This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 1

Energy-Efficient Wireless MAC Protocolsfor Railway Monitoring Applications

G. M. Shafiullah, Member, IEEE, Salahuddin A. Azad, and A. B. M. Shawkat Ali, Senior Member, IEEE

Abstract—Recent advances in wireless sensor networking(WSN) techniques have encouraged interest in the developmentof vehicle health monitoring (VHM) systems. These have thepotential for use in the monitoring of railway signaling systemsand rail tracks. Energy efficiency is one of the most impor-tant design factors for the WSNs as the typical sensor nodesare equipped with limited power batteries. In earlier research,an energy-efficient cluster-based adaptive time-division multiple-access (TDMA) medium-access-control (MAC) protocol, namedEA-TDMA, has been developed by the authors for the purposeof communication between the sensors placed in a railway wagon.This paper proposes another new protocol, named E-BMA, whichachieves even better energy efficiency for low and medium trafficby minimizing the idle time during the contention period. In addi-tion to railway applications, the EA-TDMA and E-BMA protocolsare suitable for generic wireless data communication purposes.Both analytical and simulation results for the energy consumptionof TDMA, EA-TDMA, BMA, and E-BMA have been presented inthis paper to demonstrate the superiority of the EA-TDMA andE-BMA protocols.

Index Terms—Energy efficiency, medium access control (MAC)protocol, railway wagon, vehicle health monitoring (VHM), wire-less sensor network (WSN).

I. INTRODUCTION

W ITH the increased demand for railway services, railwaymonitoring systems continue to advance at a remarkable

pace to maintain reliable, safe, and secure operation. The lack ofsafety and security monitoring of railway infrastructure runs therisk of train collision, train derailment, terrorist threats, failuresin the train wagons, etc. The performance of rail vehicles run-ning on tracks is limited by the lateral instability inherent to thedesign of the wagon’s steering and the response of the railwaywagon to individual or combined track irregularities. Railwaytrack irregularities need to be kept within safe operating mar-gins by undertaking appropriate maintenance programs. Trackgeometry inspection and monitoring enhances train-operatingsafety and reduced vehicle and track dynamic interaction. Mon-

Manuscript received March 15, 2012; revised August 15, 2012 andOctober 15, 2012; accepted October 20, 2012. This work was supported inpart by Prof. P. Wolfs and in part by Prof. C. Cole, both from the Center forRailway Engineering, Central Queensland University. The Associate Editor forthis paper was B. Ning.

G. M. Shafiullah and S. A. Azad are with the Power Engineering ResearchGroup, School of Engineering and Built Environment, Central Queensland Uni-versity, Rockhampton, Qld. 4702, Australia (e-mail: g.shafiullah@cqu.edu.au;s.azad@cqu.edu.au).

A. B. M. S. Ali is with the School of Information and Communication Tech-nology, Central Queensland University, Rockhampton, Qld. 4702, Australia(e-mail: s.ali@cqu.edu.au).

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

Digital Object Identifier 10.1109/TITS.2012.2227315

Fig. 1. Typical scenario for railway-wagon health monitoring system.

itoring vehicle characteristics in real time from track measure-ment data has been addressed by various research organizations[2]–[7]. Wireless sensor networks (WSNs) are widely usedto monitor railway tracks and irregularities, detect abandonedobjects in railway stations, develop intrusion detection systems,secure railway operations, and monitor tunnels [8]–[10].

Seifert envisioned [8] that a network of smart sensors couldbe utilized to monitor public spaces for potential invasionto alert the operators at a control center about the event. Inaddition, a WSN can be deployed to monitor large areas withgreater efficacy in video-based intrusion detection systems.Aboelela et al. [9] proposed a new approach to reduce theaccident rate and increase the efficiency of railroad maintenanceactivities. The protocol adopts a multilayered multipath routingarchitecture in which each sensor transmits the sensed datato the two nearest cluster heads (CHs). Each CH aggregatesthe data using a fuzzy logic technique and transfers it to thesink node. Cheekiralla [10] designed a wireless sensor unit forthe surveillance of a train tunnel, which measures the verticaldisplacements along the critical zone of the tunnel duringadjacent construction activity.

The potential of WSN technology to monitor the railway-wagon health condition and the vertical displacement of railwaywagons due to track irregularities has yet to be fully explored.The limited lifetime of the batteries that power the sensor nodesmakes the energy efficiency a major design issue for WSNs[11]. This paper concentrates on developing an energy-efficientWSN MAC protocol to collect data from sensor nodes thatare placed inside the railway wagons and send the data tothe locomotive for further precautionary actions to prevent anyfuture disastrous events. A prototype of the proposed railway-wagon health monitoring system is given in Fig. 1. Although theproposed energy-efficient protocol is designed with the railwayapplications in mind, it is applicable to generic wireless datacommunication purposes. Analytical and simulation modelshave been developed for the existing and proposed protocols tocompare their performances in terms of energy consumption.

1524-9050/$31.00 © 2012 IEEE

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

II. BACKGROUND OF THE STUDY

Central Queensland University, in collaboration with theCenter for Railway Engineering [4], has been working on anautonomous health card device for online analysis of car bodymotion to perceive track condition and monitor derailment.The health card devices use an accelerometer and angular ratesensors with a coordinate transform to analyze the car bodymotions into six degrees of freedom [12], [13]. These healthcard devices inspect every wagon in the fleet using low-costsmart devices [4], [12]. An algorithm was developed, whichanalyzes signals coming from accelerometers mounted on thewagon body to measure the dynamic interaction between thetrack and the rail vehicle. The algorithm was validated usingcollected field data, e.g., accelerations measured at strategicpoints on the wagon body and the bogies.

