Real-Time Communication in IEEE 802.11 Networks: Timing Analysis and aRing Management Scheme for the VTP-CSMA Architecture

Ricardo Moraes�, Paulo Portugal�, Stefano Vitturi�, Francisco Vasques�, Pedro F. Souto�

� Faculdade de Engenharia, University of Porto, Portugal� IEIIT-CNR, University of Padova, Padova, Italy

{rmoraes, pportugal, vasques, pfs}@fe.up.pt, vitturi@dei.unipd.it

Abstract

Keeping up with the timing constraints of real-time trafficin wireless environments is a hard task. One of the reasonsis that the real-time stations have to share the same commu-nication medium with stations that are out of the sphere-of-control of the real-time architecture. That is, with stationsthat generate timing unconstrained traffic. The VTP-CSMAarchitecture targets this problem in IEEE 802.11 wirelessnetworks. It is based on a Virtual Token Passing procedure(VTP) that circulates a virtual token among real-time sta-tions, enabling the coexistence of real-time and non real-time stations in a shared communication environment. Theworst-case timing analysis of the VTP-CSMA mechanismshows that the token rotation time is upper-bounded, evenwhen the communication medium is shared with timing un-constrained stations. Additionally, the simulation analy-sis shows that the token rotation mechanism behaves ade-quately, even in the presence of error-prone communicationchannels. Therefore, the VTP-CSMA architecture enablesthe support of real-time communication in shared communi-cation environments, without the need to control the timingbehavior of every communicating device. A ring manage-ment procedure for the VTP-CSMA architecture is also pro-posed, allowing real-time stations to adequately join/leavethe virtual ring. This ring management procedure is manda-tory for dynamic operating scenarios, such as those foundin VoIP applications.

1. Introduction

Presently, the IEEE 802.11 family of protocols is oneof the most used set of Wireless Local Area Networks(WLANs). It was standardized in 1999 by the IEEE, asthe IEEE 802.11 standard, which was later reaffirmed in2003 [2]. Recently, the IEEE 802.11e [3] standard was pub-lished as an amendment to the original version. This amend-

ment is intended to provide differentiated levels of Qualityof Service (QoS) to the supported applications, includingthe transport of both voice and video over WLANs.

The IEEE 802.11 architecture provides a wireless LANthat supports station mobility transparently to the upper lay-ers. The basic service set (BSS) is the building block of anIEEE 802.11 WLAN, which actually provides two types ofconfigurations: independent BSS (IBSS) and infrastructure.The IBSS is the most basic type for a IEEE 802.11 WLAN,which may be composed of, at minimum, two stations. Thismode of operation is often referred as ad hoc. The infras-tructure mode includes one or more access point (AP) [2]that convey the communication among stations. It is possi-ble to create a wireless network of arbitrary size and com-plexity, where a number of BSSs may be interconnected,appearing as a single BSS at the logical link control (LLC)[1] layer. The IEEE 802.11 standard refers to this type ofnetwork as the Extended Service Set (ESS) network.

Currently, there is a trend for the implementation of real-time (RT) communication systems on top of wireless net-works [13]. When supporting RT communication, an im-portant assumption that must be considered is that the wire-less communication medium is essentially an open commu-nication environment. That is, any new participant can try toaccess the communication medium at any instant (accord-ing to the MAC rules) and establish its own communicationchannels. As a consequence, the network load cannot bepredicted at system setup time, nor it can be effectively con-trolled during the system run-time. Therefore, a RT com-munication protocol must be able to guarantee the timingconstraints of the RT traffic in a communication environ-ment shared with timing unconstrained traffic.

Traditionally, the RT communication behavior in wiredCSMA environments is guaranteed through the tight con-trol of every communicating device [5]. The coexistence ofRT controlled stations with timing unconstrained stationsis made possible by constraining the traffic behavior of thelatter. For instance, using traffic smoothers [7, 8]. Un-fortunately, this approach is not adequate for wireless en-

32nd IEEE Conference on Local Computer Networks

0742-1303/07 $25.00 © 2007 IEEEDOI 10.1109/LCN.2007.74

107

vironments, since it is not possible to impose any trafficsmoothing strategy upon stations that are out of the sphere-of-control of the RT architecture. An interesting survey ofreal-time communication techniques for wireless networkscan be found in [13].

The VTP-CSMA architecture [10] has been proposed todeal with this problem in ad hoc networks. In [10] it hasbeen shown by simulation that the VTP-CSMA architec-ture was able to handle RT communication in shared com-munication environments. In this paper, the VTP-CSMAarchitecture is extended to infrastructure networks, where itenables the scheduling of upload RT traffic. The worst-casetiming analysis presented in Section 3 demonstrates that thetoken rotation time is upper-bounded, even in the presenceof timing unconstrained traffic. Therefore, it shows that theVTP-CSMA architecture is able to provide a RT communi-cation service, when the communication medium is sharedwith timing unconstrained stations (ST stations). The per-formance analysis represented in Section 4 illustrates theadequate behavior of the VTP-CSMA mechanism in thepresence of error-prone communication channels. One ofthe major limitations of the earlier version of the VTP-CSMA mechanism [10] was that it just considered a staticcommunication environment, with a fixed number of RTstations. In this paper (section 5) a ring management pro-cedure is presented, which allows stations to dynamicallyleave or join (rejoin) the VTP-CSMA architecture. Thisprocedure enables the support of highly dynamic communi-cation scenarios with multiple devices opening/closing RTconnections, such as those found in VoIP (voice over IP)applications.

