VTP-CSMA: A Virtual Token Passing Approach for Real-Time Communication in IEEE 802.11 Wireless...

IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 3, NO. 3, AUGUST 2007 215

VTP-CSMA: A Virtual Token Passing Approachfor Real-Time Communication in IEEE 802.11

Wireless NetworksRicardo Moraes, Francisco Vasques, Member, IEEE, Paulo Portugal, and José Alberto Fonseca, Member, IEEE

Abstract—Currently, there is a trend towards the implementa-tion of industrial communication systems using wireless networks.However, keeping up with the timing constraints of real-timetraffic in wireless environments is a hard task. The main reasonis that real-time devices must share the same communicationmedium with timing unconstrained devices. The VTP-CSMAarchitecture has been proposed to deal with this problem. Itconsiders an unified wireless system in one frequency band,where the communication bandwidth is shared by real-time andnon-real-time communicating devices. The proposed architectureis based on a virtual token passing (VTP) procedure that circu-lates a virtual token among real-time devices. This virtual tokenis complemented by an underlying traffic separation mechanismthat prioritizes the real-time traffic over the non-real-time traffic.This is one of the most innovative aspects of the proposed ar-chitecture, as most part of real-time communication approachesare not able to handle timing unconstrained traffic sharing thesame communication medium. A ring management procedure forthe VTP-CSMA architecture is also proposed, allowing real-timestations to adequately join/leave the virtual ring.

Index Terms—IEEE 802.11, IEEE 802.11e, real-time communi-cations, wireless networks.

I. INTRODUCTION

PRESENTLY, the IEEE 802.11 family of protocols is oneof the most used sets of wireless local area networks

(WLANs). It was standardized in 1999 by the IEEE, as theIEEE 802.11 standard, which was later reaffirmed in 2003[1]. Recently, the IEEE 802.11e [2] standard was publishedas an amendment to the original standard. This amendment isintended to provide differentiated levels of quality of service(QoS) to the supported applications, including the transport ofvoice and video over WLANs. Nowadays, the IEEE 802.11family of protocols is one of the main contenders to becomethe de facto standard for industrial wireless communications.

Supporting reliable real-time (RT) communications is one ofthe major requirements that is usually imposed to industrial com-munication systems [3]. In such systems, real-time control data(typically small-sized packets) must be periodically transferredbetween sensors, controllers, and actuators according to stricttransfer deadlines. Usually, the RT communication infrastruc-ture must be shared with multimedia (voice and video) traffic and

Manuscript received February 2, 2007; revised May 10, 2007. Paper no.TII-07-02-0017.R1.

R. Moraes, F. Vasques, and P. Portugal are with Faculdade de Engenharia,Universidade do Porto, 4200-465 Porto, Portugal.

J. A. Fonseca is with DETI, Universidade de Aveiro, 3810-193 Aveiro,Portugal.

Digital Object Identifier 10.1109/TII.2007.903224

data-related background traffic [4]. A fundamental assumptionthat must be considered when dealing with wireless communi-cations is that the physical medium is essentially an open com-munication environment. That is, any new participant can try toaccess the communication medium at any instant (according tothe MAC rules) and establish its own communication channels.As a consequence, the system load cannot be predicted at systemsetup time, nor controlled during the system run-time. Therefore,the underlying wireless communication protocol must be able toguarantee the timing constraints of RT traffic in a communicationmedium shared with timing unconstrained traffic.

Traditionally, the RT communication behavior in wiredCSMA environments has been guaranteed through the tight con-trol of every communicating device [5]. The coexistence of RTcontrolled stations with timing unconstrained stations has beenmade possible by constraining the traffic behavior of the latter,for instance using traffic smoothers [6], [7]. Unfortunately, thisapproach is not adequate for wireless environments, since it is notpossible to impose any traffic smoothing strategy upon stationsthat are out of the sphere-of-control of the RT architecture.

In this paper, we propose the VTP-CSMA architecture to dealwith this problem in fully meshed communication scenarios.This paper is an extended version of [8], and it is organized asfollows. The state-of-the-art on supporting real-time communi-cation with the IEEE 802.11 protocol is presented in Section II.In Section III, the VTP-CSMA proposal is presented in detail. Asimulation analysis of the proposed VTP-CSMA architecture isdiscussed in Sections IV and V. In Section VI, the VTP-CSMAproposal is extended to address dynamic communication sce-narios, allowing real-time stations to adequately join/leave thevirtual ring. Finally, in Section VII, some conclusions are drawn.

II. SUPPORTING RT COMMUNICATION WITH IEEE 802.11

There is a strong similarity between wired IEEE 802.3 andwireless IEEE 802.11 networks, since they use a similar CSMAmechanism to manage the medium access. However, and con-trarily to the case of wired networks, the CSMA collision de-tection (CSMA/CD) procedure cannot be used on wireless en-vironments, as it would require the implementation of a full-du-plex radio. As a consequence, the IEEE 802.11 standard [1]implements a collision avoidance (CSMA/CA) procedure re-ferred to as distributed coordination function (DCF). The IEEE802.11e amendment incorporates an additional hybrid coordina-tion function (HCF) that is only used in QoS network configura-tions. For further details concerning the description of the IEEE802.11/802.11e protocols, please refer to the standards [1], [2].

1551-3203/$25.00 © 2007 IEEE

Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on February 3, 2009 at 19:12 from IEEE Xplore. Restrictions apply.

216 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 3, NO. 3, AUGUST 2007

Several communication mechanisms adequate to supporttime constrained communication are already included in theIEEE 802.11/802.11e standards. The point coordination func-tion (PCF) has been proposed in the original IEEE 802.11standard as an optional access mechanism. It implements acentralized polling scheme to support synchronous data trans-mission, where the point coordinator (PC) performs the roleof polling master. Despite PCF being well suited to handledelay-sensitive applications, most part of the WLAN cardsnever actually implemented the PCF scheme, due to com-plexity reasons. The HCCA (HCF controlled channel access)mechanism has been proposed in [2] as an improvement to thePCF mechanism. However, preliminary studies [9], [10] havealready shown that the HCCA mechanism may not be suitableto guarantee the special requirements of industrial applications.Nevertheless, a number of improvements have been proposedto reduce the HCCA polling overhead. For instance, Son et al.[11] proposed a polling scheme where the hybrid coordinator(HC) punishes the stations that have no packets to transmit.When a station transmits a null frame, this station will not bepolled again during a period of time.