Each prototype health card incorporates a 27-MHz mi-crocontroller with 256 kB of onboard RAM, four dual-axisaccelerometers, a Global Positioning Satellite receiver, twolow-power radios, lithium-ion batteries, and a solar panel. ARabbit 3000 processor is used, which requires 200 mW ofpower at 40 MHz. The first generation of the health card con-sumes a total of 400 mW or energy requirement of 9.6 Wh daily.An 80-Wh lithium battery is built into the health card that canprovide energy for up to eight days. Data were collected froma ballast wagon in which dual-axis accelerometers were fittedto each corner of the body and each side frame. A personal-computer-based data acquisition system was used to store data.The main purpose of the data acquisition was to provide realdata that are represented to the health card device. Data havebeen used to validate and demonstrate the effectiveness ofsignal analysis techniques and, finally, to develop a modelto monitor typical dynamic behavior and track irregularities[12], [13].

Both the vertical and lateral conditions of the railwaywagon have been measured by each accelerometer. The aimof the sensing arrangement was to capture roll, pitch, yaw,vertical, and lateral accelerations of the wagon body. TheADXL202/ADXL210 [14] dual-axis low-power low-cost ac-celeration sensor measured 16 channel acceleration data in gunits, with eight channels for the wagon body and eight forthe wagon side frame. Four sensor nodes were placed in eachwagon body, and the locations of the sensors were front-leftbody, front-right body, rear-left body, and rear-right body. Sim-ilarly, four sensor nodes were placed in each wagon’s side fame[4]. Sensor locations and naming convention are illustratedin Fig. 2. The sampling rate of the accelerometer can be setfrom 0.01 Hz to 5 KHz through adjustable capacitors, and theclock speed of this health card device was set to 100 Hz. Datawere continuously collected from a ballast wagon, which was aconventional three-piece bogie spaced lb = 10.97 m apart. Theaccelerometers were spaced l = 14.4 m apart. The test run wasa normal ballast lying operation, starting with a full load ofballast, traveling to the maintenance site, dropping the ballaston the track, and returning empty via the same route.

This research work is an extension of the existing health cardsystem development endeavor, aiming at improving the energyefficiency of the railway-wagon health monitoring system. Inthe proposed system, there are five sensor nodes placed inside

Fig. 2. Accelerometer locations and axis naming convention [4].

each wagon, instead of four in the existing system. One sensornode is used as a CH that collects data from other nodes andsends data to the central control room or base station (BS). Inthis system, an accelerometer has been placed in each corner ofthe wagon and one accelerometer at the center of the wagon,which acts as a CH. The BS is placed in the middle of the trainfor optimal signal transmission range. If there are W wagonsin the train, then the BS is positioned between wagons W/2and W/2 + 1. This feature not only reduces the overall energyconsumption of the network significantly but reduces energydissipation of each CH as well, as they need to transmit dataover a shorter range.

Each node forwards packets to the CH in each wagon, and theCH works as a router to send the data to the BS in the middleof the train. The locomotive driver can monitor the sensor datathrough an audio/visual system and take decisions accordingly.Each node is powered by an internal battery to make this workindependently.

The cluster-based WSN deployed in the railway-wagonhealth monitoring system, as illustrated, must be designed tobe very energy efficient and reliable. In a sensor node, power isrequired for data sensing, communication, and data processing.Energy efficiency is a major issue in designing WSN to prolongthe network lifetime as the sensor nodes have limited batterylifetime. The main sources of energy loss are idle listening, col-lision, overhearing, overemitting, and control packet overhead[11], [15].

The wireless modules of the health card were developedusing Bluetooth IEEE 802.15.1 standard [16], which is anoutdated standard. Bluetooth devices are inefficient in terms ofenergy dissipation. The data communication range of Bluetoothis only 10 m, which requires more of sensors per wagonfor data communication with the locomotive. The absence ofenergy-efficient features for data collection and communica-tion between wagons to the locomotive makes this systemless reliable. Instead of Bluetooth technology, this study con-siders the IEEE 802.15.4/ZigBee standard [17], which is anultralow-power and low-data-rate radio standard. Due to itssimplicity and low cost, ZigBee is the most suitable standardto date for railway applications, e.g., data communication be-tween sensor nodes placed inside the wagons. The CC2431

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAFIULLAH et al.: WIRELESS MAC PROTOCOLS FOR RAILWAY MONITORING APPLICATIONS 3

Fig. 3. Operation diagram of TDMA [22].

System-on-Chip [18] uses an energy-efficient ZigBee-enabledCC2420 RF transceiver [19] with an enhanced 8051 micro-controller, up to 128-KB Flash memory, 8 KB of RAM, andmany other powerful features, such as low current consumption,that makes the technology an attractive solution for WSNs. TheZigBee-compliant radio operates on 16 channels in the 2.4-GHzISM band, and standard data rates are 250 kb/s. This datatransfer capability is suitable for the sensor networks placed inthe railway wagon.

In general, the maximum length of the railway wagon is17 m. However, considering redundancy, it is wise to beconsidering a radio receiver with a 34-m range that coverstwo wagons. Hence, the transmission range of the receiver isexpected to be sufficient for the railway wagon as it covers35 m of wireless range.

IEEE 802.15.4/ZigBee standard has four states: sleep orshutdown, idle or listening, transmit, and receiving. It wasshown in an experiment by Bougard et al. [20] that the Zigbeestandard consumes less than 50% of the energy for actualdata transmission, and the rest of the energy is consumed forother activities. A significant percentage (25%) of the energy isconsumed during the contention procedure. This is due to themultiplicative effect of carrier-sense multiple access/collisiondetection (CSMA/CD). The waiting for an acknowledgementconsumes 15% of the energy. Moreover, 20% of the en-ergy is used for listening for the beacon by the transceiver.Based on the energy breakdown, several ways to improve theoverall energy efficiency were proposed by the researchers.Bougard et al. [20] proposed an energy-aware radio activationpolicy to optimize the PHY and MAC layers’ parameters in adense sensor network scenario. Experimental results showedthat PHY level improvements combined with MAC optimiza-tions allow energy-efficient self-powered sensor networks [20].