2 The VTP-CSMA Architecture

The VTP-CSMA architecture is based on the control ofthe medium access right, by means of a Virtual Token Pass-ing (VTP) procedure among RT stations, complementedby a traffic separation mechanism which guarantees that,whenever a RT station is contending for the medium access,it will win the contention prior to any other ST station.

The underlying traffic separation mechanism works asfollows: whenever a collision between one RT station anda set of ST stations occurs, all the involved stations, ex-cept the RT one, will use the prioritized medium accessmechanism (EDCA) and select a random backoff intervalaccording to the access category (voice, video, best-effortand background). Conversely, the RT station (hereafteralso referred as a VTP-CSMA station) transfers its trafficat the highest access category, using the highest prioritylevel of the EDCA mode, i.e., setting the Arbitration In-terframe Space (AIFS) to AIFS[V O] = aSIFSTime +2 × aSlotT ime and the Contention Window (CW) to thevalue aCWmin[V O] = aCWmax[V O] = 0

This means that, any VTP-CSMA station will always tryto transmit its frame in the first EDCA available slot, whileall the other ST stations will wait during a time intervalevaluated by the local backoff functions. Nevertheless, iftwo or more VTP-CSMA stations simultaneously contendfor the medium access, they would collide and eventuallydiscard the frame (after the maximum number of retrans-mission attempts). This behavior is overcome by means ofa Virtual Token Passing (VTP) procedure that serializes theVTP-CSMA stations.

The VTP procedure considers a process group G withnp members. The membership is represented as L = {NA1,NA2,..., NAnp}. The notion NAi denotes the i-th station in Gand is also used as station identification (ID) for NAi. Theprocedure circulates a virtual token in L. Specifically, allmembers of group G maintain an Access Counter (ACo).The generic i-th VTP-CSMA station captures the virtualtoken when ACo equals NAi. If the station has queuedmessages, then it will immediately transfer them duringa time interval upper bounded by the transmission oppor-tunity period (TXOP). The underlying traffic separationmechanism guarantees that the VTP-CSMA station willwin the medium access contention. At the end of the cur-rent TXOP, each VTP-CSMA station will increase its ACovalue, passing the virtual token to the next station (stationwith NAi=ACo+1)1. Whenever the VTP-CSMA stationholding the token does not have any RT message to trans-fer, it will allow default stations to contend for the mediumaccess, during a time interval t2 = aSIFSTime + 3 ×aSlotT ime. As the EDCA mechanism allows any stationto start a transmission after aSIFSTime+2×aSlotT ime,it enables the coexistence of RT stations with non-real-timeST stations in the shared communication environments.

Figure 1 illustrates the VTP-CSMA mechanism, that isrepresented by four procedures: Initialization, Main, Trans-mission and Listening. According to the Initialization pro-cedure (line 2, Fig. 1), the ACo value is set to NA1 in allVTP-CSMA stations. Three variables (t1, t2, t3) of typeinteger are defined, where t1 and t2 are slot time coun-ters, whereas t3 is a special collision counter used for re-initialization purposes. The Main procedure is executed atthe beginning of each time slot, where every VTP-CSMAstation firstly verifies if the special collision counter (t3)2

1The ACo must be increased by “mod” operation, i.e. ACo =(ACo mod np) + 1. Therefore, when it is referred that the ACo is in-creased, it is executed a “mod” operation.

2When a collision resolution starts, it can be a consequence of threedifferent types of collisions: (1) among ST stations; (2) among ST sta-tions and the active VTP-CSMA station that holds the token; (3) amongVTP-CSMA stations. The first scenario can be detected if the communica-tion medium remains idle during the time interval t2. The second scenariois easily solved in favor of the VTP-CSMA station. Finally, when multi-ple VTP-CSMA stations simultaneously contend for the medium access,it means that the ACo counters lost their synchronization, forcing the re-initialization of the ring.

108

exceeds the maximum number of retransmission attempts(RN ). Whenever this number is exceeded, the Initializa-tion procedure is executed (reset mechanism). Afterwards,depending on both the channel event during the last slot andthe ACo value, each VTP-CSMA station will take a specificaction.

The Transmission procedure is only executed by theVTP-CSMA station holding the token (ACo = NAi)and, it is only initiated after the medium being idle duringAIFS[V O], i.e. the minimum AIFS value for the EDCAmode, as defined by IEEE 802.11e. Conversely, the Listen-ing procedure may be initiated after the medium being idleduring SIFS (aSIFSTime).

1 ∀ process NAi

2 Initialization: ACo ← NA1; t1 ← 0; t 2← 0; t 3← 0;3 Main: while t3 ≤ RN do4 if ACo = NAi then5 {wait AIFS[V O]};6 NAi executes Transmission;7 else8 {wait SIFS};9 NAi executes Listening;10 endif11 end while

Figure 1. VTP-CSMA mechanism.

Five channel states are defined (determined at the begin-ning of each time slot):

1. transmission from other stations: One or more mes-sages are being transmitted over the channel.

2. successful transmission from other stations: Thechannel is idle, and a successful message transmissionfrom other station finished one time slot ago.

3. channel continuing idle: The channel is idle and wasalso idle one time slot ago.

4. channel idle after collision: The channel is idle,and there was a collision one time slot ago.

5. successful transmission: The channel is idle, “I amthe transmitting station, and I finished the transmissionof one or more messages (upper bounded by TXOPinterval)” one time slot ago.