Other innovative approaches are also being developed to pro-vide real-time communication services to wireless-supportedapplications. The main technical solutions are based on polling,master-slave, and token passing approaches.

Lo et al. [12] designed a polling mechanism called con-tention period multipoll (CP-Multipoll), which incorporates theDCF access scheme into the polling scheme. It uses differentbackoff values for the multiple message streams in the pollinggroup, where each station executes the backoff procedure afterreceiving CP-Multipoll frame. The contending order of thesestations is the same as the ascending order of the assignedbackoff values.

In [13], a polling scheme based on a master-slave solution isproposed. A virtual polling list (VPL) contains the MAC addressof the wireless slaves to be polled, and a virtual polling period(VPP) defines the duration of the polling cycle. When a slavereceives a poll frame from the master, it can transmit a responseframe to the master, or directly to another slave. After polling allthe slaves registered in the VPL, the master invites other slavesinto the network through the broadcast of an entry claim frame.

In [15], Miorandi and Vitturi proposed a solution basedon a master-slave architecture on top of IEEE 802.11. In thatproposal, cyclic packets are exchanged by means of peri-odic queries sent by the master to the slaves. Three differenttechniques were proposed to handle acyclic traffic: the firsttechnique queries the slaves that signaled the presence ofacyclic data, at the end of the current polling cycle. The secondtechnique allows a slave, when polled, to send directly acyclicdata to the master. The third one exploits the decentralizednature of the IEEE 802.11 MAC protocol. When acyclic dataare generated, it allows a slave to immediately try to send datato the master.

Similarly to the VTP-CSMA mechanism proposed in thiswork, there are several other solutions based on token passingmechanisms. In [17], Ergen et al. proposed the wireless tokenring protocol (WTRP). Basically, the WTRP is a MAC protocolthat exchanges special tokens among stations and uses multiple

timers to maintain the nodes synchronized. Besides, WTRP re-quires the joining node to be connected to its logical predecessorand successor through a local connectivity table.

Cheng et al. [18] presented a wireless token-passing protocol,named Ripple. Ripple uses six types of frames: DATA, NULL,RTS, CTS, ACK, and ready-to-receive (RTR). The frame for-mats are the same as those defined in 802.11, except that Rippleonly utilizes fixed-duration DATA frames. The RTR frame hasthe same format as CTS and is used by a station to request aDATA frame from its upstream station. Basically, Ripple modi-fies the data transmission procedure of 802.11 DCF and employsRTS and RTR frames as tokens. A station can only send a DATAframe if it holds a token.

Finally, Sobrinho and Krishnakumar [19] adapted the EQuBmechanism [20] to wireless networks. The proposed black-burst(BB) scheme implements a distributed MAC scheme in an adhoc CSMA wireless networks (IEEE 802.11 DCF). It requiresthe shut-down of the random retransmission scheme. Real-timestations implementing the black burst approach contend for thechannel access after a medium interframe spacing , ratherthan after the long interframe spacing used by standardstations. Thus, real-time stations have priority over standardstations. When a real-time station wants to transmit, it sortsits access rights by jamming the channel with energy pulses(BBs) immediately after the channel becoming idle during .The length of the BB transmitted by a real-time node is an in-creasing function of the contention delay experienced by thenode. In [21], a similar scheme is presented, where voice nodes(real-time stations) use energy-burst (EB) periods to prioritizereal-time packets over data packets.

The token passing approach presented in [17] is compatiblewith IEEE 802.11 compliant devices. Conversely, the Rippleprotocol [18] modify the MAC layer, impairing the use of COTShardware. Nevertheless, both above-referred token passing andpolling schemes require a closed communication environment toguarantee the real-time communication behavior (relevant ex-ception is the BB scheme that is based on a forcing collisionresolution approach). Otherwise, they will not work properly, asthey are not able to handle messages sent by external stations.This is not a realistic assumption, as it cannot be assumed thatis possible to create a zone completely free of external wirelessstations in typical industrial environments.

Therefore, there is a need for moving from closed to opencommunication environments, when dealing with real-timecommunication. A new communication paradigm will need toemerge, as the current paradigm is still based on closed andcontrolled environments. We foresee that the most promisingsolutions to provide RT communication will allow the coex-istence of both real-time and timing unconstrained stations inthe same communication domain, imposing the prioritizationof the real-time traffic.

The VTP-CSMA architecture has been proposed to cope withthose next generation communication environments. It is basedon a forcing collision resolution mechanism, which is able toprioritize RT-traffic over timing unconstrained traffic, withoutcontrolling the latter. That is, instead of controlling all trafficgenerated by all stations, the proposed approach will only con-trol the traffic generated by RT stations, prioritizing it over the


MORAES et al.: VTP-CSMA: A VIRTUAL TOKEN PASSING APPROACH 217

timing unconstrained traffic. To our best knowledge, the pro-posed VTP-CSMA architecture and the BB schemes [19], [21]are the only solutions that enable RT communication in opencommunication environments. The main disadvantage of theBB scheme is that it compels the modification of the MAClayer and possibly also of parts of the PHY layer (e.g., radioICs), which impairs the use of COTS hardware. Although VTP-CSMA mechanism also needs to modify parts of the MAC layer,it could be easily implemented in COTS hardware (e.g., FPGA)upon standard 802.11e hardware.

III. VTP-CSMA ARCHITECTURE

The VTP-CSMA architecture is based on the control of themedium access right, by means of a VTP procedure among RTstations. This procedure is complemented by a traffic separa-tion mechanism (TSm), which guarantees that, whenever a RTstation is contending for the medium access, it will win the con-tention prior to any other standard (ST) station.