The traditional wireless MAC and routing protocols donot fulfill the requirements of WSN applications since WSNprotocols need to focus on energy-efficient design to ensureminimum power consumption and maximum battery lifetime[21], [22]. Energy-efficient MAC and routing protocol design iscurrently a prime research area in wireless data communicationapplication.

This paper concentrates on developing an energy-efficientWSN MAC protocol to collect data from sensor nodes that areplaced inside the railway wagons and send it to the locomotivefor further precautionary actions. The authors have already de-veloped an analytical model of an energy-efficient WSN MAC

protocol EA-TDMA [1], which is the most suitable for mediumto high traffic. This paper proposes another energy-efficientWSN MAC protocol, named E-BMA, which achieves evenbetter energy efficiency. Popular MAC protocols are discussedin the following section with their strengths and weaknesses.

III. SCHEDULE-BASED MEDIUM-ACCESS

CONTROL PROTOCOL

The major requirements of a wireless MAC protocol are:energy efficiency, scalability, latency, fairness, and bandwidthutilization. Contention-based protocols are scalable and adapt-able to node density or traffic load variations. However, theseschemes have a major limitation relating to an enormousamount of energy wasted due to collisions, overhearing, andidle listening [11], [21]. Schedule-based protocols are collisionfree and, hence, trim down the wastage of energy due to colli-sion. However, they lack the flexibility and scalability inherentin the contention-based protocols.

Time-division multiple access (TDMA) is a schedule-basedMAC protocol where the transmission channel is divided intoseveral time slots, and each node is assigned a time slot. Eachnode wakes up and transmits data only in its allocated time slotand remains in sleep mode in the remaining time slots [11],[21]. However, this protocol only uses the node energy effi-ciently when the traffic load is high. Nodes with empty bufferskeep their radio turned on during their scheduled slot and,hence, dissipate some of their remaining energy. The energy-efficient TDMA (E-TDMA) reduces energy consumption dueto idle listening. Sensor nodes keep their radios off when thereis no data to transmit. However, the CH has to keep the radioon all the time and hence waste energy [11], [21], [22]. Fig. 3illustrates a single round for TDMA protocol.

Low-energy adaptive clustering hierarchy (LEACH) [23], anarchitecture for wireless microsensor networks, incorporatesthe features of cluster-based routing and MAC protocol. Thisprotocol achieves energy efficiency and low latency whilemaintaining application-specific quality. LEACH allows alldata from nodes within the cluster to be locally processed inthe CH that reduces the data set. Data aggregation was doneto combine several correlated data signals into a smaller set ofinformation, and then, the resultant data were sent to the BSusing a fixed spreading code and a CSMA approach [23], [24].

The bit-map-assisted (BMA) protocol [25] is anotherschedule-based protocol that aims at reducing energy wastage

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 4. Operation diagram of the BMA protocol [25].

Fig. 5. Operation and timing diagram of the EA-TDMA protocol [1].

due to collision and idle listening. This protocol deals onlywith event-driven networks where sensor nodes forward datato the CH only if a significant event has been detected. Thecluster setup phase is similar to the LEACH [23] protocol.In the contention period, each node in the cluster transmitsa 1-bit control message to the CH node during its allocatedslot if it has data to transmit; otherwise, the transmitter radioremains idle. At the end of the contention period, the CH in theBMA protocol makes a transmission schedule and transmits theschedule only to the source nodes [25], [26]. In TDMA, oncea node is allocated a data slot, that allocation persists for allframes in that round regardless of whether the node has enoughdata packets to send in each frame. Conversely, in BMA, theallocation is done in the contention phase before the starting ofeach frame, as shown in Fig. 4. Therefore, BMA is more energy

efficient than TDMA and E-TDMA for the cases of low trafficload, relatively few sensor nodes per cluster, and relatively largepacket size [25], [26].

In railway applications, the accelerometer data are contin-uously collected while the train is in operation from sensornodes, and hence, this application is classified as a medium tohigh traffic load as the sensor collects data at the rate of 25 kb/s.Considering the application requirements, authors developedan energy-efficient protocol, named EA-TDMA [1], whichreduces the energy consumption during data transmission. Inthis protocol, every node wakes up in its allocated slot andtransmits data to the CH. If there are no data to send, it turns offthe radio immediately. The nodes move into sleep mode insteadof idle mode in the absence of data. An operation diagram anda timing diagram of the EA-TDMA protocol are illustrated in

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAFIULLAH et al.: WIRELESS MAC PROTOCOLS FOR RAILWAY MONITORING APPLICATIONS 5

Fig. 6. Operation of the E-BMA protocol.

Fig. 5(a) and (b), respectively. A detailed description of theEA-TDMA protocol is available in [1]. The energy consump-tion of EA-TDMA is significantly less than TDMA at lowtraffic loads, although this gap diminishes at high traffic loads.This protocol also outperforms BMA protocol in all trafficconditions except very low traffic [1]. In this paper, in additionto the analytical results, the superiority of the EA-TDMAprotocol has been demonstrated by the simulation model.

The railway-wagon health monitoring system requires theMAC protocol to be capable of handling steady traffic andenergy efficient. Although some of the aforementioned proto-cols were customized to achieve energy efficiency, this paperfurther explores the achievement of better energy efficiency. Inaddition to the EA-TDMA protocol in this paper, the authorspropose a new energy-efficient WSN MAC protocol, namedE-BMA. This paper explores both the analytical and simulationmodel of EA-TDMA and E-BMA protocols to demonstrate thesuperiority of these protocols compared with other conventionalprotocols. The proposed protocol achieves better energy effi-ciency for low to medium traffic load, and it is comparable withthe EA-TDMA and TDMA protocols for high traffic load. Thenewly proposed energy-efficient E-BMA protocol is describedin the next section.