According to these channel states and the ACo value,each VTP-CSMA station takes a specific action. Firstly,whenever the VTP-CSMA captures the virtual token(ACo = NAi), it will execute the Transmission proce-dure (Figure 2). This procedure works as follows. If theVTP-CSMA station holding the token have a RT messageto transfer, it will immediately start the transmission. If a

collision occurs, the VTP-CSMA station increments its t3counter and, it will retry the transmission until the maxi-mum defined number of transmission attempts. Whenever,a successful transmission occurs, the VTP-CSMA stationholding the token will execute the Listening procedure (Fig.2, lines 4-5), where each VTP-CSMA station will increaseits ACo value, passing the virtual token to the next station.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 atime interval multiple of aSlotT ime.

1 Transmission:2 if NAi has message to be transmitted then3 start the transmission; wait for transmission to complete;4 if successful transmission then5 t1← 1; t3← 0; go to Listening;6 else;7 t3++;8 if t3 ≤ RN then9 go to Transmission;10 endif11 endif12 else13 t2← 2; go to Listening;14 endif

Figure 2. Transmission procedure.

As illustrated in Figure 1, all VTP-CSMA stationsthat do not have the token (ACo �= NAi) will ex-ecute the Listening procedure (Figure 3), and depend-ing on the channel state, these VTP-CSMA stations takea specific action: a) transmission from other stations(lines 5-7): All VTP-CSMA stations wait for the end oftransmission and then update the variables t1 and t2; b)successful transmission from other stations (lines 8-9):All VTP-CSMA stations update t1, t2 and t3; c) channelcontinuing idle (lines 10-17): All VTP-CSMA stationsincrement t2 and verify the value of t1. If t1 = 1, all VTP-CSMA stations increment its ACo value. That is, t1 = 1and the channel continuing idle state means that a suc-cessful transmission occurred and a TXOP period finished.Besides, t2 will be incremented and, each time that t2 valueis greater or equal than 3, all VTP-CSMA stations must alsoincrement its ACo value; d) channel idle after collision(lines 18-19): All VTP-CSMA stations increment t2 and t3.

Most likely, whenever a collision occurs involvingjust ST stations, the next events will be followed bycontinuing idle channel states, due to the selected back-off interval (CWmin = 7, for voice category). How-ever, when a VTP-CSMA station is involved in the colli-sion, the next channel state will be always a transmission(CWmin = CWmax = 0). Therefore, the reset mecha-nism (Initialization procedure) is activated whenever there

109

1 Listening:2 At the beginning of the next slot3 event ← channel event during the last slot;4 switch (event)5 case transmission:6 wait for transmission to complete;7 t1 ← 0; t 2← 0; break;8 case successful transmission:9 t1 ← 1; t2 ← 1; t3 ← 0; break;10 case continuing idle:11 t2++;12 if t1 = 1 then13 ACo++; t1 ← 0; t2 ← 0; break;14 endif15 if t2 ≥ 3 then16 ACo++; t1← 0; t2← 0; t3← 0; break;17 endif;18 case idle after collision:19 t2++; t3++; break;20 end switch

Figure 3. Listening procedure.

are RN consecutive collisions without any idle time greaterthan t2 (aSIFSTime+3×aSlotT ime) among collisions.

3 Timing Analysis

In this section, a worst-case timing analysis of theVTP-CSMA mechanism is presented, which demonstratesthat the token rotation time is upper-bounded, even whenthe communication medium is shared with timing uncon-strained stations. This means that the non real-time mes-sages are not able to disturb the timing operation of the ring.

Consider an IEEE 802.11e network interconnecting npVTP-CSMA stations with multiple IEEE 802.11e stations(ST-stations). Consider that the VTP-CSMA stations haveaddresses ranging from 1 to np. Each VTP-CSMA stationaccesses the network according to the VTP-CSMA scheme,i.e., first station 1, then station 2, 3,. . . until station np, andthen again station 1, 2,. . . np. The default (ST) stations im-plement the traditional backoff procedure according to thedefault timing values defined in [3].

Basically, two-collision scenarios are analyzed. Firstly,it is analyzed the maximum delay to transfer a real-timemessage, when the VTP-CSMA station is holding the to-ken (Figure 4). According to the VTP-CSMA scheme,whenever a VTP-CSMA station holding the token has adata message ready to be transferred (D), it will wait anIFS (Interframe Space) before starting to transmit it (1stattempt). A station is able to detect a collision only af-ter finishing its transmission plus an aSIFSTime, i.e. if theACK frame is not received. Besides, when a transmissionstarts, all the stations set their NAV (Network AllocationVector) with the information received in the Duration/ID

field, that goes up to the end of the expected ACK frame [3].Afterwards, if the transmission is not correctly acknowl-edged, the station will wait again during another IFS (IFS:aSIFSTime + 2 × aSlotT ime) interval and, accordingto the VTP-CSMA architecture, it will immediately start totransmit its message (2nd attempt). If a second collision oc-curs, the station will wait again for the IFS before startingto transmit. The maximum time that a VTP-CSMA stationholding the token will wait before starting to transfer a mes-sage for the last attempt or eventually discard it, is given by:

Tcol = (RN − 1) × (IFS + tmessage) (1)

where RN is the maximum number of retransmission andtmessage is the duration to transfer a data message (includ-ing the ACK frame) from the VTP-CSMA station accordingto the physical (PHY) characteristics of the channel. Forinstance, considering 100 bytes for data payload in IEEE802.11a PHY mode (data rate of 36 Mbps), each attempttakes 0.128ms.

Figure 4. Collision scenario.