The TSm mechanism works as follows: whenever a col-lision between a RT station and a set of ST stations occurs,all but the RT station will use the prioritized medium accessmechanism (EDCA) and select a random backoff intervalaccording to the access category (voice, video, best-effort,and background). Conversely, the RT station (hereafter alsoreferred to as a VTP-CSMA station) transfers its traffic atthe highest priority level of the EDCA mode, i.e., usingthe same arbitration interframe space (AIFS) of the voicecategorybut, setting the contention window (CW) to the value

.This means that any VTP-CSMA station will always try to

transmit its frame in the first EDCA available slot, while allthe other ST stations will wait during a time interval evaluatedby the local backoff functions. Nevertheless, if two or moreVTP-CSMA stations simultaneously contend for the mediumaccess, they will collide and eventually discard the frame (aftera maximum number of retransmission attempts). This behavioris overcome by means of a VTP procedure, which serializes thetransmission of the VTP-CSMA stations. The VTP procedurepresented in this section just considers a static RT communica-tion environment, with a fixed number of RT stations. Later, inSection VI, the VTP procedure is extended to handle dynamicRT communication scenarios.

The VTP procedure considers a process group withnp members. The membership is represented as

. The notion denotes the th sta-tion in and is also used as station identification (ID) for .Each station in maintain a local access counter ( ), whichis used as a basic mechanism to circulate a virtual token in . Thegeneric th VTP-CSMA station captures the virtual token when

equals . If the station has queued messages, then it willimmediately transfer them during a time interval upper boundedby the transmission opportunity period (TXOP). The TSmmechanism guarantees that the VTP-CSMA station will win themedium access contention. At the end of the current TXOP, eachVTP-CSMA station will increase its value, passing the

Fig. 1. VTP-CSMA mechanism.

virtual token to the next station (station with .1

Whenever the VTP-CSMA station holding the token does nothave any RT message to transfer, the virtual token will be passedif the medium remains idle during a specific time. This time isdefined by counter . As theEDCA mechanism allows any station to start a transmissionafter , it enables the coexistenceof RT stations with non-real-time ST stations in the sharedcommunication environment.

Fig. 1 describes the VTP-CSMA mechanism, which is rep-resented by four procedures: initialization, main, transmission,and listening. According to the initialization procedure (see line2, Fig. 1), the value is set to in all VTP-CSMA sta-tions. Three local variables of type integer are de-fined, where and are slot time counters, whereas is aspecial collision counter used for re-initialization purposes. Themain procedure is executed at the beginning of each time slot

1The ACo must be increased by “mod” operation, i.e., ACo =(ACo mod np) + 1.



, verifying if the special collision counter ex-ceeds the maximum number of retransmission attempts (RN).This will occur if multiple VTP-CSMA stations simultaneouslycontend for the medium access. This means that the distributedACo got an inconsistent value, forcing the re-initialization of thering (initialization procedure). Otherwise, depending on boththe channel event during the last slot and the value, eachVTP-CSMA station will take a specific action.

The transmission procedure is only executed by the VTP-CSMA station holding the token , and it is onlyinitiated after the medium being idle during interval,i.e., the minimum AIFS value for the EDCA mode, as defined byIEEE 802.11e. Conversely, the listening procedure may be ini-tiated after the medium being idle during SIFS .Five channel states are defined (determined at the beginning ofeach time slot).

1) Transmission from other stations: One or more messagesare being transmitted over the channel.

2) Successful transmission from other stations: The channelis idle, and a successful message transmission from otherstation finished one time slot ago.

3) Channel continuing idle: The channel is idle and was alsoidle one time slot ago.

4) Channel idle after collision: The channel is idle, and therewas a collision one time slot ago.

5) Successful transmission: The channel is idle, “I am thetransmitting station, and I finished the transmission of oneor more messages (upper bounded by the TXOP interval)”one time slot ago.

According to these channel states and the value, each VTP-CSMA station takes a specific action. First, whenever the VTP-CSMA station captures the virtual token , itwill execute the transmission procedure (lines 31–44). This pro-cedure works as follows. If the VTP-CSMA station holdingthe token have a RT message to transfer, it will immediatelystart the transmission. If a collision occurs, the VTP-CSMAstation increments its counter, and it will retry the trans-mission until the maximum defined number of transmission at-tempts (RN). Whenever, a successful transmission occurs (lines34–35), the VTP-CSMA station holding the token will executethe listening procedure, where each VTP-CSMA station will in-crease its value, passing the virtual token to the next sta-tion. Conversely, whenever the VTP-CSMA station holding thetoken does not have any RT message to transfer, it will allowdefault stations to contend for the medium access, during a timeinterval multiple of .

As illustrated in Fig. 1, all VTP-CSMA stations that do nothave the token will execute the listening proce-dure (lines 10–30), and depending on the channel state, theseVTP-CSMA stations will take a specific action: a) transmissionfrom other stations (lines 14–17): All VTP-CSMA stations waitfor the end of transmission and then update the variables and

; b) successful transmission from other stations (lines 18–19):All VTP-CSMA stations update and ; c) channel con-tinuing idle (lines 20–27): All VTP-CSMA stations increment

and verify the value of . If , all VTP-CSMA sta-tions increment its value. That is, and the channelcontinuing idle state means that a successful transmission oc-

Fig. 2. Behavior of the VTP-CSMA mechanism.

curred and, a TXOP period finished. Besides, will be incre-mented, and each time that value is greater or equal than 3,all VTP-CSMA stations must also increment its value; d)channel idle after collision (lines 28–29): All VTP-CSMA sta-tions increment and .

Most likely, whenever a collision occurs involving just STstations (EDCA stations), the next events will be followed bycontinuing idle channel states, due to the selected backoff in-terval , for voice category). However, when aVTP-CSMA station is involved in the collision, the next channelstate will be always a transmission

. Therefore, the reset mechanism (initialization procedure) isactivated whenever there are RN consecutive collisions withoutany idle time greater thanamong collisions.

