-time, energy-efficient, and reliable de-livery of a packet. [39] and [40] propose techniques for pro-viding quality-of-service (QoS) support regarding to timelinessand reliability in IWSNs.In industrial environments, wireless links are normally highly

dynamic due to many factors such as RF interference, harshenvironments and vibrations. Providing reliable and efficientcommunication in IWSNs is a challenging problem [41]. Re-cently, WirelessHART [42] and ISA 100.11a [43] have beenproposed specifically for reliable and real-time wireless com-munication in industrial environments. Yang et al. [44] proposea CCA-Embedded TDMA scheme to improve the transmissionefficiency of WirelessHART MAC. The authors in [45], [46]


evaluate the potentials of flooding as a data dissemination tech-nique in IWSNs. Quang et al. [47] propose a gradient-basedrouting with two-hop information for IWSNs to enhance thereal-time performance. In this work, we address the reliablerouting problem inWSNs/IWSNs by applying the opportunisticrouting paradigm to reactive routing protocols, and jointly opti-mizing the route discovery and cooperative forwarding.

VI. CONCLUSION

In this work, we presented R3E, which can augment most ex-isting reactive routing protocols in WSNs/IWSNs to provide re-liable and energy-efficient packet delivery against the unreliablewireless links. We introduced a biased backoff scheme in theroute discovery phase to find a robust virtual path with low over-head. Without utilizing the location information, data packetscan still be greedily progressed toward the destination along thevirtual path. Therefore, R3E provides very close routing per-formance to the geographic opportunistic routing protocol. Weextended AODV with R3E to demonstrate its effectiveness andfeasibility. Simulation results showed that, as compared withother protocols, AODV-R3E can effectively improve robust-ness, end-to-end energy efficiency, and latency.

REFERENCES[1] V. Gungor and G. Hancke, “Industrial wireless sensor networks: Chal-

lenges, design principles, and technical approaches,” IEEE Trans. Ind.Electron., vol. 56, no. 10, pp. 4258–4265, Oct. 2009.

[2] S. eun Yoo, P. K. Chong, D. Kim, Y. Doh, M.-L. Pham, E. Choi, andJ. Huh, “Guaranteeing real-time services for industrial wireless sensornetworks with IEEE 802.15.4,” IEEE Trans. Ind. Electron., vol. 57, no.11, pp. 3868–3876, Nov. 2010.

[3] C. Perkins and E. Royer, “Ad-hoc on-demand distance vector routing,”in Proc. IEEE WMCSA, 1999, pp. 90–100.

[4] M. Marina and S. Das, “On-demand multipath distance vector routingin ad hoc networks,” in Proc. IEEE ICNP, Nov. 2001, pp. 14–23.

[5] D. B. Johnson and D. A. Maltz, “Dynamic source routing in ad hocwireless networks,” Mobile Computing, pp. 153–181, 1996.

[6] K. A. Agha, M.-H. Bertin, T. Dang, A. Guitton, P. Minet, T. Val, andJ.-B. Viollet, “Which wireless technology for industrial wireless sensornetworks? the development of Ocari technology,” IEEE Trans. Ind.Electron., vol. 56, no. 10, pp. 4266–4278, Oct. 2009.

[7] K. Yu, M. Gidlund, J. Åkerberg, and M. Björkman, “ReliableRSS-based routing protocol for industrial wireless sensor networks,”in Proc. IECON, 2012, pp. 3231–3237.

[8] “U. D. of Energy, Industrial wireless technology for the 21st century,”Office of Energy and Renewable Energy Report, 2002.

[9] Y. Gu and T. He, “Dynamic switching-based data forwarding for low-duty-cycle wireless sensor networks,” IEEE Trans. Mobile Comput.,vol. 10, no. 12, pp. 1741–1754, Dec. 2011.

[10] V. Gungor, O. Akan, and I. Akyildiz, “A real-time and reliable trans-port protocol for wireless sensor and actor networks,” IEEE/ACMTrans. Netw., vol. 16, no. 2, pp. 359–370, Apr. 2008.

[11] S. Biswas and R. Morris, “ExOr: opportunistic multi-hop routing forwireless networks,” in Proc. ACM SIGCOMM, 2005, pp. 133–144.