It is clear that the VTP-CSMA architecture either solvesthe collisions in a bounded time interval (within RN retrans-mission attempts), or it eventually discards the message.Therefore, a relevant focus of research is to evaluate theprobability of a message frame being discarded by the IEEE802.11 stations, whenever the number of collision resolu-tion rounds exceeds the defined number of retransmissionattempts. This topic has been addressed in [15], where ithas been shown that for a non saturated network, the packetloss rate is smaller than 10−7, when considering a collisionprobability p = 0.1. This satisfies the packet loss require-ments of traditional soft real-time applications, such as VoIP(voice over IP) or NCS (networked control systems) appli-cations.

Therefore, in the following analysis it is only consideredthe case where no message is discarded by a VTP-CSMAstation, due to an excessive retransmission number of at-tempts. That is, when the transmitting station acquires thetransmission medium, the virtual token will be passed tothe following station at the end of the acquired TXOP (inthe worst-case).

Then, the worst-case token holding time (TTHRT) occurs

when the RT station has enough RT messages to fill up to

110

the maximum TXOP interval defined for the subset of RTstations (tTXOPRT

). The value of TTHRTis given by:

TTHRT= Tcol + tTXOPRT

(2)

On the other hand, when the VTP-CSMA station hold-ing the token does not have any real-time message ready tobe transferred, one of the ST stations in the wireless domaincan successfully transfer its own messages. This means thatall the VTP-CSMA stations will wait during a time intervalt2, during which any ST station may start to transfer a mes-sage. In such a case, the ACo counters will be incrementedonly at the end of the acquired TXOP. If the medium re-mains idle during the t2 time interval, the ACo counterswill be immediately incremented (i.e., there will be a Vir-tual Token Passing), allowing the next RT station in the ringto transfer its messages.

The maximum time interval that a VTP-CSMA stationis allowed to hold the token, when it does not have anyreal-time message ready to be transferred (TTHNRT ), oc-curs when at the specific instant (aSIFSTime + 2 ×aSlotT ime) a ST station takes the decision of transmit aframe and it acquires the transmission medium and uses allthe allowed TXOP time. The value of TTHNRT

is given by:

TTHNRT = IFS + tTXOP (3)

Therefore, the worst-case token holding time TTH isgiven by:

TTH = max(TTHRT, TTHNRT

) (4)

It can be easily proved that TTHRT > TTHNRT , whenboth RT and ST stations have the same tTXOP . How-ever, most likely the value of tTXOPRT

for the RT stationswill be set to a much smaller value than the default TXOP(1.504ms, for voice category) [3]. For example, consid-ering that tTXOPRT = 0.2ms and the maximum numberof transmission attempts RN = 4, for the above describedcase (100 bytes for data payload and data rate of 36 Mbps),TTHRT

= 0.584ms, while TTHNRT= 1.547ms. There-

fore, it is expected that for most of the VTP-CSMA appli-cations, the value of TTHNRT

will prevail.Considering the token rotation time as the time interval

between two consecutive token arrivals to a particular sta-tion, the worst-case token rotation time (TRT ) is given by:

TRT = np × TTH (5)

The TRT value imposes a lower bound for the periodic-ity of the real-time message streams supported by the VTP-CSMA architecture. The study of the problem of guaran-teeing synchronous message deadlines in the VTP-CSMAarchitecture can be easily adapted from [4], where a similarstudy was done for FDDI networks.

4 Performance Analysis

The performance analysis presented in this paper aims toverify how the VTP-CSMA architecture is able to cope withthe requirements of real-time communications [11], both inthe presence of external perturbations and in the presence oferror-prone channels. By external perturbations, we meanexternal traffic sources, that are out of the sphere-of-controlof the RT architecture. Specifically, it is of strong interestto assess the average queue size that represents the aver-age output buffer occupancy and, the impact upon the tokenrotation time, since it bounds the shortest update time ofperiodic variables. Additionally, the average transmissiondelay is also assessed, as it must be kept smaller than thedeadlines of the real-time message streams.

The VTP-CSMA simulation model is implemented us-ing a Stochastic Petri Net (SPN) model, previously de-veloped to evaluate the performance of the standard IEEE802.11e EDCA function [9]. The description of the modelis out of the scope of this paper. Nevertheless, its completedocumentation is available at the website of this project3.For all the simulations, it has been used a Semi-Markov er-ror model, where the channel is always in one of two states:Good or Bad. This model assume that bit errors are inde-pendent, with a fixed error rate. For the parametrizationof the Semi-markov model, it has been used the values de-fined in [12], i.e., for all channels the mean duration of goodstate is 65ms, the mean duration of bad state is 10ms and,the coefficient of variation (CoV) for the bad state holdingtimes has been set to 10 and for the good state to 20. Thismeans that burst lengths lead to a rather bad channel [14],where the steady-state probability for finding the channel inbad state is approximately 13.3%. Two sets of simulationsare assessed, differing in their respective mean bit error rate(BER). The first set defines a mean BER of 10−4, while thesecond set defines that no bit errors occur. Therefore, dur-ing the bad channel states for the first set, the BER is about0.00075 and, for the good state no bit errors will occur.