Fig. 2 illustrates the behavior of the proposed VTP-CSMAmechanism. Assume that three VTP-CSMA stations are sharingthe physical medium with one EDCA station (ST station). Tosimplify the example, it is considered only the voice access cat-egory for the ST station . According to theinitialization procedure (see line 2, Fig. 1), the ACo value isequal to in all VTP-CSMA stations after(instant a1). Moreover, as it handles a fixed number of stations,each VTP-CSMA station knows the total number of RT stations

. Therefore, station (token holding station) runsthe transmission procedure, while stations and ex-ecute the listening procedure. However, as station has nomessages to be transferred (until instant a2), this station will alsoexecute the listening procedure. This procedure consists in con-tinuously monitoring the channel. According to the example, thenext channel states are followed by channel continuing idle pe-riods. Therefore, all VTP-CSMA stations willstart the slot counter, incrementing counter (see lines 20–27,Fig. 1). Then, counter will expire in all stations at instant a3

and the virtual token is passed to station(lines 25–27). As the channel remains idle ( station has nomessage to be transmitted), counter will expire again at in-stant a5, passing the virtual token to station .

This station may now start to transfer its own real-time mes-sages. However, considering that between instants a4 and a5,the ST station generated one packet, both stations will start thetransmission at the same time. Consequently, a collision willoccur and the VTP-CSMA station will be able to detect it



by the absence of the acknowledgment (ACK) frame. Accordingto the underlying TSm mechanism, it will retry the transmis-sion after following the end ofthe medium busy condition. This means that will start thetransfer of its real-time messages at instant a7 (considering thatthe ST station selected a backoff value greater than 0). It is worthnoting that stations and after instant a6 do not detectthe successful transmission state, and after , it isdetected the channel continuing idle state and the slot counterstarts incrementing (see line 21, Fig. 1). However, the begin-ning of a transmission is detected (instant a7) before the expira-tion of counter . Therefore, station continues the trans-mission and both and stations must wait for its com-pletion (see line 15, Fig. 1).

Afterwards, when the end of transmission is detected (suc-cessful transmission state), if there were still packets in thequeue and enough TXOP time to transmit another packet,including its response frame, the token holding stationwould start another transmission just after .Therefore, when a station transmits multiple frames, theis incremented only at the end of TXOP. As the station hasno more packets to be transmitted, all VTP-CSMA stationsdetect the successful transmission state and is reset to 1 (seelines 18–19, Fig. 1). In the next event (continuing idle state),the virtual token is passed to the next station (instant a9) byincrementing the ACo counters in all the VTP-CSMA stations.

IV. SIMULATION MODEL

The target of the presented simulations is to evaluate how theVTP-CSMA architecture is able to support real-time traffic in acommunication environment shared with timing unconstrainedsources, when compared to the IEEE 802.11e EDCA functionusing the highest access category (voice).

The VTP-CSMA simulation model is implemented using astochastic petri net (SPN) model, previously developed to as-sess the timing behavior of the EDCA function [22]. It uses asemi-Markov error model, where the channel is always in oneof two states: good or bad. This model assumes that bit errorsare independent, with a constant bit error rate in each state. Forthe parametrization of the model, the states holding times arelog-normal distributed according to [23]. The mean duration ofgood state is 65 ms, the mean duration of bad state is 10 msand, the coefficient of variation (CoV) for the bad state holdingtimes has been set to 10 and for the good state to 20. Thesemean burst lengths leading to a rather bad channel [24], wherethe steady-state probability for finding the channel in bad stateis approximately 13.3%. Two sets of simulations are assessed,differing in their respective mean bit error rate (BER). The firstset defines a mean BER of , while the second set definesthat no errors occur.

A. Simulation Scenario 1

The first simulation scenario is intended to assess the impactof the timing unconstrained traffic (generated by standard sta-tions) upon the real-time traffic. It considers an open communi-cation environment, where standard (ST) EDCA stations sharethe communication medium with a set of real-time (RT) stations.

TABLE ISIMULATION DATA

These RT stations implement either the VTP-CSMA architec-ture (RT VTP-CSMA) or the EDCA function (RT EDCA). RTstations transfer just RT messages (small sized packets at pe-riodic rate), whereas ST stations transfer three types of traffic:voice (VO), video (VI), and background (BK).

Two simulation cases are analyzed. First, a small popula-tion case considers 10 ST stations together with 10 RT stations(either RT EDCA or RT VTP-CSMA) operating in the samefrequency band. Second, a large population case extends thenumber of ST stations to 40. Each station operates at orthog-onal frequency division multiplexing (OFDM) PHY mode witha data rate of 36 Mb/s. The physical parameters used in the sim-ulations are based on the IEEE 802.11a PHY mode [1].

The RT traffic is characterized by periodic sources with asmall amount of jitter, modelled by a normal distribution with

% ( is the standard deviation and is the averageexpected value). The ST traffic is modelled by Poisson trafficsources. The maximum number of transmission attempts is setto 4 (RN). The MAC queue size is set to 50 positions. All otherrelevant simulation parameters are shown in Table I.

The generated data frames have a constant size. Each RT sta-tion generates 500 packets/s with 45 bytes for data payload,which is equivalent to 1 packet every 2 ms. Therefore, each RTstation imposes a constant network load of 180 kb/s, which rep-resents about 0.5% of the total network load (without consid-ering the MAC and PHY headers). For the set of ST stations,the offered load ranges from 5% to 95% of the total networkload. Each ST station generates voice, video, and backgroundpackets/s with the same rate, in order to impose the requestedoverall network load. Finally, for the proposed simulations, it isassumed that there is neither node mobility nor hidden stations.

B. Simulation Scenario 2

The second simulation scenario assesses the behavior of theVTP-CSMA architecture in a fixed network load environment(ST traffic load of 50%). The target of this simulation scenariois to assess the behavior of the VTP-CSMA architecture whenthere is a variable number of RT VTP-CSMA stations (5 to 80).Again two simulation cases are analyzed: the small populationcase, which considers the case of just 10 ST stations; and thelarge population case, which extends the number of ST stationsto 40. In both simulation cases, the ST stations equally share afixed network load of 50%.