[12] S. Katti, H. Rahul, W. Hu, D. Katabi, M. Medard, and J. Crowcroft,“XORs in the air: Practical wireless network coding,” IEEE/ACMTrans. Netw., vol. 16, no. 3, pp. 497–510, Jun. 2008.

[13] S. Chachulski, M. Jennings, S. Katti, and D. Katabi, “Trading structurefor randomness in wireless opportunistic routing,” in Proc. AGM SIG-COMM, 2007, pp. 169–180.

[14] K. Zeng,W. Lou, J. Yang, and D. Brown, “On geographic collaborativeforwarding in wireless ad hoc and sensor networks,” in Proc. WASA,2007, pp. 11–18.

[15] X. Mao, S. Tang, X. Xu, Li X.-Y, and H. Ma, “Energy-efficient op-portunistic routing in wireless sensor networks,” IEEE Trans. ParallelDistrib.Syst., vol. 22, no. 11, pp. 1934–1942, Nov. 2011.

[16] R. Bruno and M. Nurchis, “Survey on diversity-based routing in wire-less mesh networks: Challenges and solutions,” Computer Commun.,vol. 33, no. 3, pp. 269–282, Feb. 2010.

[17] E. Royer and C.-K. Toh, “A review of current routing protocols for adhoc mobile wireless networks,” IEEE Pers. Commun., vol. 6, no. 2, pp.46–55, 1999.

[18] R. Shah, S. Wietholter, A. Wolisz, and J. Rabaey, “When does op-portunistic routing make sense?,” in Proc. IEEE PerCom Workshops,2005, pp. 350–356.

[19] K.-H. Kim and K. G. Shin, “On accurate measurement of link qualityin multi-hop wireless mesh networks,” in Proc. ACMMobiCom, 2006,pp. 38–49.

[20] R. Fonseca, O. Gnawali, K. Jamieson, and P. Levis, “Four-bit wirelesslink estimation,” in Proc. HotNets VI, 2007, pp. 1–7.

[21] J. Sanchez, R. Marin-Perez, and P. Ruiz, “Boss: Beacon-less on de-mand strategy for geographic routing in wireless sensor networks,” inProc. IEEE MASS, 2007, pp. 1–10.

[22] X. Huang, H. Zhai, and Y. Fang, “Robust cooperative routing protocolin mobile wireless sensor networks,” IEEE Trans. Wireless Commun.,vol. 7, no. 12, pp. 5278–5285, Dec. 2008.

[23] K. Zeng, W. Lou, J. Yang, and D. R. Brown, “On throughput efficiencyof geographic opportunistic routing in multihop wireless networks,”Mobile Netw. Applicat., vol. 12, no. 5, pp. 347–357, 2007.

[24] M. Zorzi and R. R. Rao, “Geographic random forwarding (GeRaF) forad hoc and sensor networks: Energy and latency performance,” IEEETrans. Mobile Comput., vol. 2, no. 4, pp. 349–365, Apr. 2003.

[25] Y. Sun, O. Gurewitz, S. Du, L. Tang, and D. B. Johnson, “ADB: anefficient multihop broadcast protocol based on asynchronous duty-cycling in wireless sensor networks,” in Proc. ACM SenSys, 2009,pp. 43–56.

[26] Network simulator [Online]. Available: http://www.isi.edu/nsnam/ns/[27] W. A. Hoc, J. Wang, H. Zhai, W. Liu, and Y. Fang, “Reliable and

efficient packet forwarding by utilizing path diversity in wireless adhoc networks,” in Proc. IEEE Milcom, 2004, pp. 258–264.

[28] L. Cheng, J. Cao, C. Chen, J. Ma, and S. Das, “Exploiting geographicopportunistic routing for soft QoS provisioning in wireless sensor net-works,” in Proc. IEEE MASS, 2010, pp. 292–301.

[29] E. Rozner, J. Seshadri, Y. Mehta, and L. Qiu, “Soar: Simple oppor-tunistic adaptive routing protocol for wireless mesh networks,” IEEETrans. Mobile Comput., vol. 8, no. 12, pp. 1622–1635, Dec. 2009.

[30] P. Coronel, R. Doss, and W. Schott, “Geographic routing with cooper-ative relaying and leapfrogging in wireless sensor networks,” in Proc.IEEE GLOBECOM, 2007, pp. 646–651.

