Mobile Netw Appl (2011) 16:531–543DOI 10.1007/s11036-010-0241-y

DFMAC: DTN-Friendly Medium Access Controlfor Wireless Local Area Networks SupportingVoice/Data Services

Hao Liang · Weihua Zhuang

Published online: 9 May 2010© Springer Science+Business Media, LLC 2010

Abstract In this paper, we consider a wireless com-munication network for both local nodes residing indensely populated hot spots and nomadic nodes roam-ing in a large area. Wireless local area networks(WLANs) are deployed in the hot spots, while a delay/disruption tolerant network (DTN) provides services tonomadic nodes in the large area with low node density.We investigate the radio resource allocation for aDTN/WLAN integrated network, and propose a DTN-friendly medium access control (DFMAC) scheme forthe hot spots supporting voice/data services. Analyticalmodels are established to characterize the interactionsbetween a DTN and WLANs under the proposedDFMAC. Numerical and simulation results demon-strate that our proposed MAC scheme can provide adeterministic delay guarantee for the voice services oflocal nodes, while achieving an efficient performancetradeoff for the data services between nomadic nodesand local nodes.

Keywords delay/disruption tolerant network (DTN) ·wireless local area network (WLAN) · integratednetwork · medium access control (MAC)

This paper was presented in part at IEEE InternationalConference on Communications and Networking in China(CHINACOM) 2009.

H. Liang (B) · W. ZhuangDepartment of Electrical and Computer Engineering,University of Waterloo, Waterloo, ON, Canada, N2L 3G1e-mail: h8liang@ecemail.uwaterloo.ca

W. Zhuange-mail: wzhuang@bbcr.uwaterloo.ca

1 Introduction

Nowadays, wireless local area networks (WLANs) arewidely deployed at communication hot spots such aslibraries, cafeterias, classrooms, and other public areas,because of their inherent low implementation cost [1].Both delay sensitive voice service and delay tolerantdata service can be supported by WLANs to facilitateservice demands within the hot spots [2]. In order tofurther enlarge wireless service coverage, the notion ofwireless metropolitan area sharing networks (WMSNs)can be employed [3], where publicly and/or privatelyowned wireless Internet access points (APs) are shared.Similar concepts have been widely implemented incommercialized applications such as FON [4] and ve-hicular ad hoc networks (VANETs) (e.g., MobTorrent[5]). However, the main problem in WMSNs is that thedeployment of the existing APs limits the service cover-age area and hence caps system performance. Withoutsufficient APs, network connections for a mobile nodebecome intermittent when the node roams outside thehot spots. In fact, this problem exists in other networkenvironments such as wireless campus area networks(WCANs) [6] where WLANs are constrained in sep-arate buildings, and rural area networks (RANETs)[7] where hot spots are separated by large physicaldistances.

As a widely accepted solution for intermittent net-work connections, a delay/disruption tolerant network(DTN) can provide robustness for data transmissions [8].The key idea behind DTNs is the store-carry-forwardrouting [9], where data messages are stored and carriedby wireless nodes for a considerably long period oftime. Whenever a communication opportunity arises,the stored and carried data messages can be delivered.

532 Mobile Netw Appl (2011) 16:531–543

In the literature, most of the existing work on DTNrouting assumes a sparse node distribution and a lowtraffic load, whereby the issue of link-layer packet col-lisions is not taken into account. However, a recentstudy indicates that the link-layer contention problemcannot be ignored, especially in densely populated ar-eas (e.g., hot spots) [10]. How to deal with link-layercontention within the DTN framework is a challeng-ing yet open issue. On the other hand, there existextensive research results for medium access control(MAC) in WLANs supporting voice/data services. InIEEE 802.11e, the enhanced distributed channel ac-cess (EDCA) can provide service differentiation byclassifying access categories for services with differentquality of service (QoS) requirements [2]. However,the contention-based MAC of EDCA can only pro-vide stochastic delay guarantee for voice services andthus limits the voice capacity. In order to provide adeterministic delay guarantee, IEEE 802.11e hybrid co-ordination function (HCF) incorporates a superframestructure and provides a polling-based contention-freeperiod for voice services. By utilizing voice multiplex-ing, deterministic access prioritization, and voice packetaggregation, the voice/data capacity of IEEE 802.11eHCF can be further improved [11]. In addition, a prob-abilistic token-based scheme is proposed, which notonly achieves high channel utilization but also providesservice differentiation for delay sensitive voice serviceand best-effort data service [12]. However, the existingwork focuses on nodes in hot spots only. How to ensureeffective and efficient MAC in the presence of nomadicnodes needs further investigation.