Each RT VTP-CSMA station also generates periodic mes-sages according to a normal distribution with %, 100packets/s with 45 bytes for data payload, which is equivalent to1 packet every 10 ms (whereas in the first simulation scenario a



packet was generated every 2 ms). Again, the ST stations havea Poisson traffic source and transmit three types of traffic. Thepacket sizes are the same used in Scenario 1. Therefore, in orderto impose a fixed network load of 50% for the small and largepopulation cases, each ST station generates about 74 and 18.5voice, video, and background packets per second, respectively.All the other simulation parameters omitted in this subsectionare the same as those used for the Simulation Scenario 1.

V. SIMULATION RESULTS

All the simulation results have been obtained using a 95%confidence interval with a half-width interval of 5%. The per-formance metrics analyzed include: throughput, packet loss, av-erage delay, and average queue size. The throughput is the ratiobetween the total number of successfully transferred packetsand the total number of generated packets for each traffic stream.Therefore, it represents a relative throughput. The packet lossmetric is computed as (1-throughput) and represents thepercentage of packets that are lost for each traffic stream. Theaverage delay is the average delay required to transfer a packet,measured from the start of its generation at the application layerto the end of the packet transfer. The average queue size repre-sents the average output buffer occupancy.

The RT communication behavior is usually assessed forworst-case scenarios. However, when dealing with probabilisticmedium access networks, the worst-case scenarios address justrarely occurring cases. The analysis of those rarely occurringcases is definitely relevant for safety-critical applications, butit is highly pessimistic when dealing with typical industrialcommunication environments. It is well known that this typeof environments is usually loss tolerant in what concerns theloss of some message deadlines. For instance, the transfer ofa video stream may be specified to tolerate a maximum of10% deadline loss rate, if the lost frames are “adequately”spaced. Another example of relevant loss tolerant applicationsis networked control systems (NCSs) scheduled according tothe (m,k)-firm model [25], or the support of VoIP applications,where an average packet delay below 150 ms and an averagejitter ms are acceptable for most typical applications [26].For a real-time communication system, the average packetdelay must be kept smaller than message deadlines, in orderto minimize the occurrence of deadline misses. On the otherhand, throughput values as high as possible guarantee a re-duced number of message losses. Another relevant metric is theaverage queue size, which represents the ability of the protocolto dispatch periodically queued messages. Therefore, the use ofa simulation setup for the analysis of real-time communicationin probabilistic medium access networks is well justified, whendealing with loss tolerant applications.

The first simulation scenario is intended to assess the be-havior of the VTP-CSMA versus EDCA mechanisms, whensupporting real-time traffic in a communication medium sharedwith timing unconstrained traffic. The average packet delays forthe small and large population scenarios are plotted in Figs. 3and 4, considering that no communication errors occur.

Figs. 3 and 4 show that, for an external load above 15%, theRT VTP-CSMA stations have an average delay much smaller

Fig. 3. Average delay—small population.

Fig. 4. Average delay—large population.

than the RT EDCA stations. For instance, in the large popu-lation scenario (see Fig. 4), for a 55% external network load,the VTP-CSMA stations take in average 1.412 ms to transfer aRT packet, while the EDCA stations take in average 18.161 ms.More importantly, when considering the RT traffic transferredby RT VTP-CSMA stations, the average packet delay is nearlyconstant, whatever the external network load. This is a very rel-evant result when considering real-time communications, as itclearly highlights the negligible impact of the timing uncon-strained traffic upon the timing behavior of the real-time traffic.On the other hand, Figs. 3 and 4 also illustrate that the VTP-CSMA stations have a small influence upon the behavior of theother access categories, except in what concerns a small (and ex-pected) degradation of the voice (VTP-CSMA) traffic behavior.

The average throughput was also evaluated for, respectively,the small and the large population scenarios (see Figs. 5 and6). The percentage of packet losses can be inferred fromthese results, as the percentage of packet losses is given by(1-throughput)*100. Similar results were obtained for the largepopulation scenario. It is clear that RT VTP-CSMA stationsare able to transfer significantly more RT packets than RTEDCA stations. The main reason is that, using the VTP-CSMAscheme, the RT packets generated by different RT stations areglobally serialized, avoiding collisions among RT packets.



Fig. 5. Throughput—small population.

Fig. 6. Throughput—large population.

The error-prone behavior of wireless channels is usually re-garded as one of the major drawbacks when supporting real-timecommunications. In order to assess the impact of errors upon thebehavior of the VTP-CSMA architecture, a set of simulation ex-periments have been carried out, considering the channel errorparameters presented earlier. For the sake of simplicity, only thevalues for RT stations (EDCA or VTP-CSMA) are plotted forScenario 1, and just the values for the RT VTP-CSMA stationsare presented for Scenario 2.

Fig. 7 illustrates the average packet delay considering botherror-free and error-prone scenarios. In spite of being nearlyconstant (3–5 ms), the average delay is significantly larger thanfor the case of an error-free scenario (1–1.6 ms). This behaviorcan be rooted to two different causes. On the one hand, when-ever the distributed variable becomes inconsistent, sooneror later the reset mechanism will be called up (see line 2, Fig. 1)to re-synchronize the virtual ring. This mechanism is highly ef-fective, as the average packet delay remains nearly constant de-spite the external network load. On the other hand, the consid-ered packet error rate is highly pessimistic (40% of packet errorprobability during the error bursts). Therefore, the re-synchro-nization rate of the assessed scenario is much higher than ex-pected for traditional industrial environments.

Fig. 7. Average packet delay—small and large population.

Fig. 8. Average queue size—small and large population.