IV. ENERGY-EFFICIENT E-BMA PROTOCOL

The BMA protocol consumes less energy than TDMA at lowand medium traffic loads, whereas in the energy-efficient ver-sion, EA-TDMA consumes less energy than BMA, unless thetraffic load is very low. The contention phase in BMA helps tominimize the idle listening period during the data transmissionphase; however, the contention phase itself consumes a certainamount of energy before each frame transmission. The energyconsumption in the contention phase is paid off at light trafficloads. However, at high traffic loads, this contention phaseturns into an overhead as the probability of data transmissionbecomes almost certain. In the proposed E-BMA protocol, thesource nodes use piggybacking to make the reservation of thecorresponding data slot rather than sending a control messageduring its allocated contention slot, as shown in Fig. 6. UnlikeBMA, in the new protocol, a source node does not make thereservation in the contention slot as soon as the data packetbecomes available. Instead, it waits for one additional frame

duration to see if there is a successive data packet to send.There is a 1-bit field allocated in each data packet header toindicate whether the source node has a successive data packetto send. If a source node has successive data packets to send in anumber of consecutive frames, the reservation is made once forthe initial data packet in its allocated contention slot, and thesuccessive confirmations will be made through piggybacking.Note that piggybacking a control message requires only 1-bitextra space in the data packet, and hence, the additional powerrequired for piggybacking a control message on a data packetis negligible. In E-BMA, the transceiver of the source node isturned off during the contention phase when it has no controlmessage to send, whereas in BMA, the transmitter is keptidle in similar situations. This allows the E-BMA protocol tosave energy both at low and medium traffic. E-MBA is onlyoutperformed by TDMA and EA-TDMA when the traffic loadis extremely high. To achieve energy efficiency, the E-BMAprotocol compromises the latency of data transmission. Eachdata packet has to wait for one additional frame duration beforebeing transmitted to the CH. As there will be few sensor nodesper cluster in the railway-wagon health monitoring system, theframe length will be much shorter, and the latency of E-BMAwill be within the acceptable limit.

Operation of the E-MBA protocol is divided into rounds,and each round is comprised of a setup phase and a steady-state phase. The steady-state phase is comprised of a contentionphase and a data transmission phase. Both cluster formationand CH selection occur in the setup phase. All non-CH nodesreserve the data slots in the contention phase, whereas datatransmission from source nodes to the CH occurs during thedata transmission phase.

Setup Phase: Considering the specific application area andits simplicity, it is assumed that the network consists ofmultiple fixed clusters. In each of the clusters, there isone CH located in the center of the cluster. Based onthe application and cluster size, direct transmission fordata communication between source nodes and the CH isconsidered instead of multihop data transmission. In thesetup phase, the CH informs all nodes about the start of thecurrent round, frame start/stop time, and number of framesin a round.

Contention Phase: Each node is assigned a specific slot inthe contention phase. A node transmits a 1-bit control

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

message during its scheduled slot to reserve a data slotif it has a data packet to transmit; otherwise, the noderemains in sleep mode during that contention slot. Afterthe contention period is completed, the CH sets up andbroadcasts a transmission schedule for the source nodes.However, unlike BMA, the source node does not make thereservation immediately after the data becomes available.Instead, the source node keeps the data packet in the buffer,and it waits for one frame duration to see if there is aconsecutive data packet to send.

Data Transmission Phase: The data transmission phase con-tains one or more frames. The size and duration of eachframe is fixed. Nodes send their data to the CH at mostonce per frame during their allocated time slot. During thedata transmission phase, each source node turns on its radioin its allocated data slot and transmits data to the CH. Ifthere are consecutive packets, the transmitted data packetconveys that information through piggybacking.

After receiving all data from the nodes of a round, data ag-gregation takes place to reduce unwanted data. A considerableamount of energy is saved if the data are locally aggregated inthe CH first rather than when sending the raw data to the BSor central controller and aggregating them in the BS. Then, theresultant data are sent from the CH to the BS using a spreadingcode and a CSMA approach, as used in the LEACH protocol[20]. Once the CH is ready to send the aggregated data, it mustsense the channel to see if anyone else is transmitting usingthe BS spreading code. The CH waits if the channel is busy;otherwise, the CH transmits data to the BS. After a predefinedtime, the system begins the next round, and the whole processis repeated.

Analytical and simulation models were developed based onthe energy model [21], [22] for the TDMA, EA-TDMA, BMA,and E-BMA to compare their performances in terms of trafficload and energy dissipation features, which is presented in thenext section.

V. ANALYTICAL AND SIMULATION MODELING

To analyze the performance of the proposed E-BMA protocoland compare its performance with existing wireless MAC pro-tocols, including EA-TDMA, analytical and simulation modelshave developed.

This proposed protocol is analyzed in a WSN scenario wherethere are one CH and N non-CH nodes in each cluster, assum-ing that there are l frames in a round. The data slot duration isassumed to be Td. Let the probability of a node having data totransmit be p. The power consumption in the transmit mode andthe receive mode are Pt and Pr, respectively. Energy dissipationof idle listening mode is Pi. For simplicity, as stated in [21]and [23], the energy required to turn on the radio by the sourcenodes for transmission or reception is negligible and, hence, isignored in the following analysis.

As per definition, Td is the time required to transmit orreceive a data packet, and it is assumed that Tc is the timerequired to transmit/receive a control packet. The time requiredfor the CH to transmit a control message to all non-CH nodesin BMA is Tch. The time required for a node to switch on,

TABLE INOMENCLATURE

check its buffer, and turn off its radio is in E-TDMA is Te. Theparameters used in the analysis are defined in Table I.

A. Energy Consumption

The energy consumption of the TDMA, EA-TDMA, BMA,and E-BMA protocols based on the energy model in [20] and[22] is modeled as follows.

