Transcription

1 Ths materal s publshed n the open archve of Md Sweden Unversty DIVA to ensure tmely dssemnaton of scholarly and techncal work. Copyrght and all rghts theren are retaned by authors or by other copyrght holders. All persons copyng ths nformaton are expected to adhere to the terms and constrants nvoked by each author's copyrght. In most cases, these works may not be reposted wthout the explct permsson of the copyrght holder. Zheng, T.; Gdlund, M.; Åkerberg, J., "WrArb: A New MAC Protocol for Tme Crtcal Industral Wreless Sensor Network Applcatons," n IEEE Sensors Journal, vol. 16, no. 7, pp , Aprl1, IEEE. Personal use of ths materal s permtted. However, permsson to reprnt/republsh ths materal for advertsng or promotonal purposes or for creatng new collectve works for resale or redstrbuton to servers or lsts, or to reuse any copyrghted component of ths work n other works must be obtaned from the IEEE.

2 IEEE SENSORS JOURNAL, VOL. 16, NO. 7, APRIL 1, WrArb: A New MAC Protocol for Tme Crtcal Industral Wreless Sensor Network Applcatons Tao Zheng, Member, IEEE, Mkael Gdlund, Member, IEEE, and Johan Åkerberg, Senor Member, IEEE Abstract Wreless sensor networks are typcally desgned for condton montorng applcatons and to conserve energy but not for tme-crtcal applcatons wth strct real-tme constrants that can be found n the ndustral automaton and avoncs doman. In ths paper, we propose a novel medum access control MAC) protocol defned as wreless arbtraton WrArb) whch grants each user channel access based on ther dfferent prorty levels. The proposed MAC protocol supports multple users and each user s pre-assgned a specfc arbtraton frequency whch decdes the order of channel access. Wth ths mechansm, we can ensure that the user wth the hghest prorty wll mmedately gan channel access and we can guarantee a determnstc behavor. To evaluate the proposed MAC, we use a dscrete-tme Markov chan model to mathematcally formulate the WrArb protocol. Our results show that the proposed protocol provdes hgh performance to ensure determnstc real-tme communcaton and bandwdth effcency. Index Terms Wreless arbtraton, wreless sensor networks, determnstc, real-tme, cross-layer. I. INTRODUCTION WIRELESS technologes have become ncreasngly popular for emergng applcatons targetng the consumer market and the ndustral automaton doman [1]. Mergng wreless communcaton and real-tme systems s a non-trval task, especally for ndustral applcatons where the sensors and actuators are part of control loops, and predctable network performance n terms of message transfer delay and relablty s requred [2]. Ths paper wll address the problem of accessng the wreless channel and provdng tmelness guarantees. In wred networks, data packets can be effcently scheduled usng the Controller Area Network CAN) bus, whch has already been proven to be useful n ndustry. The medum access control MAC) protocol n CAN s collson-free and uses a prorty mechansm makng t possble to schedule the bus f message characterstcs transmsson tmes, jtter, etc.) Manuscrpt receved July 10, 2015; revsed September 30, 2015; accepted November 23, Date of publcaton December 2, 2015; date of current verson February 10, Ths work was supported n part by the Natonal Key Basc Research Program of Chna under Grant 2013CB and n part by the Fundamental Research Funds for the Central Unverstes under Grant 2015JBM001. The assocate edtor coordnatng the revew of ths paper and approvng t for publcaton was Dr. Ntagour P. Mahalk. T. Zheng s wth the School of Electronc and Informaton Engneerng, Bejng Jaotong Unversty, Bejng , Chna e-mal: t.zheng@ eee.org). M. Gdlund s wth Md Sweden Unversty, Sundsvall , Sweden e-mal: mkael.gdlund@mun.se). J. Åkerberg s wth ABB Corporate Research, Forskargränd 7, , Västerås, Sweden e-mal: johan.akerberg@se.abb.com). Dgtal Object Identfer /JSEN are known, whle also makng t possble to compute the upper bound on message delay. The CAN protocol belongs to the famly of domnance or bnary countdown protocols [3]. For wreless networks, the Carrer Sense Multple Access wth Collson Avodance CSMA/CA) has manly been used as a collson-free soluton to compete for channel access. CSMA/CA s used as default meda access mechansm n wreless local area networks WLAN) based on the IEEE standard and n wreless sensor networks WSNs) based on the IEEE standard. However, for tme crtcal applcatons whch requre determnstc communcaton and have strct deadlnes, t has been shown that the CSMA/CA protocol s not sutable due to unpredctable tme delay beng generated by random dstrbuton of the backoff tme [4]. CSMA/CA typcally ensures mnmum delay for low traffc loads, but as traffc load ncreases, the delay becomes unacceptable and throughput deterorates. Snce CSMA/CA s a random access scheme, t does not prortze transmssons based on the physcal processes montored by the sensors and actuators [5], [6]. In addton to the above slotted contenton-based CSMA/CA, a recent survey revews and classfes asynchronous realtme MAC protocols n [7]. Compared wth synchronous contenton-based CSMA/CA, reduced real-tme performance could be nduced as t s necessary for the asynchronous random access mechansm to adopt certan strateges to overcome the decoupled stuaton between the transmtter and the recever, whch wll ncrease transmsson delay due to overhearng, over-emttng and even packet collsons. To mprove the relablty of ndustral wreless networks, current avalable standards such as WrelessHART [8], [9], ISA a [8], WIA-PA [10] and IEEE e [11], uses a Tme Dvson Multple Access TDMA) protocol combned wth CSMA/CA to schedule the user channel access. One drawback wth usng TDMA n msson crtcal applcatons s that n the event of an emergency event wth the hghest prorty, the transmsson of the crtcal data packet needs to wat for ts transmsson tme slot, whch s unacceptable for applcatons wth strct deadlnes. Another drawback wth TDMA-based WSNs for crtcal applcatons s that f a slot for emergency messages had to be reserved n every frame, the channel utlzaton wll be reduced, especally f the refresh rate s hgh. Ths s one of the man reasons why TDMA-based WSNs are not a good choce for tme-crtcal wreless sensor network applcatons. Ths s hghlghted n the lterature [4], although a seres of real-tme MAC protocols are desgned to facltate IEEE. Personal use s permtted, but republcaton/redstrbuton requres IEEE permsson. See for more nformaton.