The simulation scenario considers an open communi-cation environment, where multiple ST stations share thesame communication medium with a subset of RT stationsimplementing the VTP-CSMA mechanism. The RT sta-tions transfer just RT traffic, whereas the ST stations trans-fer three types of traffic: voice (VO), video (VI) and back-ground (BK). Basically two simulation cases are analyzed.The first scenario (small population case) considers 10 STstations operating in the same frequency band together with10 to 50 RT VTP-CSMA stations. The second scenario(large population case) extends the number of ST stationsto 40. Each station operates at OFDM PHY mode and thePHY data rate is set to 36 Mbps. The physical parametersused in the simulations are the default values of the IEEE

3http://www.fe.up.pt/∼vasques/ieee80211e/

111

802.11a PHY mode [2]. In both simulation cases, each STstation generates λ voice, video and background packets/sat the same rate, in order to impose a network traffic loadranging from 5% to 95%.

The RT traffic is characterized by periodic traffic sourceswith a small amount of jitter (normal distribution withσ/µ ≤ 1%). The ST stations have Poisson traffic sources.The maximum number of transmission attempts is set to 4.The MAC queue size is set to 50 positions. For the proposedsimulations is assumed that there is no node mobility, norhidden stations. All other relevant simulation parametersare shown in Table 1.

Table 1. Simulation data.Parameters VTP-CSMA

ST stationsVO VI BK

CWmin 0 7 15 31CWmax 0 15 31 1023AIFSN 2 2 3 7

TXOP (ms) 1.504 1.504 3.008 0Packet Size - bytes 45 160 1280 1600

Interarrival time (ms) 10, 20 Variable Variable Variable

4.1 Simulation Results

All the simulation results have been obtained using theSPN simulation model, with 95% confidence interval and amaximum half-width interval of 5%. The target of the simu-lation scenarios is to assess the behavior of the VTP-CSMAarchitecture, when a set of RT stations is transferring real-time data (45 bytes for data payload) with message streamperiods (MSP) of 10ms and 20ms in the same frequencyband of 10(40) timing unconstrained ST stations, in eithererror-free and error-prone channels.

A. The impact of timing unconstrained traffic upon thetoken rotation time.

An important parameter that must be carefully evaluatedis the token rotation time, as the ring stability cannot be af-fected by the external perturbations. We have assessed thebehavior of the token rotation time for different real-timeconfigurations (10, 20, 30, 40 and 50 RT stations). Fig-ure 5 illustrates the impact of external perturbations uponthe token rotation time, when supporting real-time messagestreams with periods of 10ms in the large population sce-nario (40 ST). For the sake of cleanness is only plotted theresults for 10, 20 and 50 RT stations. Similar results wereobtained in the small population scenario.

As expected, the token rotation time have a reducedvalue while the network load is kept at a small value, evenwhen considering error-prone channels (dashed lines). Themain reason is that the medium remains idle during longtime intervals, and therefore each RT station will capture

5 15 25 35 45 55 65 75 85 9510

−1

100

101

102

103

network load imposed by external (ST) stations (%)

toke

n ro

tatio

n tim

e (m

s)

10 RT

20 RT

50 RT

upper−bound − 10 RT stations

upper−bound − 20 RT stations

upper−bound − 50 RT stations

Figure 5. Token rotation time - 10 ms.

the token almost always after t2 (aSIFSTime + 3 ×aSlotT ime). This means that the probability of having aST station contending for the medium access is small. Con-versely, when the network load increases, whenever a RTstation does not have any RT message to transfer, it is ex-pected that one ST station will be able to capture the trans-mission medium during one TXOP time interval.

In Figure 5, it is also plotted the upper-bound for theTRT obtained from Equation 5. It must be considered thatthis upper-bound addresses a rarely occurring case, as it isbased on the assumption that during a token rotation cycle,none of the RT stations had RT-messages to transfer, andthat every ST-station transferred messages up to the maxi-mum allowed TXOP (1.504ms for voice category).

From the simulation results, it can be concluded thatthe VTP-CSMA architecture guarantees a stable ring opera-tion for up to 40 RT stations (supporting real-time messagestreams of 10ms and 20ms), as the token rotation time iskept under the MSP (10ms or 20ms), whatever the timingunconstrained traffic load.

B. The impact of timing unconstrained traffic upon theaverage queue size.

A second simulation analysis concerns the assessment ofthe average queue size in a RT station, when the traffic inthe wireless domain is disturbed by the presence of timingunconstrained traffic from ST stations and from physical in-terferences (error-prone channel). Figure 6 shows the aver-age queue size for MSP of 10ms in both error-free (solidlines) and error-prone (dashed lines) channels, where it isplotted the cases of 10, 20 or 40 RT stations operating inthe small population scenario (10 ST stations). It is clearin Figure 6 that the average number of packets waiting to

112

be transmitted is kept under 1 packet. This indicates thatthe VTP-CSMA mechanism can be suitable to support real-time traffic with MSP of 10ms, as the pending RT messagesare always transferred (in average) before the generation ofnew RT messages. Similar results were obtained for MSPof 20ms in both small and large population scenarios.

5 15 25 35 45 55 65 75 85 9510

−1

100

101

network load imposed by external (ST) stations (%)

aver

age

queu

e si

ze (p

acke

ts)

10 RT

20 RT 40 RT

Figure 6. Average queue size - 10 ms.

C. The impact of timing unconstrained traffic upon theaverage packet delay.

Finally, the average packet delay for RT messages isplotted in Figure 7, for the large population scenario. Forthe sake of simplicity, only the values for 10, 20 and 40RT stations are plotted in the figure. From Figure 7, itcan be seen that the average packet delay remains smallerthan the period of the related message stream, except for 40RT stations where it is slightly above 10ms. However, itis worth noting that the set of simulation experiments con-sider a highly pessimist scenario for the channel error pa-rameters. For the simulated scenarios, it has been used aBER = 7.5 × 10−4 and a mean duration for the Bad stateof 10ms. Therefore, the successful transmission probabilityto transfer a RT packet during the error periods (10ms) isjust 60% (such probability is equal to (1 − BER)n, wheren is the total length (bits) of a RT packet).