Energy Consumption of TDMA Protocol: During the con-tention, the CH and all non-CH nodes keep their radioson, and communication takes place between the CH andall non-CH nodes. In this period, the CH assigns data slotsto individual nodes for data transmission and informs allnodes in the cluster. Therefore, energy consumption bythe CH to send a control message is PtTc, and energyconsumption by each node to receive a control message isPrTc. Therefore, the energy consumption in a contentionperiod is given by

Econt = NPrTc + PtTc. (1)

Each node transmits, at most, one packet per frame inter-val. During a frame transmission, energy consumption bya source node is PtTd. The energy consumed by the CHwhile receiving the data packet is PrTd. A nonsource nodeturns on its radio and keeps it idle during its scheduled timeslots. The energy consumed by a nonsource node is PiTd.As the CH also stays in idle mode when there are no datato receive from the non-CH node during a data slot, theenergy consumed by the CH is also PiTd.

In a data slot, a node sends data with probability p andremains idle with probability (1 − p). The expected energyconsumption during a single frame transmission consistingof N data slots is [pPtTd + (1 − p)PiTd + pPrTd + (1 −p)PiTd]N . The expected energy consumption in a trans-mission round is given by

Etrans=[pPtTd+(1−p)PeTe+pPrTd+(1−p)PiTd] lN. (2)

As each round is comprised of l frames, the average energyconsumption per round in the TDMA protocol can be

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAFIULLAH et al.: WIRELESS MAC PROTOCOLS FOR RAILWAY MONITORING APPLICATIONS 7

formulated as

ETDMA=[NPrTc + PtTc]

+ [pPtTd + 2(1 − p)PiTd + pPrTd] lN. (3)

Energy Consumption of EA-TDMA Protocol: Similar to theTDMA protocol, the energy consumption in a contentionperiod is given by

Econt = NPrTc + PtTc. (4)

EA-TDMA differs from TDMA in that every non-CH nodein EA-TDMA wakes up in its allocated slot and checkstransmit buffers. If there are no data to send, it turns off theradio immediately. Hence, the energy consumed by a non-CH node that has no data to transmit is PeTe. The energyPeTe is used to switch on, check the transmit buffers,and then turn off the radio module. The expected energyconsumption in a transmission round is given by

Etrans = [[pPtTd + (1 − p)PeTe]

+ [pPrTd + (1 − p)PiTd]] lN. (5)

As each round is comprised of l frames, the average energyconsumption per round in the EA-TDMA protocol can beformulated as

EEA−TDMA = [NPrTc + PtTc] + [[pPtTd + (1 − p)PeTe]

+ [pPrTd + (1 − p)PiTd]] lN. (6)

Energy Consumption of BMA Protocol: In BMA, there isa contention period in each session when all nodes keeptheir radios on. Each source node transmits a controlmessage during its scheduled slot, as well as its remainsidle (N − 1) slots. Each nonsource node stays idle duringthe contention period. During a contention slot, the CHnode receives control packets when there is a source nodesending a control packet; otherwise, the CH stays idle. Theexpected energy consumption during a contention period isgiven by

Econt=[pPtTc+(1 − p)PiTc+(N − 1)PiTc+PrTch]N

+ [pPrTc + (1 − p)PiTc]N + PtTch. (7)

During a frame transmission, each source node sends thedata packet in its allocated slot, whereas the nonsourcenodes keep their radios turned off. The expected energyconsumption during a frame transmission is given by

Eframe = [pPtTd + pPrTd]N. (8)

The average energy consumption per round in the BMAprotocol can be formulated as

EBMA = [[pPtTc + pPrTc + 2(1 − p)PiTc + (N − 1)PiTc

+PrTch + pPtTd + pPrTd]N + PtTch] l. (9)

Energy Consumption of E-BMA Protocol: A source nodesends a control message in its respective contention slot(unless the reservation is done by the preceding data packet

sent by the same source node in the previous frame) andremains idle in the remaining (N − 1) contention slots.The nonsource nodes keep their radio turned off duringthe entire contention period. A control message cannotbe piggybacked if there is no data packet sent in theprevious frame by the same node. The probability of adata packet not being piggybacked is p(1 − p). If a controlmessage is piggybacked, the source node keeps the radioturned off in the respective contention slot, whereas the CHnode remains in idle listening mode. The expected energyconsumption during a contention period is given by

Econt=[p(1 − p)PtTc + PrTch]N

+ [p(1 − p)PrTc + (1 − p(1 − p))PiTc]N + PtTch. (10)

During a frame transmission, each source node sends thedata packet in its allocated slot, whereas the nonsourcenodes keep their radios turned off. Note that piggybackinga control message only requires 1-bit extra space in the datapacket. Hence, it is assumed that no additional power isrequired for piggybacking. The expected energy consump-tion during a frame transmission is given by

Eframe = [pPtTd + pPrTd]N. (11)

The average energy consumption per round in the E-BMAprotocol can be formulated as

EE−BMA = [[p(1 − p)PtTc + p(1 − p)PrTc

+ (1 − p(1 − p))PiTc + PrTch + pPtTd

+ pPrTd]N + PtTch] l. (12)

B. Transmission Latency

The maximum transmission latency of TDMA and EA-TDMA protocols is given by Tc +NTd as both protocols havesimilar frame structure. The maximum transmission latency ofBMA is Tch + (Tc + Td)N . The maximum transmission la-tency of E-BMA is 2[Tch + 2(Tc + Td)N ] as each data packethas to wait for one additional frame duration before beingtransmitted.

Simulation models have been developed for the TDMA, EA-TDMA, BMA, and E-BMA protocols to verify the correctnessof the analytical models using Java programming languageand SimJava Package version 2.0 [27]. SimJava is a process-oriented discrete-event simulation package developed by theUniversity of Edinburgh. The simulation results represent thegeneral characteristics of the existing and proposed protocols.The simulation for each model was run for 10 000 rounds. Theexpected energy consumption was calculated, averaging energyconsumption over the entire simulation period.

Both analytical and simulation results confirm that E-BMAis more energy efficient than the other three protocols at lowto medium traffic. It is only outperformed by the TDMA andEA-TDMA protocols when the traffic is extremely high.