3 2128 IEEE SENSORS JOURNAL, VOL. 16, NO. 7, APRIL 1, 2016 the expected transmsson latency for WSNs, they only manage to meet the real-tme requrement by attemptng to reduce the data processng tme between transmtter and recever rather than mmedately reactng to dfferent emergency events. Regardless of whether CSMA/CA or TDMA nduces hgher latency by avodng addtonal channel conflct and contenton, nether s sutable for the real-tme control system where dfferent trggered events are lmted wthn dfferent hard tme-bounds. In ths paper, we address the problem of wreless channel access by proposng WrArb whch has smlar functons as the CAN bus. The proposal combnes the physcal layer and MAC layer to guarantee real-tme performance. Before the perod of data transmsson, channel access arbtraton wll be performed frst. Each user n the network s pre-assgned a dedcated arbtraton frequency wth the purpose to dentfy channel access prortes whch s utlzed to determnstcally determne the channel access order. Ths procedure wll ensure that n each arbtraton cycle the most crtcal event can be the only one to obtan the hghest prorty, and mmedately gan access to the wreless medum. Through the proposed scheme we can guarantee real-tme performance, where all messages wll meet ther ndvdual deadlnes, for tme crtcal ndustral wreless applcatons. We provde a stochastc theoretcal model that focuses on the performance evaluaton of the WrArb scheme by employng a dscrete tme Markov chan DTMC) analyss. As formulated analytcal expressons for the theoretcal analyss, t can be used to calculate the performance n terms of system throughput and communcaton delay. Furthermore, we compare the channel utlzaton between WrArb and TDMA-based WSNs such as WrelessHART and GnMAC for tme-crtcal applcatons for example safety) that requre hgh refresh rate. The obtaned results show that the proposed soluton outperforms TDMA-based WSNs n terms of latency, throughput and channel utlzaton. The remander of the paper s organzed as follows. Secton II dscusses related work and n Secton III we descrbe the medum access mechansm for the WrArb n detal. In Secton IV, we model WrArb nto a dscrete tme Markov chan. In Secton V, we valdate system models from worst-case scenaros, whch s followed by an analyss of performance evaluatons wth regard to throughput, delay and channel utlzaton. Fnally, the results of the research are summarzed and we propose future work on WrArb. II. RELATED WORK In order to serve the strngent real-tme requrements for tme-crtcal event-based wreless applcatons, and overcome mentoned drawbacks on nondetermnstc tme delay as well as channel utlzaton, several MAC protocols supportng real-tme have prevously been presented [12]. Arbtrary contenton s the man reason for the transmsson delay caused by random backoff and retransmssons. Thus, CSMA/CA-based technologes are used to reduce the harmful mpacts from unexpected data collson. In [13], Ye et al. presented the SMAC protocol for wreless sensor networks where the sensor nodes utlze the benefts of collson avodance of the RTS/CTS handshake mechansm. However, because of the schedulng of the fxed duty cycle, ths wll cause an unacceptable delay whch makes ths MAC protocol unsutable for tme-crtcal applcatons, especally for applcatons n ndustral automaton. There are smlar protocols based on SMAC n the lterature, such as SMAC-AL [14] and DSMAC [15], whch replace an adaptve duty cycle to reduce the forwardng delay. However, ths class of protocols nvolves end-to-end delay due to the sleep perod to avod collson wth others. In order to reduce the delay from unnecessary sleep, a wake-up schedulng method such as LEEMAC [16], DMAC [17], DW-MAC [18] or SPEED-MAC [19] s developed to decrease sleepng delay. Although the aforementoned protocols can reduce the delvery delay by schedulng the sleep slot, they are not desgned for determnstc delay guarantee due to random backoff. All these MAC protocols ncludng but not lmted to [13] [19]) nvolves the CSMA/CA mechansm beng more valuable to protect ordnary wreless sensor networks from packet collsons rather than tme-crtcal Wreless Sensor and Actuator Networks WSAN) wth hard requrements for determnstc delay guarantees. HyMAC [20] s a class of MAC protocols mergng TDMA and FDMA together. Although HyMAC guarantees a certan end-to-end delay, the man drawback s the lack of ablty to adapt to the harsh wreless channel condtons of ndustral automaton or avoncs. In [21], Suryacha et al. presented the GnMAC protocol whch they clam can gve support for real-tme guarantees for tme-crtcal applcatons n ndustral wreless sensor networks. However, ther approach to TDMA schedulng wll result n exclusve tme slot usage whch may prevent an emergency task from sendng data mmedately because t needs to wat for transmsson slots. III. WIRELESS ARBITRATION MEDIUM ACCESS PROTOCOL A. System Model The centralzed control model s commonly used n ndustral applcatons. Perpheral networked devces form a complete system where dfferent performance levels are requred accordng to the ndustral process needs. Therefore, the star network s frstly consdered n ths paper snce t s a typcal centralzed network topology. It s shown n Fg. 1 whch conssts of the gateway as the central controller and several network users whch nclude dfferent knds of ndustral devces, such as sensor nodes and actuators. The WrArb medum access protocol uses an arbtraton cycle also called the arbtraton phase). An arbtraton phase conssts of two parts, whch are arbtraton decson perod and arbtraton executon perod. The arbtraton decson perod s the frst step n an arbtraton phase. Ths s to process channel access requests and determne a determnstc channel access order. Then, the arbtraton executon perod occurs when the actual data s transmtted. Specfcally, dependng on ts prorty, the user wth the hghest prorty can mmedately access a channel to transfer data, whle the remanng users wth lower prorty cannot access the channel. The percent that an arbtraton phase spends n the decson perod s a

4 ZHENG et al.: A NEW MAC PROTOCOL FOR TIME CRITICAL INDUSTRIAL WSN APPLICATIONS 2129 Fg. 2. The flow dagram for the WrArb. Fg. 1. The network consdered for the WrArb and the meda access control mechansm. fracton of the total arbtraton cycle, and should be small enough to ensure that the devce wth the hghest prorty wll have enough tme to complete ts data transmsson. Ths can be acheved because t s possble for a devce to use a sgnal short enough to delver the arbtraton request to the gateway. We defne a repeatng perod T s as the fxed arbtraton nterval. All the network users need to send ther channel access request sgnals to the gateway before data transmsson. Therefore, the users should support the followng functons: synchronzaton wth the gateway; transmsson of channel access request sgnals before attemptng to access the channel; transmsson of data packets. The gateway handles two key functons ncludng network synchronzaton and channel access request arbtraton. In order to let the gateway dentfy dfferent channel access request sgnals, some new research works on PHY layer should be carred out for both the gateway s recever and the network user s transmtter. In ths paper, we let the Arbtraton Frequency AF) to represent event prorty, whch s desgned wth equvalence to the subcarrer frequency. Each user n, N ={1, 2,...,N} s only preassgned one subcarrer frequency f, whch mples that the desgnated arbtraton frequences should reman orthogonal wth each other,.e., { f f j =,, j N and = j}. Regardng the gateway, because several channel access request sgnals modulated over dfferent subcarrers arrve at ts recever randomly, the gateway should have the capablty to contnuously keep sensng the arbtraton sgnals from all subcarrers n a random manner. Therefore, a new recever model s ntroduced brefly snce PHY layer modelng s not the focus of ths paper. The new recever of the gateway should consst of a pluralty of modules whch ncludes a bandpass flter, a sampler, a Fourer transformer, as well as a comparson and decson block. Specfcally, the bandpass flter s used to allow channel access request sgnals wthn a selected range of arbtraton frequences to be sensed and decoded, whle preventng nterferng sgnals at unwanted frequences from gettng through. After spectral samplng of the fltered sgnals, these channel access request sgnals captured by the recever are stll mxed. Therefore, the Fast Fourer Transform FFT) s used to extract the sgnals nformaton of dfferent frequences. After FFT processng, the peak magntude values over dfferent frequences are gven. In order to elmnate nterferences and noses from the harsh ndustral envronment, an arbtraton decson threshold Y arb s predefned to compare wth the ampltudes of the FFT sgnals over dfferent arbtraton frequences wth purposed to accurately dentfy the users. Fnally, accordng to the preassgned subcarrer frequences, a channel access order can be arbtrated. A seres of nteractve steps are executed n each arbtraton cycle see Fg. 1 for 1,,4). For easer mplementaton, WrArb defnes a beacon-enabled channel access mode where perods can be announced n the beacon ssued by the gateway. B. Arbtraton Decson Process One of the core parts of WrArb s the arbtraton decson process n Fg. 2. Four varables are ntroduced n the arbtraton process: 1) AF;2) AF); 3)NB;4)S. AF represents the orthogonal arbtraton frequency whch s pre-assgned to a dedcated user for transmttng ts data; AF) represents the subcarrer frequency set whch s dynamcally updated n the gateway. Ths set s used to save arbtraton frequences dentfed from the comparson block by the gateway n each arbtraton cycle. After the fnal arbtraton decson, ths set wll be represented as a queue, where all the remanng users wth lower prortes are queung n the ascendng order of arbtraton frequences accordng to ther prortes; NB represents the number of delayed arbtraton cycles before the user tred to access the channel for ts data transmsson, whch s defned as the watng stage; S represents the On-Off state of the user n each wreless arbtraton cycle.