[31] Q. Cao, T. Abdelzaher, T. He, and R. Kravets, “Cluster-based for-warding for reliable end-to-end delivery in wireless sensor networks,”in Proc. IEEE INFOCOM, 2007, pp. 1928–1936.

[32] A.Willig, “Recent and emerging topics in wireless industrial communi-cations: A selection,” IEEE Trans. Ind. Inf., vol. 4, no. 2, pp. 102–124,May 2008.

[33] K. Low, W. Win, and M. Er, “Wireless sensor networks for industrialenvironments,” in Proc. Int. Conf. Intell. Agents, Web Technol. InternetCommerce, Nov. 2005, vol. 2, pp. 271–276.

[34] V. C. Gungor and F. C. Lambert, “A survey on communication net-works for electric system automation,” Comput. Netw., vol. 50, no. 7,pp. 877–897, May 2006.

[35] B. Lu and V. Gungor, “Online and remote motor energy monitoringand fault diagnostics using wireless sensor networks,” IEEE Trans. Ind.Electron., vol. 56, no. 11, pp. 4651–4659, Nov. 2009.

[36] T. Chiwewe and G. Hancke, “A distributed topology control techniquefor low interference and energy efficiency in wireless sensor networks,”IEEE Trans. Ind. Inf., vol. 8, no. 1, pp. 11–19, Feb. 2012.

[37] J. Akerberg, M. Gidlund, and M. Bjorkman, “Future research chal-lenges in wireless sensor and actuator networks targeting industrial au-tomation,” in Proc. IEEE INDIN , 2011, pp. 410–415.

[38] J. Heo, J. Hong, and Y. Cho, “Earq: Energy aware routing for real-timeand reliable communication in wireless industrial sensor networks,”IEEE Trans. Ind. Inf., vol. 5, no. 1, pp. 3–11, Feb. 2009.

[39] L. Xue, X. Guan, Z. Liu, and B. Yang, “Tree: Routing strategy withguarantee of QoS for industrial wireless sensor networks,” Int. J.Commun. Syst., 2012, DOI 10.1002/dac.2376.

[40] Y. Li, C. S. Chen, Y.-Q. Song, Z. Wang, and Y. Sun, “Enhancing real-time delivery in wireless sensor networks with two-hop information,”IEEE Trans. Ind. Inf., vol. 5, no. 2, pp. 113–122, May 2009.


[41] “A QoS aware route selection algorithm for industrial wireless sensornetworks,” Ad Hoc Networks, vol. 10, no. 3, pp. 458–478, 2012.

[42] “Hart Specification,” [Online]. Available: http://www.hartcomm.org[43] “Isa-100, Wireless Systems for Automation,” [Online]. Available:

http://www.isa.org/isa100[44] “CCA-embedded TDMA enabling acyclic traffic in industrial wireless

sensor networks,” Ad Hoc Networks, vol. 11, no. 3, pp. 1105–1121,2013.

[45] F. Barac, J. Akerberg, andM. Gidlund, “A lightweight routing protocolfor industrial wireless sensor and actuator networks,” in Proc. IECON,2011, pp. 2980–2985.

[46] F. Barac, K. Yu, M. Gidlund, J. Akerberg, and M. Bjorkman, “Towardsreliable and lightweight communication in industrial wireless sensornetworks,” in Proc. IEEE INDIN , 2012, pp. 1218–1224.

[47] P. T. A. Quang and D.-S. Kim, “Enhancing real-time delivery of gra-dient routing for industrial wireless sensor networks,” IEEE Trans. Ind.Inf., vol. 8, no. 1, pp. 61–68, Feb. 2012.

Jianwei Niu (M’06) received the Ph.D. degreein computer science from Beijing University ofAeronautics and Astronautics, Beijing, China, in2002.He was a Visiting Scholar with the School of

Computer Science, Carnegie Mellon University,Pittsburgh, PA, USA from January 2010 to February2011. He is currently a Professor with the Schoolof Computer Science and Engineering, BeijingUniversity of Aeronautics and Astronautics, Beijing,China. He has authored and coauthored more than

80 refereed papers and filed more than 30 patents. He has served as an editor ofJournal of Network and Computer Applications. His current research interestsinclude mobile and pervasive computing.Dr. Niu served as the Program Co-chair of IEEE Sec 2008, Vice Co-chair of

TPC of CPSCom 2013, TPC members of ICC, WCNC, Globecom, LCN, andetc. He won the Best Paper Award in WCNC 2013, ICACT 2013, CWSN 2012,and GreenCom 2010.