In this work, we consider a DTN/WLAN inter-working environment [13]. Different from other inter-working paradigms such as celluar/WLAN integratednetworks [1, 14], in the DTN/WLAN interworkingenvironment under consideration, ubiquitous wirelesscoverage is not provided or not even available tonomadic nodes. Nomadic nodes can only get real-time Internet access within the hot spots serviced byWLANs. We focus on resource allocation within hotspots and propose a DTN-friendly MAC (DFMAC)scheme based on a superframe structure and tokenpassing strategies. Under the DFMAC scheme, a deter-ministic delay guarantee can be provided to the voiceservices of local nodes, while a performance tradeofffor data services between nomadic nodes and localnodes can be achieved by tuning phase duration pa-rameters. Analytical results are provided to demon-strate the performance of DFMAC. We further validateour performance analysis with extensive computer sim-ulations. Performance comparisons with the existingschemes are also provided to show the effectiveness and

Table 1 Summary of important symbols used

Symbol Definition

BU Buffer size for the uplink delay tolerantdata service at the tagged nomadic node

Hk kth hot spot in the network coverage areaK Number of packets in each messageMH Number of hot spotsMLD,k (MLV,k) Number of local data (voice) nodes within

hot spot Hk

NV The maximum number of voice packetsthat can be generated by a voice sourceduring each superframe

POFF,i (PON,i) The probability that there are i voicepacket arrivals during a superframe,while at the beginning of this superframethe voice source is at the OFF (ON) state

RTR Radio transmission rangeT Duration of superframeTB Duration of beaconTD(k) The maximum duration of dedicated phase

within each superframe in hot spot Hk

TDL(k) (TUL(k)) The maximum duration of DL (UL)subphase within each superframe inhot spot Hk

TO Duration for transmitting the overheadof a voice packet

TP Duration for data packet transmissionTPL,D (TPL,V) Duration for the payload transmission of a

data (voice) packetTR,i,k Token rotation time of local data node i

within hot spot Hk

TSJ Sojourn durationTT Duration for token passingTV(k) Average channel time occupied by local

voice services in a superframe1/α (1/β) Average duration of the ON (OFF)

period of local voice traffic flowsβDL(k) (βUL(k)) The maximum fraction of DL (UL)

subphase within each superframe inhot spot Hk

γV Inter-visiting rateλi,k Average packet arrival rate of local data

node i within hot spot Hk

λU Average message arrival rate of thetagged nomadic node

φ Diameter of hot spotψ Network coverage areaτI Waiting time for the wireless channel

to be idle before voice/data packettransmission

τIV Waiting time for the wireless channelto be idle before the transmission of alocal voice node switching from theOFF state to the ON state

τAD Waiting time for the wireless channelto be idle before sending theARRIVAL or DEPARTURE message

τV Inter-arrival time of voice packets

Mobile Netw Appl (2011) 16:531–543 533

efficiency of the newly proposed DFMAC. To the bestof our knowledge, this is the first work in the literatureto capture the interworking of DTN/WLAN and devisea strategy to balance radio resource allocation betweenthese two network components.

The remainder of the paper is organized as follows.In Section 2, we present the system model under con-sideration. In Section 3, the proposed DFMAC schemeis discussed in details. The performance analysis ofDFMAC is presented in Section 4, followed by numer-ical results in Section 5. We conclude this research inSection 6. As many symbols are used in this paper,Table 1 summarizes the important ones.