5 2130 IEEE SENSORS JOURNAL, VOL. 16, NO. 7, APRIL 1, 2016 In the ntalzaton stage when random users jon a network, AF, NB and S respectvely are defaulted to f,0and0, where {1, 2,...,N} represents the numeral ID of the preassgned arbtraton frequency. AF) stored n the gateway that s used to represent a queue ncludng all users n the prevous arbtraton phase. The boundary of the next perod of the arbtraton phase s assgned n ths ntalzaton stage marked as step 1). After ntalzaton, users begn to send ther channel access request sgnals also called the arbtraton sgnals) to the gateway. Before arbtraton sgnals recevng tme expres, the gateway contnually scans and receves ncomng arbtraton sgnals. If the receved sgnal spectrum ampltude a Y [] over the arbtraton frequency f s not smaller than the predefned arbtraton threshold Y arb stored n the gateway, the user s actvaton status for data transmsson can be detected by the gateway n step 2. Step 3 creates a strategc decson based on the prorty whch determnes how the wreless channels should be scheduled wthout contenton. In each arbtraton cycle, f an actvated user n ntends to jon the network and prepares to transmt ts data, the MAC layer of the gateway wll ncrement the frequency set AF) wth a new arbtraton frequency f, whch corresponds to ths actvated user, and the cardnalty of frequency set AF) s ncreased by 1. We defne Mn AF) as the mnmum element of the frequency set AF) whch could equal to f. If the arbtraton frequency f s the mnmum of the frequency set AF), data transmsson wll start n the followng tme perod. Ths s because the wreless channel must be allocated to the user wth the hghest prorty only; otherwse, NB wll be ncremented by an nteger AF), where we defne the cardnalty of the subset AF) as AF), wth the purpose of ensurng that the user carryng the most tmecrtcal event wll access a channel mmedately and complete ts data transmsson wthn the requred hard tme-bound. As a subset of the frequency set AF), AF) means that all ncluded members are smaller than arbtraton frequency f. As a result, a user wth lower prorty wll have to wat at least AF) watng stages untl NB s decreased to 0 marked as step 4). If there s no other user wth hgher prorty to access the channel n the next arbtraton phase, user n can access the channel mmedately; otherwse, step 4 wll be repeated. Durng the perods of watng, users wth lower prorty should reman n off state to save energy consumpton. As an example, we consder a network havng three users n A, n B, n C whch are granted arbtraton frequency f A, f B, f C, respectvely. For smplcty, arbtraton frequences are n ascendng order. In the current arbtraton phase [kt s,k + 1)T s ], after tme-synchronzaton, the gateway captures three arbtraton sgnals x A t), x B t), x C t) whch are respectvely modulated on the sub-carrer frequency f A, f B, f C ; after arbtraton comparson and decson, output 1 s granted to user n A, whle the remanng two users get output 0. The fnal channel access s thereby gven to user n A frst, snce two necessary condtons are satsfed n that 1) the spectrum ampltude of the receved arbtraton sgnal from user n A s hgher than the predefned arbtraton level as well as 2) arbtraton frequency f A s the lowest. Once user n A completes data transmsson n the current arbtraton phase, the channel resource wll be mmedately released. Followng ths, the next arbtraton phase [k +1)T s,k +2)T s ] wll start. If no new actve user wth lower arbtraton frequency than that of user n B apples for the channel, or f there are no other users n queue before user n B, the second channel access wll be assgned to user n B because the arbtraton frequency of user n B s the lowest of all the current partcpants. Next, the channel control passes from n B to n C, and then to the next actve user wth the hghest arbtraton frequency. IV. MATHEMATICAL FORMULATION In order to evaluate the performance of WrArb we model the stochastc behavor of each devce as a dscrete tme Markov chan DTMC). We assume that the probablty to start sensng the wreless channel s constant and ndependent across all users. A. Stochastc Markov Decson Process Consder that a sequence of random state varables X t wth the tme epoch t s assumed to ntegrate a dscrete tme state space S ={X t, t = 0, 1, 2,...}, the Markov property should be satsfed as follows PX t = x t X 0 = x 0, X 1 = x 1,...,X t 1 = x t 1 ) = PX t = x t X t 1 = x t 1 ) 1) where x k S for all k ={0, 1, 2,...,t} are realzed states of the stochastc process. Ths condton can be descrbed n words as the future state only depends on the present state, and s ndependent on the full hstory state n the dscrete tme stochastc process {X t, t = 0, 1, 2,...}. Here, we assume that X t 1, t 1 s denoted as the current state, the future state followng tme t 1 s lsted n set {X t, X t+1,...}, whle the past state s collected n set {X 0,...,X t 3, X t 2 }.Ifthevalue of the current state x t 1 s known, then the evoluton value of the future state x t only reles upon x t 1, whch ndcates that t s stochastcally ndependent of the known value of the past state {x 0,...,x t 3, x t 2 }. The ntalzaton state of any user ntendng to jon the network can be constructed by usng a two-state Markov decson process whch s of the form {s t1, s t2 }. Power control s assumed to be used n order to mnmze energy consumpton. If the wreless channel s not dle n the present tme slot, users need to reman n sleep mode by temporarly preventng transmsson before the next round of attemptng to access the wreless channel. Hence, n each ntalzaton perod, the user s n ether off state or on state whch represents the nactve or sleep) and actve or awake) state, respectvely. At each decson epoch nstant tmes), the off state s dentfed by default as the last hstorcal state of the on state. For ease of descrpton, the off and on states are marked as 1 and1 respectvely. s t1 { 1} represents the nactve state at tme t 1 and s t2 {1} represents the actve state at tme t 2. Fg. 3 shows the symbolc representaton of the two-state Markov decson process and possble transtons for any newly actvated user n each decson epoch. In state s t2,the decson maker selects an acton a t2,t 1 provdes the user wth a state change after an event trgger, whch refers to the user