Therefore, the VTP-CSMA architecture is able to pro-vide an adequate RT communication service to soft real-time applications, such as VoIP or NCS applications. Voice(and video) conversation need to be conducted with the min-imum of delay in transmission. Many VoIP systems usemessage stream periods (MSP) of 20ms and, an averagepacket delay below 150ms is acceptable for most user ap-plications [6]. Besides, it is recommended a packet loss rate

< 2% and average jitter < 50ms, what is easily reached bythe VTP-CSMA architecture.

5 15 25 35 45 55 65 75 85 9510

−1

100

101

102

103

network load imposed by external (ST) stations (%)

aver

age

pack

et d

elay

(ms)

10 RT

20 RT

40 RT

Figure 7. Average packet delay - 10 ms.

5 Virtual Ring Management

The original version of the VTP-CSMA mechanism [10]was able to handle only a fixed number of stations (np sta-tions). In this paper, an enhanced ring management proce-dure is proposed, allowing the VTP-CSMA architecture tobe an open group. Thus, a station can dynamically join orleave the Virtual Token Ring (group G), enabling the sup-port of dynamic communication scenarios, such as thosethat are usually found in VoIP applications. The ring man-agement includes procedures to (i) add RT-stations to thering, (ii) remove RT-stations from the ring. These proce-dures must ensure the two following properties. Agree-ment: All VTP-CSMA stations must agree on the valuesof ACo and np. That is, at whatever instant of time, allstations know the address of the token holder and the totalnumber of stations belonging to the group G. Uniqueness:Each station must be assigned an unique NA, which rangesbetween 1 and np.

Unless these two properties are satisfied, mutual accessto the medium by the RT-stations cannot be ensured. Be-sides these properties, some further assumptions are maderegarding the capabilities of the communication system.

The first assumption is that stations belonging to groupG are able to exchange messages. Each message (msg)contains the fields sender and type; sender identifies thesource, whereas type is concerned with the function of themessage itself. Depending on the value assigned to thetype parameter, messages may be used either to transfer

113

real-time data or to manage the group membership. Thefield type may assume the following values: {REMOVE,JOIN, ADD, UPDATE, HB, RT}. The msg.JOIN, msg.ADDand msg.UPDATE messages are used to add a station tothe group; msg.REMOVE is used for removing a station;msg.HB is used to send a “heartbeat message” and msg.RTare default RT data frames.

The second assumption specifies that the logical ring hasalready been initialized with some VTP-CSMA stations. Fi-nally, it is desirable that all VTP-CSMA stations remain inthe range of the Access Point (AP); conversely, for ad hocnetworks, it is desirable that stations can always hear eachother. However, both the hidden and exposed scenarios areconsidered in the analysis. Communication errors can occurdue to collisions and/or interferences.

5.1 Adding a Station to the Virtual TokenRing (VTR)

A station willing to join the VTR, will broadcast amsg.JOIN message using the default EDCA mechanism.As a consequence, the message will be received by all thenetwork stations, but queued only by those belonging togroup G (all the other stations will discard it). However,only one response has to be issued to the requesting station.It will be provided by station NA1, via the msg.ADDmessage4. Therefore, when ACo=NA1, the token holderstation will allow the requesting station to join the groupthrough an acknowledged point-to-point control messageof type msg.ADD, with the following parameters: ACcurr� integer � (current value of the Access Counter); NAA� integer � (NA assigned to the requesting station):NAA← NP+1; MAC � HEX � (MAC address of therequesting station).

Then, the requesting station (i.e. the station that issuedthe msg.JOIN) sets ACo←ACcurr, NA←NAA, NP←NAAand enters the group G. The MAC address of the request-ing stations will be used if more than one join request ar-rives. In this case, station NA1 has to answer separatelyto each station that issued the request. The value of theMAC address may be obtained from the msg.JOIN. Aftereach Adding procedure, the membership can be representedas L = {NA1, NA2, ..., NAnp, NAnp+1}.

After processing all the received join requests, in orderto inform all the VTP-CSMA stations of the new entry, thetoken holding station (NA1) broadcasts an unacknowledgedmessage of type msg.UPDATE, with the following param-eters: ACcurr � integer � (current value of the AccessCounter); NPcurr � integer � (current value of the num-ber of stations).

4It is worth noting that, in an infrastructure network, station NA1 willbe, most likely, the Access Point.

Consequently, all stations belonging to the group G willupdate the value of their VTR parameters (ACo← NA1,if different from the previous one; NP←NPcurr). Set-ting the value of all the ACo counters to NA1 ensures there-synchronization of potentially inconsistent values of theACo distributed variable (if any), when compared to stationNA1.

5.2 Removing a Station from the VirtualToken Ring

There are two possibilities for removing a station fromthe virtual token ring (group G). In the first, the stationdecides autonomously to leave the group G. In this case,it will communicate its decision via a remove message(msg.REMOVE), whenever it receives the virtual token. Inthe second, the station is compelled to leave the VTR, ei-ther due to a “crash failure” or because it becomes unableto transfer its own VTP messages. As an immediate effect,there will be no more RT messages in the slot assigned tothat station. In such a case, the unused NA address will belater recovered by an address reclaim procedure (based ona “heartbeat” approach).