Another important aspect in RT communication is related tothe average queue size. From Fig. 8, for an error-free scenario,the average queue size of the RT VTP-CSMA stations is nearlyconstant and smaller than 1. This result is consistent with thepacket generation periodicity of 2 ms and the average packetdelay of (1–1.6 ms). For the error-prone scenario, as the averagepacket delay is (3–5 ms), the average queue size will be largerthan 1 (there will be frequent deadlines misses). These resultsare significantly better than the results for the EDCA function(also plotted in Fig. 8).

The target of the second simulation experiment is to assess thebehavior of the VTP-CSMA architecture in a fixed network loadenvironment (external traffic load of 50%) and an increasingnumber of VTP-CSMA stations ( . Fig. 9 il-lustrates the average packet delay in the case of 10 ST (smallpopulation case) or 40 ST (large population case) stations.

The simulation results clearly show that, for a fixed externaltraffic load, the average packet delay is nearly constant (2–3ms) for a number of RT stations up to 65–70 stations in anerror-free scenario. However, in an error-prone scenario, the av-erage packet delay significantly increases. This means that theratio between the average packet delay for error-prone versuserror-free scenarios is highly influenced by the number of RTstations in the virtual ring. This is an expected result, as after



Fig. 9. Average delay—small and large population.

Fig. 10. Average queue size—small and large population.

each re-synchronization of the virtual ring, the virtual token willbe assigned to station . Thus, the re-synchronization activ-ities have a strong impact upon the communication service pro-vided to the stations with higher values for NA.

Finally, Fig. 10 illustrates the values for the average queuesize. In an error-free channel, the VTP-CSMA mechanism isable to efficiently serve the RT traffic for up to 65–70 RT sta-tions, as the average queue size is kept smaller than 1 packet.Whenever the highly pessimistic error characteristics are con-sidered, the VTP-CSMA mechanism is able to support just upto 35 VTP-CSMA stations in the virtual ring.

VI. VIRTUAL RING MANAGEMENT

The VTP-CSMA architecture presented in Section III is ableto handle only a fixed number of RT stations (np stations).In this section, an enhanced ring management procedure isproposed, allowing the VTP-CSMA architecture to be an opengroup. Thus, a station can dynamically join or leave the virtualtoken ring, enabling the support of dynamic communicationscenarios. The ring management includes procedures to 1)add RT-stations to the ring; and 2) remove RT-stations fromthe ring. These procedures must ensure the two followingproperties. Agreement: All VTP-CSMA stations must agree

on the values of the distributed variables and np. That is,at whatever instant of time, all stations know the address of thetoken holder and the total number of stations belonging to thegroup . Uniqueness: Each station must be assigned a uniqueNA, which ranges between 1 and .

Unless these two properties are satisfied, mutual access to themedium by the RT-stations cannot be ensured. Besides theseproperties, some further assumptions are made regarding the ca-pabilities of the communication system.

The first assumption is that stations belonging to groupare able to exchange ring management messages. Each message(msg) contains the fields sender and type;sender identifies thesource, whereas type is concerned with the function of the mes-sage itself. Depending on the value assigned to the type param-eter, messages may be used either to transfer real-time data orto manage the group membership. The field type may assumethe following values: {REMOVE, JOIN, ADD, UPDATE, HB,RT}. The msg.JOIN, msg.ADD, and msg. UPDATE messagesare used to add a station to the group; msg.REMOVE is usedfor removing a station; msg.HB is a “heartbeat message”; andmsg.RT are default RT data messages. The second assumptionspecifies that the logical ring has already been initialized withsome VTP-CSMA stations. Communication errors can occurdue to collisions and/or interferences.

A. Adding a Station to the Virtual Token Ring

A station willing to join the VTR will broadcast a msg.JOINmessage using the default EDCA mechanism. As a conse-quence, the message will be received by all the network stationsbut queued only by those belonging to group (all the otherstations will discard it). However, only one response has tobe issued to the requesting station. It will be provided bystation , via the msg.ADD message.2 Therefore, when

, the token holder station will allow the re-questing station to join the group through an acknowledgedpoint-to-point control message of type msg.ADD, containingthe following parameters: ACcurr (current value of the Ac-cess Counter); NAA (NA assigned to the requesting station:

MAC (MAC address of the requestingstation).

After receiving the msg.ADD, the requesting station (i.e., thestation that issued the msg.JOIN) sets

and enters the group . The MAC ad-dress of the requesting stations will be used if more than onejoin request arrives. In this case, station has to answerseparately to each station that issued the request. The value ofthe MAC address may be obtained from the msg.JOIN. Aftereach adding procedure, the membership can be represented as

.After processing all the received join requests, in order to

inform all the VTP-CSMA stations of the new entry, the tokenholding station ( broadcasts an unacknowledged messageof type msg.UPDATE, with the following parameters: ACcurr(current value of the Access Counter); NPcurr (current value ofthe number of stations).

2It is worth noting that, in an infrastructure network, station NA will be,most likely, the access point.



Consequently, all stations belonging to the group will up-date the value of their VTR parameters , if dif-ferent from the previous one; . Setting the valueof all the counters to ensures the re-synchronizationof potentially inconsistent values of the distributed vari-able (if any), when compared to station .

B. Removing a Station From the Virtual Token Ring

There are two possibilities for removing a station from thevirtual token ring (group ). In the first, the station decides au-tonomously to leave the group . In this case, it will commu-nicate its decision via a msg.REMOVE, whenever it receivesthe virtual token. In the second case, the station is compelledto leave the VTR, either due to a “crash failure” or because itbecomes unable to transfer its own VTP messages. As an im-mediate effect, there will be no more RT messages in the slotassigned to that station. In this situation, the unused NA addresswill be later recovered by an address reclaim procedure (basedon a “heartbeat” approach).

In order to autonomously leave the group , the stationholding the virtual token broadcasts an unacknowledgedmsg.REMOVE message, which has the following parameter:NAR (NA to be removed from the VTR).