In the following section, detailed analyses of the resultsare presented, and protocol performances in terms of energydissipation are compared.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 7. Energy dissipation of EA-TDMA, BMA, and TDMA protocols as a function of probability p (N = 10 and l = 2).

VI. RESULTS AND ANALYSIS

This section analyzes the performance of the proposedE-BMA protocol in terms of energy efficiency and transmissionlatency. In addition, the performance of the E-BMA protocolhas been compared with that of the TDMA, EA-TDMA, andBMA protocols in terms of energy dissipation and transmis-sion latency. As aforementioned, the IEEE 802.15.4 standardand ZigBee wireless module are used for the proposed MACprotocol. The ZigBee-enabled 2.4-GHz CC2420 RF transceiver[19] is used for this analytical and simulation analysis. Foranalytical and simulation modeling purposes, it is assumedthat the power consumption is 50 mW for transmitting,54 mW for receiving, and 50 mW for idle listening. Thesepower ratings are comparable with that of the CC2420 RFtransceiver specification. The data rate is 25 kb/s, and thecontrol packet size is 5 bytes. For simplicity, it is assumed thatTe = Td/10 and that Pe = Pi.

A. Energy Consumption

Energy consumption of the protocols has been evaluated andcompared for different parameter settings. For analytical andsimulation analysis, four cases have considered the estimationof the energy dissipation of the protocols being analyzed.Case 1: In this case, the average energy consumption of the

aforementioned four WSN MAC protocols has been de-rived for various transmission probabilities. It is assumedthat the total number of non-CH nodes is N = 10, thenumber of frames is l = 2 per round, and the data packetsize is 100 bytes.

Fig. 7(a) and (b) shows the average energy consump-tion of the TDMA, EA-TDMA, BMA, and E-BMA proto-cols for the transmission probability varying from p=0.1to 1.0. The graphs reveal that the energy consumption ofthe TDMA protocol is almost constant as the difference be-tween transmission power and idle listening power is verysmall. The EA-TDMA protocol consumes less energy forlow to medium traffic, i.e., from p=0.1 to 0.5, whereas it isas good as the TDMA protocol for medium to high traffic,1

1Low data traffic refers to transmission probability, i.e., p<0.3, whereas highdata traffic refers to transmission probability, i.e., p>0.7. Medium traffic refersto the transmission probability in between.

i.e., from p = 0.5 to 1.0. This is because, in EA-TDMA,if a node has no data to send in its allocated slot, thetransceiver is turned off to save energy. The lower thetraffic, the higher the savings. The BMA protocol is com-parable with the EA-TDMA protocol for low traffic, i.e.,p = 0.1; however, it consumes more energy than TDMAand EA-TDMA for medium to high traffic because thecontention phase in BMA consumes a certain amountof energy, depending on the traffic load. At high traf-fic loads, the contention phase turns into an overheadfor BMA.

The proposed E-BMA protocol outperforms all threeprotocols significantly. The E-BMA is only outperformedby TDMA and EA-TDMA when p ≥ 0.8, i.e., when thetraffic load is high. In E-BMA, the transceiver of the sourcenode is turned off during a contention slot when it has nocontrol message to send, whereas in BMA, the transceiveris kept idle in similar situations. This allows the E-BMAprotocol to save energy both at low and medium traffic.However, at high traffic loads, the overhead of the con-tention phase surpasses the savings and that why E-MBAis only outperformed by TDMA and EA-TDMA when thetraffic load is extremely high. It is to be noted that there isa constant difference between the energy consumption ofBMA and that of E-BMA.

Case 2: In this experiment, the average energy consumption ofthe aforementioned four WSN MAC protocols has been de-rived for a different number of non-CH nodes. It is assumedthat the transmission probability is p = 0.4, the number offrames is l = 2, and data packet size is 100 bytes.

Fig. 8(a) and (b) shows the average energy consump-tion of TDMA, EA-TDMA, BMA, and E-BMA protocolsfor the total number of non-CH nodes varying from N = 5to 50. As the traffic load is medium (p = 0.4) in thissituation, the performance of E-BMA is the best, whichis followed by EA-TDMA and TDMA. The BMA protocolhas the maximum energy consumption at medium trafficload. The energy consumption of the BMA protocol dra-matically rises as N increases, signifying the overhead dueto contention. The proposed E-BMA protocol minimizesthis overhead through piggybacking at medium traffic load.There is a moderate increase in the energy consumption inTDMA and E-TDMA as the number of nodes increases

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAFIULLAH et al.: WIRELESS MAC PROTOCOLS FOR RAILWAY MONITORING APPLICATIONS 9

Fig. 8. Energy dissipation of EA-TDMA, BMA, and TDMA protocols as a function of number of nodes in a cluster: N (p = 0.4, l = 10).

Fig. 9. Energy dissipation of EA-TDMA, BMA, and TDMA protocols as a function of number of frames: l (N = 20, p = 0.3).

since the overhead in the contention period is minimal forthese two protocols.

Case 3: In this case, the aforementioned four WSN MACprotocols have been evaluated in terms of average energyconsumption for various numbers of frames per round. It isassumed that the total number of non-CH nodes is N = 10,the transmission probability is p = 0.3, and the data packetsize is 100 bytes.

Fig. 9(a) and (b) shows the average energy consump-tion of TDMA, EA-TDMA, BMA, and E-BMA proto-cols for the number of frames per round changing froml = 2 to 20. As the graphs reveal, for medium traffic andsmall number of nodes, E-BMA performs the best amongthese four WSN MAC protocols. In this case, the energyconsumption of TDMA is the highest, whereas the energyconsumption of EA-TDMA and BMA is in between. Sincethe number of nodes is small, the contention overhead inBMA moderately increases, and hence, its energy con-sumption is slightly lower than EA-TDMA. The proposedE-BMA protocol has the least contention overhead due topiggybacking.

Case 4: In this experiment, the impact of data packet size on theoverall energy dissipation has been measured. It is assumedthat the total number of nodes is N = 10, the transmissionprobability is p = 0.4, and the number of frames is l = 2per round.