In order to autonomously leave the group G, the stationholding the virtual token broadcasts an unacknowledgedmsg.REMOVE message, which has the following parame-ters: NAR � integer � (NA to be removed).

As a consequence, all the VTP-CSMA stations that re-ceive a msg.REMOVE will update their variables in the fol-lowing way. Stations with NAi < NAR will update their NP(NP←NP-1). Stations with NAi > NAR will update boththeir NP (NP←NP-1) and their NAi (NAi ←NAi-1). Fi-nally, stations with NA = NAR are considered to be out ofthe ring.

The latter may occur only in the case of an inconsistentvalue of the ACo distributed variable. In such a case, thestation must consider itself out of the ring. If it wants to re-enter the ring, in must later “rejoin” it. As a further check,stations belonging to group G may verify the correctness oftheir ACo values. Indeed, it should be NAR=ACo for eachstation. After the removal of a station, the membership ofgroup G can be represented as L = {NA1, NA2,..., NAnp−1}.

A VTP-CSMA station, may also leave the VTR withoutbeing able to broadcast the msg.REMOVE message (stationcrash). This event has first to be detected and then the net-work address of the station (NA) must be recovered (via theaddress reclaiming procedure), in order to ensure the con-sistency of the ring. The detection procedure works as fol-lows. In the VTR, each station knows the address of itspredecessor (PS={(i-1) mod NP}). Whenever a station de-tects that its predecessor has not sent any message duringa Tlive time interval, it concludes that the station has un-expectedly exited the virtual ring. Thus, a station detecting

114

1 if ACo = NAi−1 then2 THBpred

← 0;3 endif

4 if ACo = NAi−1 then5 start THBpred

;6 endif

Figure 8. Additional Listening items.

such an exit, invokes the address reclaiming procedure. Inparticular, it broadcasts a msg.REMOVE message to removeits predecessor from group G.

It is worth mentioning that, in order to avoid wrong re-movals, every station must transfer, at least, one messageevery TLive. The technique adopted to avoid wrong re-movals uses a heartbeat approach, where the THBown

=12TLive (lines 23-28, Figure 9). In practice, a station thathas not sent any message during the last 1

2TLive interval,the next time it gets the virtual token, it will transfer ei-ther a heartbeat message (msg.HB), or a real-time message(msg.RT).

5.3 Implementation Details

In order to implement the Adding an Removing proce-dures, it is necessary to add some functionalities to the orig-inal procedures. This section describes the main modifica-tions that must be done to allow real-time stations to dynam-ically join or leave the VTP-CSMA architecture. Firstly,to implement the address reclaiming procedure, it has beennecessary to include a THBpred

timer (heartbeat timer) inthe Listening procedure, which will be reset each time thepredecessor station (PS) sends a message. Therefore, it isnecessary to include lines 1-3 from Figure 8 after line 7of the Listening procedure (Figure 3). Furthermore, when-ever ACo = NAi−1 and the ACo value is incremented bycontinuing idle state (t2), it indicates the PS station doesnot send any message and THBpred

timer must be started.Then, it is necessary to include lines 4-6 from Figure 8 afterline 16 of the Listening procedure (Figure 3). Besides, thevariable t3 is no longer necessary and, consequently lines18-19 from Listening procedure must be excluded. This isone of the advantages of the ring management procedure.Whenever a VTP-CSMA station becomes un-synchronized(e.g. when it did not receive an unacknowledge control mes-sage), the station will consider itself out of the ring, and willlater re-join it (Figure 9, lines 16-17). This means that thereis no longer the need to re-initialize the VTP-CSMA archi-tecture from scratch each time.

The main modification to implement the ring manage-ment proposal must be done in the Transmission procedure.

1 Transmission:2 if (ACo=NAi=1) then3 while NAi has msg.JOIN buffered do4 send msg.ADD;5 end while6 send msg.UPDATE7 endif8 if THBpred

≥ TLive then9 send msg.REMOVE;10 endif11 if NAi has msg.RT to be transmitted then12 THBown ← 0; send msg.RT;13 if successful transmission then14 t1← 1; go to Listening;15 else16 if too many attempts then17 NAi do “rejoin”;18 else19 go to Transmission;20 endif21 endif22 else23 start THBown ;24 if THBown ≥ 1

2TLive then

25 send msg.HB; t1← 1; THBown ← 0; go to Listening;26 else27 t2← 2; go to Listening;28 endif29 endif

Figure 9. Transmission procedure (ring man-agement).

For the sake of cleanness, this procedure is rewritten incor-porating such additional modifications (Figure 9). One ofthe modifications considers the case when the token holdingstation detected that its PS did not send any message duringa TLive interval. In such a case, it sends a msg.REMOVEmessage removing its predecessor (lines 8-10). On the otherhand, if the token holding station NAi inside of group G hadno message to transmit during 1

2TLive, it will send a msg.HBmessage in order to avoid wrong removals (lines 24-25).

In case multiple stations issue a request to join the VTR,there is the need to distinguish between the subsequentmsg.ADD messages that have to be delivered, possibly, todifferent stations, in response to the join requests. This re-quires to send, as a parameter, the MAC address of the re-questing station (lines 2-7).

Finally, in the Initialization procedure it is necessary toinitialize the variable THBown and THBpred

and, the Mainprocedure must be executed whenever the VTP-CSMA ar-chitecture is running, instead of while t3 ≤ RN .