6 ZHENG et al.: A NEW MAC PROTOCOL FOR TIME CRITICAL INDUSTRIAL WSN APPLICATIONS 2131 Fg. 3. Symbolc representaton of the two-state Markov decson process. Fg. 4. Dscrete tme Markov chan model for the WrArb. beng n state s t1 wth probablty α at the next tme slot; f nstead an acton a t2,t 2 s chosen n state s t2, the user mantans ths orgnal state wth probablty ᾱ at the next tme slot. In state s t1, the decson maker can lkewse choose ether acton a t1,t 2 or acton a t1,t 1. Choosng acton a t1,t 2 n state s t1 makes the user move to state s t2 wth probablty at the next tme slot; state s t1 remans unchanged wth probablty, f acton a t1,t 1 s chosen at the next tme slot. B. Modelng Dscrete Tme Markov Chan We assume a network of a fxed number N of users and each user always has a packet avalable for transmsson. In ths way we can model the behavor of users usng the DTMC shown n Fg. 4. The DTMC descrbes the state transton dagram of dfferent users, where { f t), w t), s t)} s the three-dmensonal state. We use a dscrete and nteger tme scale where t and t + 1 correspond to the startng pont of two consecutve tme slots. Let f t) be the arbtraton frequency assgned to the user n at tme t and w t) s the stochastc process representng the watng stage of the user n at tme t. Ifm s defned as the maxmum watng stage, regardng the user ownng arbtraton frequency f, t s reasonable to estmate that the value of the watng stage should be less than or equal to the result of sequence number mnus 1. Because the user wth arbtraton frequency f +1 wll not obtan a hgher prorty from the gateway to access the channel than the user wth arbtraton frequency f. If user n and user n +1 both have a data packet to transfer n the same arbtraton phase, user n wll send the data frst due to ts hgher prorty. Therefore, w t) falls nto the set [0, 1, 2,..., 1], = 0, 1, 2,...,N. Let s t) be the stochastc process representng the actve-nactve state of user n at tme t. Next, as a result of any user changes to the state of the data transmsson, l [1, L] represents data transmsson phase, where L s the maxmum packet transmsson duraton measured n the nteger number of tme slots that are not allowed to be nterrupted. The dle state { f, 0, 1} stands for the nactve perod when the user n has no data to transfer. The statonary dstrbuton of the DTMC can be descrbed as b{ f,w,s} = lm P{ f t) =,w t) = w, s t) = s}, t [1, N], w [0, 1], s { 1, 0} 2) Gven the DTMC n Fg. 4 the one-step transton probabltes can be wrtten as b{ f, 0, 1 f, 0, 1} =, [1, N] b{ f, 2, 1 f, 1, 1} =, [3, N] b{ f, 0, 1 f, 1, 1} =, [2, N] b{ f, L, 0 f, 0, 1} =ρ, 3) The frst equaton n 3) descrbes the probablty of startng a new data transmsson request at the begnnng of each arbtraton phase. It represents each arbtraton user that transfers from nactve state to actve state. The second equaton descrbes a stuaton where the user wth a smaller arbtraton frequency needs to wat for a number of cycles of arbtraton phase to allow the hgh-prorty users to transfer data frst, meanwhle the correspondng watng stage s reduced one by one. The thrd equaton states the probablty of transtng from the last watng stage to the next arbtraton phase snce the counter for the watng stage s cleared. The last equaton mples that snce the data transmsson request from user n has been detected n the earler arbtraton phase, the gateway pre-assgned a prorty to ths user; f no hgherprorty users are detected by the gateway n the current arbtraton phase, user n obtans the hghest prorty; we defne the followng data transmsson perod wll be used by user n wth probablty ρ. We defne the DTMC s m-step transton probablty as the probablty of transtng from state { f, 0, 1} to state { f, m, 1} [2, N], m [1, 1]) nm steps as b m) = P{ f, m, 1 f, 0, 1}, [2, N], m [1, 1] 4) then, the state transton probablty matrx for the DTMC can be expressed as b 1) b 1) 3 b 2) b 1) 4 b 2) 4 b 3) b 1) N 2 b 2) N 2 b 3) N b 1) N 1 b 2) N 1 b 3) N 1 b N 2) N 1 0 b 1) N b 2) N b 3) N b N) N b N 1) N

7 2132 IEEE SENSORS JOURNAL, VOL. 16, NO. 7, APRIL 1, 2016 b{ f, 0, 1} = b{ f, 0, 1} + ᾱb{ f, 0, 1} = 1 b{ f, 0, 1} = b{ f, 1, 1} + b{ f, 0, 1} + ᾱb{ f, 0, 1} b{ f, 1, 1} = b 1) b{ f, 0, 1} + b{ f, 2, 1} + b{ f, 1, 1} b{ f, 2, 1} = b 2) b{ f, 0, 1} + b{ f, 3, 1} + b{ f, 2, 1} 2 N... b{ f, 2, 1} =b 2) b{ f, 0, 1} + b{ f, 1, 1} + b{ f, 2, 1} b{ f, 1, 1} =b 1) b{ f, 0, 1} + + b{ f, 1, 1} 5) From the DTMC, we can note that the matrx s a N 1)-step lower trangular matrx. C. Transton Probabltes For any user wth arbtraton frequency f, we can create a set of steady-state lnear functons as n Equaton 5), as shownatthetopofthspage. In general, owng to the chan regulartes, we can determne the steady-state probablty by b{ f, 1, 1} =b{ f, 0, 1} j=2 For condton = 1, we obtan 1 j 1 1 ) j 2 b j 1) 6) b{ f, 0, 1} = α b{ f, 0, 1}, = 1 7) Accordng to the second relatonshp n 5), we obtan the steady-state probablty for condton 2 N whch can be determned by b{ f, 0, 1} = 1 ) 1 j 2 α b j 1) b{ f, 0, 1} j=2 8) From the gven relatons, all the values b{ f, 0, 1}, N are expressed as functons of the value b{ f, 0, 1} and of the condtonal transton probabltes between on and off state α and ). For the case of 2, the m-step transton probablty b m, 1 m N 1) should be consdered to be a condton to determne the value b{ f, 0, 1}, 2 N. By usng normalzaton the steady-state probablty b{ f, 0, 1} can be smplfed and t s gven by Equaton 9), as shown at the bottom of ths page. The steady-state probablty n queue shown n Fg. 4 can be calculated by w 1 b{ f,w, 1} =b{ f, 0, 1} j 1 ) j 1 w+ j 1) b 10) j=1 where w [1, 1] s equvalent to the count of the watng stage. Regardng the tmelne for data transmsson, the followng equatons for any user [1, N] are satsfed. αb{ f, l 1, 0} =b{ f, l, 0}, l = 1 11) b{ f, l, 0} =b{ f, l 1, 0}, l [2, L] 12) αb{ f, l 1, 0} =b{ f, 0, 1}, l = 1 13) b{ f, L, 0} =ρ b{ f, 0, 1} 14) Accordng to 7), 9), 11) 14), for = 1, we obtan b{ f 1, 0, 1} = L + 1 ) ρ 1 + α α 15) Furthermore, accordng to 8), 9), 11) 14), for 2 N, we obtan 16), as shown at the bottom of ths page. Then, substtutng the expresson n Equaton 15) and 16) leads to a new expresson for the total probablty α α + α L + ) ρ 1 + α 2, = 1 α b{ f, 0, 1} = α + α L + ) ρ 2 + α 2, = 2 α α + α L + ) ρ + α 2, 3 N + αφ 17) b{ f, 0, 1} = L b{ f, 0, 1}+ b{ f, l, 0}+b{ f, 0, 1}, = 1 l=0 1 = 1 L b{ f, 0, 1}+ b{ f,w, 1}+ b{ f, l, 0}+b{ f, 0, 1}, 2 N w=1 1 + L + 1 ) ρ + α 1 w w=1 j=1 l=0 1 j 1 ) j 1 b 1 w+ j 1) + 1 α ) 1 j 2 b j 1) j=2 9) 16)

8 ZHENG et al.: A NEW MAC PROTOCOL FOR TIME CRITICAL INDUSTRIAL WSN APPLICATIONS 2133 where 1 ) 1 w 1 φ = w=2 j=w 1 ) j b j) 18) whch can be smplfed to 1 φ = b m) m 2 ) 1 k 19) m=2 k=0 In addton, for = 1, we note ρ = α, andfor2 N, we note ρ = α 1 j) j=1 b. A new data transmsson wll occur when all users wth hgher prorty have completed ther tasks. When the watng stage counter s equal to zero llustrated n Fg. 2), we can measure the probablty of the user begnnng to access the channel n a random arbtraton phase by τ = ρ b{ f, 0, 1} =b{ f, L, 0} 20) Consderng the best-case scenaro where there s only one actve communcaton lnk between a unque user and the gateway, the probablty of allocatng the followng data transmsson slot tme to ths user should be constant ρ 1 = 1. Hence, the DTMC can be smplfed as a two-state Markov chan b{ f 1, 0, 1} b{ f 1, 0, 1}. Regardng the worst-case scenaro, τ depends on the m-step state transton probablty b m whch s derved from both the probablty ρ and the steady-state probablty b{ f, 0, 1} whch s non-ndependent of the state transton probablty n queue. Accordng to the above descrpton of the meda access mechansm, the m-step state transton probablty b m s related wth the probablty p whch mples a new transmsson attempt occasonally encounters a collson wth the current user s channel access, whch s due to the presence of hgher-prorty users n the current arbtraton phase. m actve users are consdered watng for data transmsson n ther ndependent steady state, ths probablty p can be defned as p = 1 ) ) m τ. We know that the total tme delay s manly caused by the watng stage whch s formed by several