In Fig. 10(a) and (b), it is evident that the E-BMAprotocol is the most energy-conservative protocol amongthe four protocols, whereas TDMA is the most energy-consuming protocol. However, when the data packet size isless than 50 bytes, the energy dissipation of EA-TDMA issimilar to that of E-BMA. This is because the overhead ofthe wakeup period in the EA-TDMA protocol diminisheswith the reduction in packet size. Although the energyconsumption of BMA is lower than that of TDMA andEA-TDMA, it is as worse as TDMA for packet sizes lessthan 50 bytes. The reason is the energy wastage due to idlelistening in the data transmission period of TDMA exceedsthe contention overhead of BMA when the packet sizeis small.

B. Transmission Latency

The maximum transmission latency of the TDMA, EA-TDMA, BMA, and E-BMA is presented for different numbersof nodes and packet sizes.

Fig. 11(a) demonstrates that the maximum transmission la-tency in all four protocols increases with the number of nodes,as the length of a frame is directly related to the number ofnodes. The packet transmission latency of TDMA and EA-TDMA is the lowest among all four protocols. Due to the exis-tence of the contention period in each frame, the transmission

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 10. Energy dissipation of EA-TDMA, BMA, and TDMA protocols as a function of data packet size (N = 20, l = 10, p = 0.4).

Fig. 11. Transmission latency of EA-TDMA, BMA, and TDMA protocols for (a) different number of nodes and (b) different data packet sizes.

latency of BMA is slightly higher. The transmission latency ofE-BMA is twice that of BMA as each packet has to wait forone additional frame duration in E-BMA. Fig. 11(b) demon-strates that the maximum transmission latency in all protocolsincreases with the data packet size, as the length of a frame isdirectly related to the data packet size. Similar to the previouscase, the packet transmission latency of TDMA and EA-TDMAis the lowest. The transmission latency of BMA is slightlyhigher, and the transmission latency of E-BMA is twice thatof BMA.

Summarizing the analytical and simulation results, the fol-lowing can be concluded.

• The E-BMA protocol is the most energy-efficient protocol,particularly in the case of low and medium traffic applica-tions. However, at extremely high traffic conditions, theEA-TDMA protocol performs better.

• The E-BMA protocol is more energy efficient than theother three protocols for any number of sensor nodes ina cluster when the traffic load is medium.

• The E-BMA protocol dissipates less energy than the otherthree protocols, regardless of the number of frames perround for medium traffic.

• The performance of the E-BMA protocol is superior tothe other three protocols when data packet size is equal toor greater than 50 bytes. For small data packet size (lessthan 50 bytes), the energy dissipation of EA-TDMA iscomparable with E-BMA.

• Although the transmission latency of E-BMA is higherthan other protocols, it will not impact the system

performance significantly when the number of nodesis small.

VII. CONCLUSION

The performance of rail vehicles running on railway tracks isgoverned by the dynamic behaviors of railway wagons, partic-ularly in the cases of lateral instability and track irregularities.In this paper, considering the traffic conditions of the intendedapplication, an energy-efficient WSN MAC protocol has beeninvestigated to monitor typical dynamic behavior of railwaywagons. Simulation and mathematical models have been devel-oped for the proposed E-BMA protocol, and its performancehas been compared with the EA-TDMA, TDMA, and BMAprotocols in terms of energy efficiency.

Analytical and simulation results show that the E-BMA andEA-TDMA protocols outperform both the TDMA and BMAprotocols for all traffic conditions. The results revealed thatthe E-BMA protocol outperformed other protocols for low tomedium traffic, whereas the EA-TDMA protocol outperformedthe TDMA and BMA protocols for medium to high traffic.The E-BMA protocol is only outperformed by EA-TDMA andTDMA protocols for high traffic.

ACKNOWLEDGMENT

The authors would like to thank the reviewers for theirvaluable and informative suggestions that improved the qualityof this paper.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAFIULLAH et al.: WIRELESS MAC PROTOCOLS FOR RAILWAY MONITORING APPLICATIONS 11

REFERENCES

[1] G. M. Shafiullah, A. Thompson, P. Wolfs, and S. Ali, “Energy-efficientTDMA MAC protocol for wireless sensor networks applications,” in Proc.5th ICECE, Dhaka, Bangladesh, Dec. 24–27, 2008, pp. 85–90.

[2] J. Smith, S. Russel, and M. Looi, “Security as a safety issue in railcommunications,” in Proc. 8th Aust. Workshop SCS, Canberra, Australia,2003, pp. 79–88.

[3] V. K. Garg and R. V. Dukkipati, Dynamics of Railway Vehicle Systems.New York: Academic, 1984.

[4] S. S. Bleakley, “Time frequency analysis of railway wagon body acceler-ations for a low-power autonomous device,” M.S. thesis, Fac. Eng. Phys.Syst., Central Queensland Univ., Rockhampton, Australia, Oct. 2006.

[5] G. M. Shafiullah, A. Thompson, P. Wolfs, and S. Ali, “Predicting verticalacceleration of railway wagons using regression algorithms,” IEEE Trans.Intell. Transp. Syst., vol. 11, no. 2, pp. 290–299, Jun. 2010.

[6] Y. Sato and M. Miwa, “Measurement and analysis of track irregularity onsuper-high speed train—TRIPS,” Int. J. Heavy Veh. Syst., vol. 7, no. 1,pp. 22–33, May 2000.

[7] C. W. Tsang and T. K. Ho, “Optimal track access rights allocation foragent negotiation in an open railway market,” IEEE Trans. Intell. Transp.Syst., vol. 9, no. 1, pp. 68–82, Mar. 2008.

[8] M. Seifert, “Using modern technology to enhance rail security,” in Proc.Conf. Railway Eng., Melbourne, Australia, Apr./May 2006, pp. 430–436.