As a consequence, all the VTP-CSMA stations that receivea msg.REMOVE will update their variables in the followingway: Stations with will update their NP

. Stations with will update both their NPand their . Finally,

stations with are considered to be out of the ring.The latter may occur only in the case of an inconsistent value

of the distributed variable. In such a case, the station mustconsider itself out of the ring. If it wants to re-enter the ring, inmust later “rejoin” it. As a further check, stations belonging togroup may verify the correctness of the image of their local

. Indeed, it should be for each station. Afterthe removal of a station, the membership of group can berepresented as .

A VTP-CSMA station may also leave the VTR without beingable to broadcast the msg.REMOVE message (station crash).Such an event has first to be detected and then the network ad-dress of the station (NA) must be recovered, in order to ensurethe consistency of the ring. The detection procedure works asfollows. In the VTR, each station knows the logical address ofits predecessor . Whenever a stationdetects that its predecessor has not sent any message during a

time interval, it concludes that the station has unexpect-edly exited the virtual ring. Thus, a station detecting such anexit invokes the address reclaiming procedure. In this case, itbroadcasts a msg.REMOVE message to remove its predecessorfrom group .

It is worth mentioning that, in order to avoid wrong removals,every station must transfer at least one message every .The technique adopted to avoid wrong removals uses a heartbeatapproach, where the . In practice, for a stationthat has not sent any message during the last interval,the next time it gets the virtual token, it will transfer either aheartbeat message (msg.HB) or a real-time message (msg.RT).

C. Validation of the VTP-CSMA Architecture

A particularity of the proposed ring management scheme thatmust be carefully checked is the use of unacknowledged broad-cast messages. The loss of the msg.REMOVE and/or msg.UP-DATE messages by some nodes may lead to the violation ofany, or even both, of the correctness properties (agreement anduniqueness). For example, if one of the VTP-CSMA stationsdoes not receive a msg.REMOVE message, then it will not up-date its image of the distributed variables . As aconsequence, there will be an inconsistency in the distributedvariables, and sooner or later, two RT-messages will collide.This collision will force both stations to consider themselvesout of the ring. This means that the structural behavior of theVTP-CSMA architecture is ensured by means of a self-removalmechanism that removes the two colliding stations. It is ex-pected that its impact upon the performance of the ring remainsnegligible (as just the inconsistent station plus one consistentstation are removed from the ring). That is, conversely to othertoken passing schemes, the ring does not need to be rebuilt fromscratch. On the other hand, as unacknowledged broadcast mes-sages are extensively used to manage the virtual ring, the latencyto add/remove a RT station is kept at a reduced value (about twotoken cycle times).

The proposed ring management procedure has to be carefullyassessed via an adequate performance analysis, in order to vali-date the effectiveness of the self-removal reset mechanism. Pre-liminary results of the performance analysis have already high-lighted the effectiveness of such “self-removal” approach uponthe original reset mechanism.

VII. CONCLUSION

The major motivation of this work was to “propose a so-lution enabling the support of real-time (RT) communicationin wireless IEEE 802.11 environments, where unconstrainedtiming devices would be able to coexist with real-time devices.”The VTP-CSMA approach targets this problem in fully meshedcommunication scenarios.

The obtained results clearly demonstrate that the proposedmechanism guarantees the highest transmitting probability toVTP-CSMA stations in a wireless environment, where the com-munication medium is shared with timing unconstrained trafficsources. More importantly, whatever the network load, both theaverage packet delay and the related average queue size arenearly constant for the RT traffic. This means that the uncon-strained network traffic has a negligible impact upon the timingbehavior of the RT traffic.

Additionally, the capability of the VTP-CSMA mechanismto handle a large amount of RT stations in error-free and error-prone scenarios has also been assessed. The obtained resultsclearly show that even for stringent scenarios, the VTP-CSMAmechanism is able to support up to 35 RT stations (in error-prone scenarios) or up to 70 stations (in error-free scenarios).

Finally, the use of an adequate virtual ring managementprocedure enables the support of dynamic communicationscenarios.



REFERENCES

[1] IEEE Standard for Information Technology—Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) Specifications,ANSI/IEEE Std. 802.11, 1999, Edition (R2003).

[2] IEEE Standard for Information Technology—Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) SpecificationsAmendment 8: MAC Quality of Service Enhancements, IEEE Std.802.11e-2005, 2005.

[3] A. Willig, K. Matheus, and A. Wolisz, “Wireless technology in indus-trial networks,” Proce. IEEE, vol. 93, no. 6, pp. 1130–1151, Jun. 2005.

[4] T. Sauter and F. Vasques, “Editorial special section on communicationin automation,” IEEE Trans. Ind. Informat., vol. 2, no. 2, pp. 73–77,May 2006.

[5] J.-D. Decotignie, “Ethernet-based real-time and industrial communica-tions,” Proc. IEEE, vol. 93, no. 6, pp. 1102–1117, Jun. 2005.

[6] S.-K. Kweon and K. Shin, “Achieving real-time communication overEthernet with adaptive traffic smoothing,” in Proc. 6th IEEE Real-TimeTechnology and Applications Symp., 2000, pp. 90–100.

[7] A. Carpenzano, R. Caponetto, L. Lo Bello, and O. Mirabella, “Fuzzytraffic smoothing: An approach for real-time communication over eth-ernet networks,” in Proc. 4th IEEE Int. Workshop Factory Communi-cation Systems, 2002, pp. 241–248.

[8] R. Moraes, F. Vasques, P. Portugal, and J. Fonseca, “Real-time com-munication in 802.11 networks: The virtual token passing VTP-CSMAapproach,” in Proc. 31st IEEE Conf. Local Computer Networks (LCN),2006, pp. 389–396.

[9] C. Casetti, C.-F. Chiasserini, M. Fiore, and M. Garetto, Notes on theinefficiency on 802.11e HCCA Dipartimento de Elettronica, Politec-nico di Torino. Torino, Italy, 2005.

[10] P. Garg, R. Doshi, R. Greene, M. Baker, M. Malek, and X. Cheng,“Using IEEE 802.11e MAC for QoS over wireless,” in Proc. IEEEInt. Performance, Computing, and Communications Conf., 2003, pp.537–542.