of the nactve states { f,w, 1}. The busy mode of the wreless channel for data transmsson can be modeled to be a seres of state transfer processes whch consst of L steady-state probablty whch means that a user transmts a data packet wth constant length L. b{ f, l, 0} s defned as the probablty of an event whereas the user n s transmttng n the lth tme slot, whle b{ f, 0, 1} s the probablty of the case that certan events requre contnuous channel occupancy n the followng arbtraton phase. We have τ = L b{ f, l, 0} = l=0 L + 1 α ) ρ b{ f, 0, 1} 21) In order to evaluate mutual effects between users, when they access a channel n a random manner, we provde the expressons for the remanng unknown m-step transton probabltes b m). Frst, we defne two new events, one s denoted by Sw) = 1 as the event when the user s queued for a future arbtraton phase; another event s denoted by S j n) = 0, j 2 as the event when the user wth arbtraton frequency f j s preparng to access the medum as the watng stage counter s equal to zero. Accordng to the analyss of the DTMC, S j w) = 1 s equvalent to state { f j,w, 1} whle S j n) = 0 s equvalent to state { f j, 0, 1}. ThevarableSw) = 1 can be mapped as an event-based acton, whch mples that n ths decson-makng process the expected reward gven to the user n j can be expressed as tme delay here, the reward works as a knd of penalty to the system performance). The maxmum watng stage under ths acton Sw) = 1 should be m = AF), where m s the maxmum transton steps from state { f j, 0, 1} to state { f j,w, 1}. Then, the probablty θ j s defned as the condtonal probablty that an actve user n j meets a random non-empty watng queue set AF) because there are users wth lower arbtraton frequences that can be defned as PS j w) = 1 S j n) = 0 Sw) = 1). Gven the channel state transton probabltes, we can calculate θ j as θ j = PS j w) = 1 S j n) = 0 Sw) = 1) = PS j w) = 1 S j n) = 0 j 1 k=1 Sk w) = 1))) 22) where the superscrpt k represents the arbtraton frequency number. Accordance wth provsons of the DTMC, and only when the condtons that user n k wth a lower arbtraton frequency than f k s stll queung n the watng stage at current tme t whle user n j wth arbtraton frequency f j s attemptng to access the channel are both satsfed, a state transton process for user n j from S j n) = 0toS j w) = 1 wll happen the next tme t +1. At ths moment, f k AF), 1 k j 1 s notable. Wth regard to the transton steps m mentoned n the state transton process, t only depends on the number of users who meet the above mentoned condtons, and t mples that the maxmum transton step wll not exceed ths lmt AF), wherem AF) s true. Snce a key assumpton n ths paper s that the probablty of sensng the channel s ndependent across all users, takng a 1-step transton nto consderaton, we have θ 1) j ) = P S j n) = 0 j 1 P k=1 S k w) = 1) 23) Regardng the 2-step transton, we have ) = P S j n) = 0 { ) ) P S 1 w) = 1 P S 2 w) = 1 ) ) + P S 1 w) = 1 P S 3 w) = 1 ) ) + P S 2 w) = 1 P S 3 w) = 1 ) ) + P S 2 w) = 1 P S 4 w) = 1 ) + P S j 2 w) = 1 P θ 2) j S j 1 w) = 1 ) } 24) whch mples that n the current tme when user n j s attemptng to access the channel, there are 2 random users wth lower arbtraton frequency n queue. As a

9 2134 IEEE SENSORS JOURNAL, VOL. 16, NO. 7, APRIL 1, 2016 result, the user n j has to be subsequently transted for two watng stages. To express ths transmsson pattern more ntutvely, we defne a new random matrx as C 2 j 1 =[ζ j 1 2 ] j 1 whch means that j 1 partcpants 2 ) 2 that have lower arbtraton frequences than user n j contans 2 randomly actvated users who are queung n the current arbtraton phase. In 24), C 2 j 1 can be expressed by [ f 1, f 2 ), f 1, f 3 ),...,f 2, f 3 ), f 2, f 4 ),...,f j 2, f j 1 )] T. Through mathematcal nducton, the general form of condton probablty θ j n terms of m-step transton can be rewrtten as ) θ m) j 1 m ) j = P S j n) = 0 P S k w) = 1) q=1 f k rq) 25) where rq) represents the row elements of the matrx C m j 1 =[ζ j 1 m ], q s the row number. j 1) m returns a bnomal coeffcent contanng all possble combnatons of j 1 tems taken m at a tme. Matrx j 1) m has m columns and j 1)!/ j 1 m)!m!) rows. When there are random m users wth lower arbtraton frequences n queue whle user n j attempts to access the channel n the same arbtraton phase. The formula varables PS k w) = 1) are defned as the exstence probablty of the random users wth lower arbtraton frequency f k than f j n the current arbtraton phase. For j = 2, the exstence probablty s PS k w) = 1) = b{ f 1, 0, 1} 26) For 3 j N, the exstence probablty can be derved from 5) and s gven by k 1 PS k w) = 1) = b{ f k, 0, 1}+ b{ f k,w, 1}) 27) we get PS k w) = 1) k 1 = b{ f k, 0, 1}+ w=1 + b{ f k, 0, 1} w=1 b w) k b{ f k, 0, 1}) k 1 j 1 1 ) j 2 b j 1) k 28) j=2 then, { k 1 PS k w) = 1)=b{ f k, 0, 1} 1+ w= ) w ) } b w) k. 29) V. PERFORMANCE ANALYSIS To valdate the performance of the proposed WrArb medum access protocol we consder a star network used for an applcaton n the ndustral automaton doman whch n general has hard real-tme requrements [22]. In ths artcle we assume that f the star network approaches complexty TABLE I SYSTEM PARAMETER USED TO OBTAIN NUMERICAL RESULTS wth more than one user, τ s reduced to less than 1 due to postve transton probablty, 0 P tr < 1 0 < P s 1s notable. As a consequence, the MAC throughput wll decrease due to sgnal collsons wth data packets and/or channel errors caused by nterference [23]. To valdate the performance of the proposed WrArb protocol we use the values and parameters that can be found n the specfcatons of the MAC sub-layer for IEEE The maxmum expected data transmsson rate s equal to 250kbps under the QPSK physcal layer. The data format, such as the PHY header and MAC header, s defned by the standard specfcatons. Specfcally, the MAC address ssue s not consdered n ths analyss, and due to the lmtaton of the MAC protocol data unt length, the maxmum length approprated by the MAC payload s 960bts.Thevalue of the parameters used to obtan the analyss results are lsted n Table I. A. System Delay Analyss The flow chart for WrArb shown n Fg. 2 shows that the total duraton of a data transmsson should be consdered for the followng two cases: need no watng and need watng. The frst case s based on a stuaton where the hghest prorty user the so-called master) has the lowest arbtraton frequency and can access the channel drectly n the present arbtraton phase as there are no other users queueng for the channel. The second case to consder s when users wth lower prorty are forced to let hgher prorty users wth lower arbtraton frequency access the channel frst, whch results n low prorty users havng to queue for random watng stages. Therefore, the total transmsson tme delay of the hghest prorty user conssts of three man parts,.e., 1) tme for arbtraton sgnal transmsson T arb ; 2) tme for fnal arbtraton decson recevng T TxRxarb ; 3) tme for data transmsson T data. For other low prorty users, the total transmsson tme delay needs to be added to a random delay T watng ) n addton to the prevous three parts. Ths random tme delay should be a functon of the ndex varable of the user s arbtraton frequency f.