[9] C. P. E. Aboelela, W. Edberg, and V. Vokkarane, “Wireless sensor networkbased model for secure railway operations,” in Proc. 25th IEEE Perform.,Comput. Commun. Conf., Apr. 2006, p. 6.

[10] S. M. S. L. Cheekiralla, “Development of a wireless sensor unit fortunnel monitoring,” M.S. thesis, Civil Environ. Eng., Mass. Inst. Technol.,Cambridge, MA, Feb. 2004.

[11] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “A surveyof sensor networks,” IEEE Commun. Mag., vol. 40, no. 8, pp. 102–114,Aug. 2002.

[12] P. J. Wolfs, S. Bleakley, S. T. Senini, and P. Thomas, “An autonomous,low cost, distributed method for observing vehicle track interactions,” inProc. Rail Conf., Atlanta, GA, Apr. 2006, pp. 279–286.

[13] P. Wolfs, S. Bleakley, S. Seninin, and P. Thomas, “A distributed lowcost device for the remote observation of track and vehicle interactions,”in Proc. Conf. Railway Eng., RTSA, Melbourne, Australia, Apr. 2006,pp. 280–286.

[14] ADXL202E-Low-Cost-2 g Dual-Axis Accelerometer With Duty CycleOutput—Analog Devices, Analogue Devices. [Online]. Available: http://www.alldatasheet.com/datasheet-pdf/pdf/48922/AD/ADXL202E.html

[15] D. S. S. Rhee and S. Liu, “Techniques for minimizing power consumptionin low data-rate wireless sensor networks,” in Proc. Wireless Commun.Netw. Conf., Mar. 2004, pp. 1727–1731.

[16] “IEEE P802.15 Working Group for Wireless Personal Area Networks(WPANs),” IEEE Comput. Soc., New York, Jul. 2000, IEEE Tech. Rep.

[17] “IEEE Standard 15.4: Wireless Medium Access Control (MAC) and Phys-ical Layer (PHY) Specifications for Low-Rate Wireless Personal AreaNetworks (LR-WPAN),” IEEE Comput. Soc., New York, 2003, IEEETech. Rep.

[18] Low-Power RF System-on-Chip for ZigBee Applications: CC2431.[19] “An IEEE 802.15.4 complaint and ZigBee-ready 2.4 GHz RF transceiver,”

Microw. J., vol. 47, no. 6, pp. 130–135, Jun. 2004.[20] B. Bougard, F. Catthoor, D. Daly, A. Chandrakasan, and W. Dehaene,

“Energy efficiency of the IEEE 802.15.4 Standard in dense wireless mi-crosensor networks: Modeling and improvement perspectives,” in Proc.Des., Autom. Test Europe Conf. Exhib., 2005, pp. 196–201.

[21] W. Ye, J. Heidemann, and D. Estrin, “An energy-efficient MAC protocolfor wireless sensor networks,” in Proc. IEEE INFOCOM, New York,Jun. 2002, pp. 1567–1576.

[22] V. Raghunathan, C. Schurgers, S. Park, and M. B. Srivastava, “Energy-aware wireless microsensor networks,” IEEE Signal Process. Mag.,vol. 19, no. 2, pp. 40–50, Mar. 2002.

[23] W. B. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan, “Anapplication-specific protocol architecture for wireless microsensor net-works,” IEEE Wireless Commun. Trans., vol. 1, no. 4, pp. 660–670,Oct. 2002.

[24] G. M. Shafiullah, A. Gyasi-Agyei, and P. Wolfs, “A survey of energy-efficient and QoS-aware routing protocols for wireless sensor networks,”in Proc. Int. Joint Conf. CISSE, Dec. 2007, pp. 352–357.

[25] J. Li, “A bit-map assisted energy-efficient MAC scheme for wireless sen-sor networks,” M.S. thesis, Elect. Eng., Mississippi State Univ., Starkville,MS, May 2004.

[26] J. Li and G. Y. Lazarou, “A bit-map-assisted energy-efficient MACscheme for Wireless Sensor Networks,” in Proc. 3rd Int. Symp. IPSN,Berkeley, CA, Apr. 2004, pp. 55–60.

[27] Simjava Ver. 2.0.

G. M. Shafiullah (M’12) received the B.Sc.Eng. degree in electrical and electronics engineer-ing from Chittagong University of Engineeringand Technology, Chittagong, Bangladesh, and theM. Eng. degree from Central Queensland University(CQUniversity), Rockhampton, Australia. He is cur-rently working toward the Ph.D. degree with theSchool of Engineering and Built Environment,CQUniversity.

He is the author of 25 referred book chapters,journal papers, and conference papers. His research

interests include networking and data communication, sensor networking,power engineering, renewable energy, smart grid technology, and computer-aided technology.

Salahuddin A. Azad received the B.Sc. Eng. de-gree in computer science and engineering fromBangladesh University of Engineering and Tech-nology, Dhaka, Bangladesh, in 1999 and the Ph.D.degree in information and technology from MonashUniversity, Clayton, Australia, in 2007.

He is currently a Postdoctoral Research Fel-low with the Power Engineering Group, CentralQueensland University, Rockhampton, Australia. Heis the author of 17 refereed conference papers, IEEEjournal papers, and book chapters. His major re-

search interests include renewable energy, smart grid, image processing, ma-chine learning, data mining, and network security.

A. B. M. Shawkat Ali (SM’10) received the Ph.D.degree in information technology from Monash Uni-versity, Clayton, Australia.

He is currently with the School of Information andCommunication Technology, Central QueenslandUniversity, Rockhampton, Australia. In particular,he is currently leading a research group on com-putational intelligence. He is the author of morethan 100 research papers in international journalsand conferences, as well as several book chaptersand books. His research interests include computa-

tional intelligence, data mining, smart grids, cloud computing, and biomedicalengineering.

Dr. Ali is currently the Editor-in-Chief for the International Journal ofEmerging Technologies in Sciences and Engineering.