115

6 Validation of the VTP-CSMA Architecture

In order to validate the proposed VTP-CSMA architec-ture, one major issue need to be investigated: the “struc-tural” behavior of the VTP-CSMA architecture.

One of the weaknesses of the proposed ring manage-ment scheme is rooted to the use of unacknowledged broad-cast messages. The loss of the msg.REMOVE and/ormsg.UPDATE messages by some nodes may lead to the vio-lation of any, or even both, of the two correctness properties(Agreement and Uniqueness). For example, if one of theVTP-CSMA stations do not receive a msg.REMOVE mes-sage, then it will not update its distributed variables (NA,ACo, NP). As a consequence, there will be an inconsistencyin the distributed variables, and sooner or later there willbe the collision of RT-messages which, in turn, forces bothstations to consider themselves out of the ring. This meansthat the structural behavior of the VTP-CSMA architectureis ensured by means of a self-removal mechanism that re-moves the two colliding stations. Thus, it is expected thatits impact upon the performance of the ring remains neg-ligible (as just the inconsistent station plus one consistentstation are removed from the ring). That is, conversely toother token passing schemes, the ring does not needs to bere-built from scratch.

The proposed ring management procedure has to be care-fully assessed via an adequate performance analysis to val-idate the effectiveness of the self-removal reset mechanismwhen dealing with inconsistent distributed variables. Pre-liminary results of the performance analysis have alreadyhighlighted the effectiveness of such “self-removal” ap-proach upon the original ACo reset mechanism.

7. Conclusions

This paper addresses the VTP-CSMA architecture, thathas been previously proposed [10] to support real-time com-munication in IEEE 802.11e wireless networks with a fixednumber of real-time stations (static environments).

The timing analysis carried out in this paper demon-strates that the token rotation time of the VTP-CSMA archi-tecture is upper-bounded, even in the presence of externalperturbations. This means that the VTP-CSMA architec-ture is able to provide a real-time communication service,even when the communication medium is shared with tim-ing unconstrained traffic sources. The performance analysishighlights the adequate behavior of the VTP-CSMA mech-anism in error-prone channels.

A Ring Management procedure for the VTP-CSMA ar-chitecture has also been proposed, which allows real-timestations to dynamically join or leave the VTP-CSMA archi-tecture. This ring management procedure extends the origi-nally proposed VTP-CSMA architecture, enabling its oper-

ation in dynamic scenarios, such as those found in state-of-the-art VoIP applications.

References

[1] IEEE Standard for Information Technology - ”Logical LinkControl”. ANSI/IEEE 802.2 Std, 1998 edition (Incorporat-ing ANSI/IEEE Stds 802.2c-1997, 802.2f-1997), 1998.

[2] IEEE Standard for Information Technology - Wireless LANMedium Access Control (MAC) and Physical Layer (PHY)Specifications. ANSI/IEEE Std 802.11, 1999 Edition(R2003), 2003.

[3] IEEE Standard for Information Technology - Wireless LANMedium Access Control (MAC) and Physical Layer (PHY)Specifications Amendment 8: MAC Quality of Service En-hancements. IEEE Std 802.11e-2005, 2005.

[4] G. Agrawal, B. Chen, and S. Davari. Guaranteeing syn-chronous message deadlines with the timed token mediumaccess control protocol. IEEE Transactions on Computers,43(3):327 – 339, 1994.

[5] J.-D. Decotignie. Ethernet-based real-time and industrialcommunications. Proceedings of the IEEE, 93(6):1102 –1117, 2005.

[6] B. Goode. Voice over Internet Protocol (VoIP). Proceedingsof the IEEE, 90(9):1495 – 1517, 2002.

[7] S.-K. Kweon and K. G. Shin. Statistical real-time commu-nication over Ethernet. IEEE Transactions on Parallel andDistributed Systems, 14(3):322 – 335, 2003.

[8] L. Lo Bello, G. A. Kaczynski, and O. Mirabella. Improv-ing the real-time behavior of Ethernet networks using trafficsmoothing. IEEE Transactions on Industrial Informatics,1(3):151 – 161, 2005.

[9] R. Moraes, P. Portugal, and F. Vasques. A stochastic petrinet model for the simulation analysis of the IEEE 802.11eEDCA communication protocol. 10th IEEE InternationalConference on Emerging Technologies and Factory Automa-tion - ETFA, pages 38 – 45, 2006.

[10] R. Moraes, F. Vasques, P. Portugal, and J. Fonseca. Real-time communication in 802.11 networks: The virtual tokenpassing VTP-CSMA approach. 31st IEEE Conference onLocal Computer Networks (LCN), pages 389 – 396, 2006.

[11] J.-P. Thomesse. Fieldbus technology in industrial automa-tion. Proceedings of the IEEE, 93(6):1073 – 1101, 2005.

[12] A. Willig. Redundancy concepts to increase transmissionreliability in wireless industrial LANs. IEEE Transactionson Industrial Informatics, 1(3):173 – 182, 2005.

[13] A. Willig, K. Matheus, and A. Wolisz. Wireless technologyin industrial networks. Proceedings of the IEEE, 93(6):1130– 1151, 2005.

[14] A. Willig and A. Wolisz. Ring stability of the PROFIBUStoken-passing protocol over error-prone links. IEEE Trans-actions on Industrial Electronics, 48(5):1025 – 1033, 2001.

[15] H. Zhai, X. Chen, and Y. Fang. How well can the IEEE802.11 wireless LAN support quality of service? IEEETransactions on Wireless Communications, 4(6):3084 – 94,2005.

116