10 ZHENG et al.: A NEW MAC PROTOCOL FOR TIME CRITICAL INDUSTRIAL WSN APPLICATIONS 2135 The maxmum total tme delay can be expressed as T max ) { T arb + T TxRxarb + T data, = 1 = T arb + T TxRxarb + T data + T watng ), 2 N 30) For a user wth arbtraton frequency f 1, whch s supposed to be the lowest of the entre arbtraton frequency set, or for the network as a specal type of the star network where there s only one actvated user ntendng to transfer data, the possble total transmsson delay s statstcally determnstc. We consder the above case the best-case scenaro. However, n the general case when the packet s transmtted n a consdered tme slot, we can calculate the maxmum watng duraton for any user as T watng ) = 1)T s 31) As T arb, T TxRxarb and T data are constants, when all users wth lower arbtraton frequency than that of user n are wllng to access the channel for data transmsson n the same consdered tme slot, the total communcaton delay T max ) of user n depends on the value T watng ), whch determnes the upper lmt of the queue, whch s determned by the perod of each arbtraton phase T s. B. Normalzed Throughput Analyss In ths secton we wll analyze the system saturaton throughput for the proposed WrArb protocol. We consder the proposed DTMC model to analyze the throughput performance of the WrArb protocol. To calculate the average of the normalzed system throughput S) we follow the procedure n [5] and S s defned as the fracton of tme the channel s used to successfully transmt payload bts and ths can be expressed as E[payload nformaton transmtted n a slot tme] S = 32) E[length of a slot tme] Consder that E[T P ] s the average duraton used for a packet payload transmsson and E[T suc ] s the average duraton of a successful transmsson, E[T wat ] s the average duraton of a transmsson falure because of non-empty set AF), whch means there s at least one user n queue wth a lower arbtraton frequency. In addton, E[T null ] s the average duraton of a non-transmsson tme slot. Hence, the normalzed system throughput can be expressed as P tr P s r data E[T P ] S = 33) P tr P s E[T suc ]+P tr P s E[T wat ]+P tr E[T null ] where the probablty P tr s used to statstcally estmate how the shared channel can be randomly occuped by users from the system pont of channel state, and descrbe how only one transmsson n a randomly consdered tme slot s expected to occur. Accordngly, P s s the probablty that a transmsson occurrng on the channel s successful. When t comes to empty tme slots, the average length of ths non-transmsson tme slot s obtaned wth probablty P tr. Then, we obtan P tr = 1 P tr = N 1 ρ b{ f, 0, 1}) 34) =1 If exactly one user wth the hghest prorty transmts on the channel n the current arbtraton phase, we have Nj=1 N= τ j j+1 1 ρ b{ f, 0, 1})) P s = 1 P s = 35) P tr where P s represents the current user encounterng a transmsson falure due to the presence of other hgher-prorty users n queue and has to wat. If we consder the case where all data packets have the same fxed packet sze, E[P] =r data, E[T P ]=P. Wedefne E[P] as the average packet payload sze, the average amount of payload nformaton successfully transmtted n a slot tme should be P tr P s E[P], snce a successful transmsson occurs n a slot tme wth probablty P tr P s. In the general case, t s thus necessary to assume a sutable probablty dstrbuton functon f ) for the payload sze. To calculate the normalzed system throughput for ths case t s necessary to specfy the correspondng values E[T suc ] and E[T wat ].Fromtheflow chart shown n Fg. 2, we obtan { E[T suc ]=T setup + δ + T arb + δ + T TxRxarb + δ + T data E[T wat ]=T wat 36) where T setup s used to receve the notfcaton from the gateway to algn tme boundares for network synchronzaton. Each user has a perod of tme T arb used to transmt ts arbtraton sgnal. Then, T TxRxarb s used to set the output bt accordng to the fnal arbtraton decson from gateway. It s reasonable to add δ as the turnaround tme.e., transmttng/recevng swtches). Fnally, T data s a perod of tme when the channel remans busy because of successful channel access and the master completes data transmsson. Then, we obtan T data = T phy + T mac + E[T P ] 37) where T phy and T mac are defned as the tme delay for the physcal layer and MAC layer header, respectvely. Furthermore, T wat s a perod of tme durng whch the channel s allocated to another user wth hgher prorty,.e., T wat = T s. C. Worst-Case Scenaro Wthout loss of the generalty, we assume that the network s composed of at least two users, whch are equpped wth default functons to swtch to actve state from nactve state wth a random transton probablty 0, 1] whch represents dfferent events,.e., wake-up from sleep wth a certan probablty, and mgrated nto the current network from other networks. Fg. 5 shows the performance evaluaton n terms of the normalzed system throughput wth dfferent numbers of users.

11 2136 IEEE SENSORS JOURNAL, VOL. 16, NO. 7, APRIL 1, 2016 Fg. 5. Normalzed saturaton system throughput versus dfferent numbers of users. Fg. 7. Tme delay versus dfferent numbers of users. Fg. 6. Normalzed system throughput versus dfferent probablty. Fg. 8. Tme delay versus probablty P tr under dfferent numbers of users. From ths fgure, t can be noted that when the user s actvated from the nactve state at a greater probablty, WrArb can enhance the system throughput performance. Ths dstrbuton s derved from the ablty to successfully transmt more MAC payload bts n each arbtraton phase. As explaned above, WrArb mples that more actvated users wth data transmsson requests wll results n a larger probablty whch shows how the wreless channel s expected to be randomly occuped by users; and also, how the channel can be ensured to be successfully assgned to the user the so-called master) wth the hghest-prorty wthout any collsons. Therefore, more actvated users n the network yeld hgher system throughput performance. Ths fact wll enhance the probablty of channel occupancy whch results n hgher channel utlzaton effcency. Fg. 6 and Fg. 7 show the performance of the WrArb focusng on the normalzed saturaton system throughput and the transmsson delay, respectvely, versus dfferent transton probabltes from nactve state to actve state and dfferent numbers of actvated users. It shows that the saturaton throughput of the system can be ncreased when the user can swtch to be actvated wth a larger transton probablty, or when the network sze ncreases. The multple access scheme and numercal results above descrbe that, the prob- ablty of occupyng the channel n one arbtraton phase wll ncrease wth the number of users, as shown n Fg. 7. Meanwhle n Fg. 7, t can be noted that for any non-hghest prorty user the total transmsson duraton wll be nevtably extended as dscussed n DTMC of WrArb, whch s a result of queung for arbtraton n order to avod nterference from other users wth hgher prorty than the current user. In the same fgure, regardng the user wth the lowest arbtraton frequency n any worst-case scenaro, t can be noted that the tme delay s always mantaned at a value of 5.248ms. Another dstrbuton we can obtan from the same fgure s that f any new user wth a lower AF s actvated, t wll affect other users of data transmsson,.e, the data transmsson of other users wth greater arbtraton frequences wll be affected, whch s reflected n the total transmsson delay beng magnfed by several multples of the perod of arbtraton phase. The maxmum total transmsson duraton s determnstc and predctable rather than ndstnct, due to dfferent AFs representng dfferent tme crtcalty boundares. From Fg. 8, t s clear that the transmsson duraton ncreases wth the total number of users n the network. A larger network sze leads to ncreasng the probablty P tr by ncreasng the transton probablty from nactve state to actve state. The longest transmsson duraton s generated