2 System model

Consider a DTN/WLAN integrated network as shownin Fig. 1, where a number of isolated small hot spots(Hk, k = 1, 2) are scattered around a large networkregion. Wireless Internet APs attached to the Internetbackbone provide wireless access to the nodes in theirhot spots. Under this network architecture, we considertwo kinds of wireless nodes, namely (1) the local nodesresiding in hot spots and (2) the nomadic nodes whichcan roam within the entire network region. All mobilenodes and APs are equipped with a short range radiotransceiver. Nomadic nodes comprise a DTN, whereaslocal nodes in a hot spot comprise a WLAN. A nomadicnode can access a WLAN when it comes into a hot spot.The WLAN within a hot spot is fully connected and theconnections are reliable, but the network connectionsbecome intermittent for nodes outside hot spots. Thenomadic nodes can stay disconnected for hours andtravel through a hot spot within several minutes. TheDTN and WLANs operate in different frequency chan-nels to prevent the transmission interference betweenthese two network components. With unique featuresof a DTN and a WLAN, a DTN/WLAN integratednetwork is imperative in order to better serve nomadicnodes when they travel within hot spots.

In the DTN/WLAN integrated network, two cate-gories of services should be differentiated in resourceallocation of a WLAN, i.e., local services and delaytolerant data services. The local services are generatedby local nodes and destined for a server in the Internet(uplink) or vice versa (downlink), while delay tolerantdata services are generated by nomadic nodes and des-tined for a server in the Internet (uplink) or vice versa(downlink). The local services are further categorizedinto local voice services and local data services. Each lo-cal voice service can be described by a two-way (uplinkand downlink) ON/OFF model. At the ON state, voice

H2

Internet Backbone

H1

AP

Local Node

Nomadic Node

WLAN

DTN

WLAN

Fig. 1 An illustration of the DTN/WLAN integrated network

packets are generated by a voice source with a constantinter-arrival time, while at the OFF state, no voicepacket is generated. In the DTN framework, delaytolerant data services are represented in self-containedmessages (referred to as bundles [8]). Segmentationof a message into packets is necessary for link-layertransmissions in a WLAN. Here, we focus on the casewhere messages are delivered by direct transmissionsbetween nomadic nodes and the APs, while the impactof message relaying among nomadic nodes is left forfurther work.

The two categories of services have different QoSrequirements. For local nodes, the packet delay of voiceservices and the throughput of data services are themajor QoS measures. On the other hand, for delay tol-erant data services to nomadic nodes, low throughputand large delay are acceptable. As storage capability isusually limited for nomadic nodes, delivery probabilityunder a limited buffer space becomes the most impor-tant performance metric. In this work, we consider thelocal voice services have the highest priority. In otherwords, the delay requirement of local voice servicesshould be consistently satisfied regardless of the ar-rivals and departures of nomadic nodes in a WLAN.On the other hand, for data services, as the time that anomadic node remains in a hot spot is expected to bevery brief, its messages should be transmitted with highpriority when the node is connected via the WLAN.In order to provide service differentiation betweendelay tolerant data services and local data services, anefficient resource reservation for higher-priority no-madic nodes is indispensable [15], to be discussed inSection 3.

534 Mobile Netw Appl (2011) 16:531–543

3 DTN-friendly medium access control (DFMAC)

The DFMAC is a token-based scheme designed forWLANs where local nodes and nomadic nodes coexist.A superframe structure is implemented to provide adeterministic delay guarantee for local voice serviceswhile achieving service differentiation between delaytolerant data services and local data services. Intra-phase and inter-phase token passing mechanisms areprovided to achieve efficient resource allocation.

3.1 Superframe structure

As shown in Fig. 2, under the DFMAC scheme, timeis partitioned into superframes of constant duration T.The duration T can be selected based on the delayrequirement of local voice services [11, 16]. At thebeginning of each superframe, there is a short beaconperiod TB for synchronization, followed by a local voicephase, a dedicated phase, and a local data phase. Thesethree phases are dedicated to the local voice services,delay tolerant data services, and local data services,respectively.