[11] J. Son, I.-G. Lee, H.-J. Yoo, and S.-C. Park, “An effective pollingscheme for IEEE 802.11e,” IEICE Trans. Commun., vol. E88-B, no.12, pp. 4690–4693, 2005.

[12] S.-C. Lo, G. Lee, and W.-T. Chen, “An efficient multipolling mecha-nism for IEEE 802.11 wireless LANs,” IEEE Trans. Comput., vol. 52,no. 6, pp. 764–768, Jun. 2003.

[13] S. Lee, K. N. Ha, J. H. Park, and K. C. Lee, “NDIS-based virtualpolling algorithm of IEEE 802.11b for guaranteeing the real-time re-quirements,” in Proc. Industrial Electronics Conf. (IECON), 2005, pp.2427–2432.

[14] A. Willig, “A MAC protocol and a scheduling approach as elements ofa lower layers architecture in wireless industrial LANs,” in Proc. IEEEInt. Workshop Factory Communication Systems (WFCS’97), 1997, pp.139–1488.

[15] D. Miorandi and S. Vitturi, “Analysis of master-slave protocols for real-time industrial communications over IEEE802.11 WLANs,” in Proc.2nd IEEE Int. Conf. Industrial Informatics, 2004, pp. 143–148.

[16] A. Hamidian and U. Korner, “An enhancement to the IEEE 802.11eEDCA providing QoS guarantees,” Telecommun. Syst., vol. 31, no. 2-3,pp. 195–212, 2006.

[17] M. Ergen, D. Lee, R. Sengupta, and P. Varaiya, “WTRP-wirelesstoken ring protocol,” IEEE Trans. Veh. Technol., vol. 53, no. 6, pp.1863–1881, Nov. 2004.

[18] R.-G. Cheng, C.-Y. Wang, L.-H. Liao, and J.-S. Yang, “Ripple: A wire-less token-passing protocol for multi-hop wireless mesh networks,”IEEE Commun. Lett., vol. 10, no. 2, pp. 123–125, Feb. 2006.

[19] J. Sobrinho and A. Krishnakumar, “Quality-of-service in ad hoc carriersense multiple access wireless networks,” IEEE J. Sel. Areas Commun.,vol. 17, no. 8, pp. 1353–1368, Aug. 1999.

[20] J. L. Sobrinho and A. Krishnakumar, “EQuB—Ethernet quality of ser-vice using black bursts,” in Proc. Conf. Local Computer Networks,1998, pp. 286–296.

[21] G.-H. Hwang and D.-H. Cho, “New access scheme for VoIP packets inIEEE 802.11e wireless LANs,” IEEE Commun. Lett., vol. 9, no. 7, pp.667–66, Jul. 2005.

[22] R. Moraes, P. Portugal, and F. Vasques, “A stochastic petri net modelfor the simulation analysis of the IEEE 802.11e EDCA communicationprotocol,” in Proc. 10th IEEE Int. Conf. Emerging Technologies andFactory Automation (ETFA), 2006, pp. 38–45.

[23] A. Willig, “Redundancy concepts to increase transmission reliabilityin wireless industrial LANs,” IEEE Trans. Ind. Informat., vol. 1, no. 3,pp. 173–182, Aug. 2005.

[24] A. Willig and A. Wolisz, “Ring stability of the PROFIBUS token-passing protocol over error-prone links,” IEEE Trans. Ind. Electron.,vol. 48, no. 5, pp. 1025–1033, Oct. 2001.

[25] M. Hamdaoui and P. Ramanathan, “A dynamic priority assign-ment technique for streams with (m,k)-firm deadlines,” IEEE Trans.Comput., vol. 44, no. 12, pp. 443–451, Dec. 1995.

[26] B. Goode, “Voice over Internet Protocol (VoIP),” Proc. IEEE, vol. 90,no. 9, pp. 1495–1517, Sep. 2002.

Ricardo Moraes received the B.Sc. degree in math-ematics from University of Planalto Catarinense,Lages, Brazil, in 1997, and the M.Sc. degree incomputer science from Federal University of SantaCatarina, Florianopolis, Brazil, in 1999. Currently,he is pursuing the Ph.D. degree at the University ofPorto, Porto, Portugal.

His research interests include real-time communi-cation systems and real-time system architectures.

Francisco Vasques (M’00) received the licen-ciatura degree in electrical engineering from theUniversity of Porto, Porto, Portugal, in 1987 and theM.Sc. and Ph.D. degrees in computer science fromLAAS-CNRS, Toulouse, France, in 1992 and 1996,respectively.

Since January 2004, he has been an Associate Pro-fessor in the Mechanical Engineering Department ofthe University of Porto. He is author or coauthor ofmore than 100 technical papers in the areas of real-time systems and industrial communication systems.

Dr. Vasques is an Associate Editor of the IEEE TRANSACTIONS ON

INDUSTRIAL INFORMATICS for the topic of industrial communications since2007.

Paulo Portugal received the licenciatura, M.Sc., andPh.D. degrees in electrical and computer engineeringfrom the University of Porto, Porto, Portugal, in1992, 1995, and 2005, respectively.

Currently, he is an Assistant Professor in the De-partment of Electrical and Computer Engineering ofthe University of Porto. His research interests includedependability and performability modeling and eval-uation, and industrial communication systems.

José Alberto Fonseca (M’00) received the licen-ciatura degree in electronics and telecommunicationsengineering and the Ph.D. degree in electrical en-gineering from the University of Aveiro, Aveiro,Portugal, in 1980 and 1992, respectively.

He has been an Associate Professor in the Elec-tronics and Telecommunications Department of theUniversity of Aveiro since 2000. His current researchinterests are embedded systems, distributed systems,and industrial communications.


Date post:	09-Feb-2023
Category:	Documents
Upload:	up-pt
View:	0 times
Download:	0 times

VTP-CSMA: A Virtual Token Passing Approach for Real-Time Communication in IEEE 802.11 Wireless...

Documents