12 ZHENG et al.: A NEW MAC PROTOCOL FOR TIME CRITICAL INDUSTRIAL WSN APPLICATIONS 2137 Fg. 9. Tme delay versus probablty τ under dfferent numbers of users. by the user who owns the hghest arbtraton frequency. The mnmum total transmsson delay s defntely guaranteed n 5.248ms, whch s the lowest lmt for data transfer used by the user who owns the lowest arbtraton frequency. As the number of users ncreases, the maxmum transmsson tme delay of the user wth the hghest arbtraton frequency s supermposed by nteger multple of 10ms. Snce user n ownng the arbtraton frequency f should avod to nterfere wth the hghest-prorty user s data transmsson, t has to queue up to 1 tmes the perod of the arbtraton cycle. Lke Fg. 8, Fg. 9 descrbes a dstnct group of curves, whch concentrates on the relatonshp between tme delay and user state. Here, we take the probablty τ nto consderaton whch represents a statstcal evaluaton on the channel access and transmsson n a random arbtraton phase. Focusng on the user who owns the hghest arbtraton frequency, t has to avod a non-competton and non-conflct dle channel n the current tme slot for data transmsson of other hgher-prorty users, the probablty of accessng a channel n the current arbtraton phase ncreases as the number of hgher-prorty users ncreases. As explaned above, WrArb mples that more actvated users havng data transmsson requests wll result n a larger probablty whch shows how the wreless channel s expected to be randomly occuped by users; and also, how the channel can be ensured to be successfully assgned to the user the so-called master) wth the hghest-prorty wthout any collsons. D. Channel Utlzaton In ths secton we analyze the channel utlzaton by comparng WrArb wth a TDMA-based WSN, such as WrelessHART. We defne the channel utlzaton n the average number of transmssons per tme slot n a superframe that have been receved successfully by the gateway. The channel utlzaton can be calculated as U = E[ N =1 N data )] 38) E[N slot ] where N data ) represents the total number of data transmssons n a superframe by user n, and N slot represents the total number of tme slots n a superframe. Fgure 10 shows a typcal TDMA structure whch s scheduled wth several ndependent tme slots. The superframe s dvded nto n perodc frames of equvalent perod, whch s determned by the users update rate refresh rate or samplng rate). Consderng WSAN where the regular data s forced to be exchanged perodcally, regardng uplnk servce, the sensor should report envronmental nformaton or mechancal operaton to the gateway perodcally. For the downlnk servce, actuators should execute output nstructons from the gateway perodcally, and the typcal perod tme can vary from 250ms to 1s dependng on the applcatons n the ndustral control system. In ths paper we compare WrArb wth WrelessHART, whch s based on TDMA. In WrelessHART each event s at most allocated one fxed length tme slot whch s lmted to 10ms. As can be seen from Fg. 10, event e 1, e 2, e 3,...,e,...,e N can be correspondngly acqured by the 1 st, 2 nd, 3 rd,..., th, N th tme slot wthn each frame perodcally. As a result, the total number of tme slots allocated to any actve user n, {1, 2,...,N} s n. In addton, consder the tme crtcal applcatons where an emergency event could be trggered when a specfc condton s satsfed, such as n safety and securty servces. In the event of emergency, actuators must react wthn the requred deadlne. For nstance, f the response tme for the emergency event s lmted to 250ms, after the frame of TDMA s rescheduled, the N + 1) th tme slot wthn each frame s assgned to ths emergency e e. Although a total of n tme slots are scheduled to e e, only the N +1) th slot n the current frame when the emergency was trggered s effectvely used whle other n 1 slots wll reman null. That s because the emergency event could not happen perodcally. Let us consder the delay boundary for regular data transmsson whch s fxed to 250ms, and the tme slot s lmted to 10ms. Therefore, n the case of TDMA-based network, we can calculate that the maxmum number of network users n one frame that can be effectvely scheduled s N = 25 under the constrant that no more than one user can use a channel n a gven tme slot. We assume that regardless of perodc regular data or aperodc emergency data, the lnks are scheduled to communcate contnuously. Fg. 11 and 12 llustrate a comparson result of the channel utlzaton of WrArb wth TDMA. It can be seen that full channel utlzaton can always be acheved by the WrArb protocol regardless of the number of network users or the length of superframe. The maxmum number of network users s determned by the prorty of emergency events, whch mples that any new user can access a channel at any tme, once t obtans the hghest prorty assumng that the recever has enough capacty to dstngush arbtraton frequences). It s not necessary to reserve any tme slots to use the non-perodc schedulng strategy of WrArb. The drawback of TDMA s poor channel utlzaton manly due to handlng a temporary emergency event; the TDMA schedulng reserves perodc tme slots for each event. Our proposed WrArb s an on-demand schedulng MAC procotol, whch means that the current tme slot s only assgned to the user wth the hghest prorty, whle those followng tme slots are sequentally assgned to the remanng users n accordance wth the predefned prorty order. Before the perod of data

13 2138 IEEE SENSORS JOURNAL, VOL. 16, NO. 7, APRIL 1, 2016 Fg. 10. The tme slots schedulng for regular data and emergency data wth TDMA and WrArb. channel access prortes whch s utlzed to determne channel access order. It ensures that n each data transmsson stage the most crtcal event s the only one to mmedately gan access to the channel, whle the remanng users have to sleep. There s no channel contenton and nterference for the connectons between the hgh prorty node and the gateway. Therefore, the channel utlzaton by WrArb s 100%. Ths means that WrArb can gve us a hard real-tme guarantee wth determnstc and predctable performance as well as better channel utlzaton than TDMA. Fg. 11. The channel utlzaton of varous protocols versus dfferent numbers of users under delay boundary 250ms. Fg. 12. The upper boundary of channel utlzaton of varous protocols versus dfferent superframe szes. transmsson, a channel access arbtraton procedure should be processed frst. Each user n the network s pre-assgned a dedcated arbtraton frequency wth the purposed to dentfy VI. CONCLUSION Wreless sensor networks have ganed acceptance n a number of domans, and n the future WSNs mght be consdered for tme-crtcal real-tme applcatons that can be found wthn ndustral automaton and avoncs. Control applcatons n ndustral automaton typcally have a hard deadlne on data packet delvery and requre determnstc communcaton. Unfortunately, today s WSNs based on CSMA/CA random backoff and TDMA perodc schedulng cannot offer that fully. Therefore, n ths artcle we propose a new medum access protocol called WrArb whch acheves tme-crtcal data delvery n wreless sensor networks wth low energy expendture. In WrArb each user s pre-assgned a dedcated arbtraton frequency to determne the order for users to gan channel access. In ths way we can offer a collson-free and determnstc communcaton over the wreless medum. To evaluate the performance of the proposed protocol we use a dscrete tme Markov chan. The results obtaned show that the proposed WrArb performs well n terms of delay snce the man bottlenecks n current WSNs are removed. The protocol s ntended for short-range communcaton, and for ths purpose, our protocol s relable, even n harsh ndustral envronments. Snce the proposed protocol s n the early stages of development there are several challenges that need to be nvestgated before a full-scale mplementaton can be acheved. For nstance, more work s needed on settng the arbtraton