Long Cheng (M’12) received the B.S. degree incomputer Science from Xi’an TelecommunicationInstitute, Xi’an, China, 2004, the M.S. degreein telecommunication engineering from XiDianUniversity, Xi’an, China, in 2007 and the Ph.D.degree from the State Key Lab of Network andSwitching Technology, Beijing University of Postsand Telecommunications, Beijing, China, in 2012.He is a Postdoctoral Researcher with Beihang Uni-

versity and Singapore University of Technology andDesign, Singapore. During the period from June 2009

to December 2009, he was with the Hong Kong Polytechnic University as a Re-search Assistant. From January 2010 to September 2011, he was a Joint PhDstudent in the Department of Computer Science and Engineering, University ofTexas at Arlington. His current research interests include wireless sensor net-work, Internet of things, mobile computing, and pervasive computing.

Yu Gu (M’10) received the Ph.D. degree from theUniversity of Minnesota, Twin Cities, MN, USA, in2010.He is currently an Assistant Professor with

the Pillar of Information System Technology andDesign, Singapore University of Technology andDesign, Singapore. He is the author and coauthor ofover 40 papers in premier journals and conferences.His publications have been selected as graduate-levelcourse materials by over 20 universities in the UnitedStates and other countries. His research includes

networked embedded systems, wireless sensor networks, cyber-physicalsystems, wireless networking, real-time and embedded systems, distributedsystems, vehicular ad-hoc networks and stream computing systems.Dr. Gu is a member of the Association for Computing Machinery.

Lei Shu (M’10) received the B.Sc. degree in com-puter science from South Central University for Na-tionalities, Wuhan, China, in 2002, the M.Sc. degreein computer engineering fromKyungHeeUniversity,Seoul, Korea, in 2005, and the Ph.D. degree from theDigital Enterprise Research Institute, National Uni-versity of Ireland, Galway, Ireland, in 2010.He is a Professor with the Guangdong University

of Petrochemical Technology, Maoming, China. Hewas a Specially Assigned Research Fellow with theDepartment of Multimedia Engineering, Graduate

School of Information Science and Technology, Osaka University, Osaka,Japan, and was a Research Scientist with the Digital Enterprise ResearchInstitute (DERI), National University of Ireland, Galway, Ireland. He hasauthored and coauthored over 100 papers in related conferences, journals, andbooks.Prof. Shu is a member of the Association for Computing Machinery.

Sajal K. Das (SM’08) is a University Distin-guished Scholar Professor of Computer Scienceand Engineering and the Founding Director ofthe Center for Research in Wireless Mobility andNetworking (CReWMaN), University of Texas atArlington, Arlington, TX, USA. During 2008–2011,he served the National Science Foundation (NSF) asa Program Director with the Division of ComputerNetworks and Systems. He holds five U.S. patentsand coauthored books, including Smart Environ-ments: Technology, Protocols, and Applications

(Wiley, 2005), Handbook on Securing Cyber-Physical Critical Infrastructure:Foundations and Challenges (Morgan Kaufmann, 2012), and Mobile Agentsin Distributed Computing and Networking (Wiley, 2012). He serves as thefounding editor-in-chief of Pervasive and Mobile Computing and also as anassociate editor ACM/Springer Wireless Networks, Journal of Parallel andDistributed Computing, and Journal of Peer-to-Peer Networking. His currentresearch interests include wireless and sensor networks, mobile and pervasivecomputing, cloud computing, cyberphysical systems and smart heath care,security and privacy, biological and social networks, applied graph theory andgame theory. He has authored and coauthored extensively in these areas andmade fundamental contributions.Prof. Das was a recipient of the IEEE Computer Society Technical Achieve-

ment Award for pioneering contributions to sensor networks and mobile com-puting; IEEE Region 5 Outstanding Engineering Educator Award; and EightBest Paper Awards including those at IEEE SmartGridComm’12, QShine’09,EWSN’08, IEEE PerCom’06, and ACM MobiCom’99. He is the cofounder ofIEEE PerCom, WoWMoM and ICDCN conferences.

784 ieee transactions on industrial … · 784 ieee transactions on industrial informatics, vol....

Documents