14 ZHENG et al.: A NEW MAC PROTOCOL FOR TIME CRITICAL INDUSTRIAL WSN APPLICATIONS 2139 threshold, evaluaton of response tme and nvestgaton of how the proposed protocol could work n multhop scenaros. REFERENCES [1] V. C. Gungor and G. P. Hancke, Industral wreless sensor networks: Challenges, desgn prncples, and techncal approaches, IEEE Trans. Ind. Electron., vol. 56, no. 10, pp , Oct [2] J. Åkerberg, M. Gdlund, T. Lennvall, J. Neander, and M. Björkman, Effcent ntegraton of secure and safety crtcal ndustral wreless sensor networks, EURASIP J. Wreless Commun. Netw., Nov [3] A. K. Mok and S. A. Ward, Dstrbuted broadcast channel access, Comput. Netw., vol. 3, no. 5, pp , [4] M. Khanafer, M. Guennoun, and H. T. Mouftah, A survey of beaconenabled IEEE MAC protocols n wreless sensor networks, IEEE Commun. Surveys Tuts., vol. 16, no. 2, pp , Second Quarter [5] G. Banch, Performance analyss of the IEEE dstrbuted coordnaton functon, IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp , Mar [6] S. Polln et al., Performance analyss of slotted carrer sense IEEE medum access layer, IEEE Trans. Wreless Commun., vol. 7, no. 9, pp , Sep [7] M. Doudou, D. Djenour, and N. Badache, Survey on latency ssues of asynchronous MAC protocols n delay-senstve wreless sensor networks, IEEE Commun. Surveys Tuts., vol. 15, no. 2, pp , Second Quarter [8] S. Petersen and S. Carlsen, WrelessHART versus ISA100.11a: The format war hts the factory floor, IEEE Ind. Electron. Mag., vol. 5, no. 4, pp , Dec [9] J. Åkerberg, F. Rechenbach, M. Gdlund, and M. Bjorkman, Measurements on an ndustral wreless HART network supportng PROFIsafe: A case study, n Proc. IEEE 16th Conf. Emerg. Technol. Factory Autom. ETFA), Sep. 2011, pp [10] M. We and K. Km, Intruson detecton scheme usng traffc predcton for wreless ndustral networks, J. Commun. Netw., vol. 14, no. 3, pp , Jun [11] IEEE Standard for Local and Metropoltan Area Networks Part 15.4: Low-Rate Wreless Personal Area Networks LR-WPANs) Amendment 1: MAC Sublayer, IEEE Standard e-2012, 2012, pp [12] P. Suryacha, U. Roedg, and A. Scott, A survey of MAC protocols for msson-crtcal applcatons n wreless sensor networks, IEEE Commun. Surveys Tuts., vol. 14, no. 2, pp , Second Quarter [13] W. Ye, J. Hedemann, and D. Estrn, An energy-effcent MAC protocol for wreless sensor networks, n Proc. IEEE 21st Annu. Jont Conf. Comput. Commun. Soc. INFOCOM, vol. 3. Jun , 2002, pp [14] W. Ye, J. Hedemann, and D. Estrn, Medum access control wth coordnated adaptve sleepng for wreless sensor networks, IEEE/ACM Trans. Netw., vol. 12, no. 3, pp , Jun [15] P. Ln, C. Qao, and X. Wang, Medum access control wth a dynamc duty cycle for sensor networks, n Proc. IEEE Wreless Commun. Netw. Conf. WCNC), vol. 3, Mar. 2004, pp [16] S. W. Hussan, T. Khan, and S. M. H. Zad, Latency and energy effcent MAC LEEMAC) protocol for event crtcal applcatons n WSNs, n Proc. IEEE Int. Symp. Collaboratve Technol. Syst. CTS), May 2006, pp [17] G. Lu, B. Krshnamachar, and C. S. Raghavendra, An adaptve energy-effcent and low-latency MAC for tree-based data gatherng n sensor networks, Wreless Commun. Moble Comput., vol. 7, no. 7, pp , [18] Y. Sun, S. Du, O. Gurewtz, and D. B. Johnson, DW-MAC: A low latency, energy effcent demand-wakeup MAC protocol for wreless sensor networks, n Proc. 9th ACM Int. Symp. Moble Ad Hoc Netw. Comput., 2008, pp [19] L. Cho, S. H. Lee, and J.-A. Jun, SPEED-MAC: Speedy and energy effcent data delvery MAC protocol for real-tme sensor network applcatons, n Proc. IEEE Int. Conf. Commun. ICC), May 2010, pp [20] M. Salajegheh, H. Soroush, and A. Kals, HYMAC: Hybrd TDMA/FDMA medum access control protocol for wreless sensor networks, n Proc. IEEE 18th Int. Symp. Pers., Indoor Moble Rado Commun. PIMRC), Sep. 2007, pp [21] P. Suryacha, J. Brown, and U. Roedg, Tme-crtcal data delvery n wreless sensor networks, n Dstrbuted Computng n Sensor Systems Lecture Notes n Computer Scence). Hedelberg, Germany: Sprnger-Verlag, 2010, pp [22] J. Åkerberg, M. Gdlund, and M. Björkman, Future research challenges n wreless sensor and actuator networks targetng ndustral automaton, n Proc. 9th IEEE Int. Conf. Ind. Inform. INDIN), Jul. 2011, pp [23] L. Angrsan, M. Bertocco, D. Fortn, and A. Sona, Expermental study of coexstence ssues between IEEE b and IEEE wreless networks, IEEE Trans. Instrum. Meas., vol. 57, no. 8, pp , Aug Tao Zheng S 09 M 15) receved the B.S. degree n communcatons engneerng and the Ph.D. degree n communcaton and nformaton systems from Bejng Jaotong Unversty, Bejng, Chna, n 2006 and 2014, respectvely. He was a Vstng Researcher wth the ABB Corporate Research, Sweden, from 2012 to 2013, where he was nvolved n the Industral Communcaton and Embedded Systems. He s currently a Faculty Member wth Bejng Jaotong Unversty. Hs specfc areas of research nterest manly focus on physcal layer and MAC layer communcaton technology n ndustral network and Internet of Thngs, nterference avodance technology and coexstence network optmzaton, and hardware mplementaton of wreless sensor networks. Mkael Gdlund M 99) receved the M.Sc. and Ph.D. degrees n electrcal engneerng from Md Sweden Unversty, Sweden, n 2000 and 2005, respectvely. He s currently a Full Professor of Computer Engneerng wth Md Sweden Unversty, and snce 2014, he s also workng as a Scentfc Advsor wth ABB Corporate Research. In 2005, he was a Vstng Researcher wth the Department of Informatcs, Unversty of Bergen, Norway. From 2006 to 2007, he was a Research Engneer and Project Manager, responsble for wreless broadband communcaton at Acreo AB, Sweden. From 2007 to 2008, he was a Senor Specalst and Project Manager wth responsblty for next-generaton IPbased rado solutons at Nera Networks AS, Bergen, Norway. From 2008 to 2013, he was a Senor Prncpal Scentst and Global Research Area Coordnator of Wreless Technologes wth ABB Corporate Research. He holds more than 20 patents granted and pendng applcatons) n the area of wreless communcatons, and has authored or co-authored over 100 scentfc publcatons n refereed fora. Hs research nterest are wreless communcaton and networks, wreless sensor networks, access protocols, and securty. He won the Best Paper Award at the IEEE Internatonal Conference on Industral IT n He s an Assocate Edtor of the IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS. Johan Åkerberg M 08 SM 11) receved the M.Sc. and Ph.D. degrees n computer scence and engneerng from Mälardalen Unversty, Sweden. He has close to 20 years experence wthn ABB n varous postons, such as Global Research Area Coordnator, Research and Development Project Manager, Industral Communcaton Specalst, and Product Manager. He s currently a Prncpal Scentst wth ABB Corporate Research, Sweden. He s also a Vce Char n the IEEE-IES Techncal Commttee on Factory Automaton. He s manly nvolved n communcaton for embedded real-tme systems n ndustral automaton and s frequently nvted to gve talks to governmental bodes, nternatonal unverstes, and automaton fars. He holds more than ten patents granted and pendng applcatons) n the area of wred/wreless ndustral automaton, and has authored or co-authored numerous scentfc publcatons n refereed conferences and journals.