In order to achieve the deterministic delay guaranteefor local voice services, the duration of the local voicephase within a superframe varies and depends on thenumber of voice sources at the ON states. Note that,if the local voice services can tolerate some loss, ourwork can be directly generalized to provide a stochasticdelay guarantee by setting a maximum duration for thelocal voice phase according to the loss tolerance [16].The dedicated phase can have a maximum durationTD(k) within each superframe, where k is the indexof a hot spot. Each dedicated phase is divided into anuplink (UL) subphase for uplink delay tolerant dataservices and a downlink (DL) subphase for downlinkdelay tolerant data services with maximum durationsTUL(k) and TDL(k), respectively. Obviously, we haveTUL(k) + TDL(k) = TD(k). The remaining period in the

Local VoicePhase

Local Data Phase

UL Subphase

TUL (k)

DL Subphase

TDL(k)

T

TB

Beacon

DedicatedPhase

TD (k)

Fig. 2 Superframe structure of DFMAC

superframe is the local data phase. At any time, if thereis no local voice node at the ON state, a dedicatedphase will begin after the beacon period; if there is nonomadic node in the hot spot, the unutilized dedicatedphase is merged to the local data phase. In this research,we assume a basic admission strategy which ensuresthat the local data phase has a non-zero share of eachsuperframe [16], and that a maximum fraction βUL(k) =TUL(k)/T and βDL(k) = TDL(k)/T of each superframecan be guaranteed for the UL and DL subphases,respectively.

There is a token in each WLAN. Only the tokenholder can have an opportunity for packet transmission.The token is circulated among local voice nodes andthe AP in the local voice phase, among nomadic nodesand the AP in the dedicated phase, and among localdata nodes and the AP in the local data phase. Onlynomadic nodes can access the UL subphase and onlythe AP can access the DL subphase. Since all downlinkdelay tolerant data service queues reside at the AP,when the AP holds the token in a DL subphase, thetoken is circulated among these queues. By introducingthe dedicated phase, radio resources can be reservedfor nomadic nodes since the time for them to travelthrough a hot spot is in general very brief.

For intra-phase token passing, three different mech-anisms are adopted for the three phases. An inter-phasetoken passing mechanism is also provided to pass thetoken among different phases.

3.2 Deterministic token passing in the local voice phase

As discussed in Section 2, the delay guarantee of localvoice services should be consistent regardless of thetraffic load fluctuation caused by the occasional pres-ence of nomadic nodes in a WLAN. The polling-basedMAC can provide a deterministically bounded delayfor voice services [16]. However, the polling overheadshould be minimized to provide high resource utiliza-tion by considering voice multiplexing [11]. In thisresearch, we use a deterministic token passing mech-anism for the local voice phase to reduce the pollingoverhead in terms of the polling frames. Moreover,with a superframe structure, voice packet aggregationcan be performed for overhead reduction, where thevoice packets generated by a voice source during eachsuperframe are combined and transmitted by a singleMAC frame. Therefore, the overhead caused by RTP(real-time transport protocol)/UDP (user datagramprotocol)/IP headers and physical (PHY) preamblescan be reduced [11].

For deterministic token passing, each local node(including the AP) records the next token holder to

Mobile Netw Appl (2011) 16:531–543 535

which the token should be passed. The next tokenholder is updated each time a local voice call arrives/completes, or a local voice node is switched from theOFF state to the ON state or vice versa. When a voicenode holds a token, it aggregates all the voice packetsbacklogged and transmits them by a single MAC frame.The token is piggybacked to the packets and can bedecoded by the next token holder through overhearing.Before each transmission, the current token holdershould wait for the wireless channel to be idle for afixed period of time τI. If the current token holderis switched from the ON state to the OFF state, itfurther piggybacks a VOICE OFF message to the cur-rent transmission. After overhearing the current trans-mission, the previous token holder changes the nexttoken holder to the one indicated in the piggybackedtoken of the current transmission. On the other hand,a local voice node switching from the OFF state to theON state transmits a voice packet with a piggybackedVOICE ON message after sensing the channel idle fora short period of time τIV (τIV < τI), and passes thetoken to the node whose transmission is delayed. Uponoverhearing the current transmission, the previous to-ken holder sets the local voice node (from the OFFstate to the ON state) as the next token holder. If thelocal voice node switching from OFF to ON transmits avoice packet during a dedicated phase, the UL or DLsubphase is delayed by the time of the voice packettransmission since the dedicated phase is reserved fornomadic nodes.

3.3 Probabilistic token passing for local data phase

The framework of probabilistic token passing proposedin [12] is adopted for the local data phase, due toits good performance and its capability of supportingservice differentiation. For a group G of nodes (orqueues) contending for wireless channel access, thetoken passing scheme can be characterized by two pa-rameters: ppp = {pi, 1 ≤ i ≤ |G|} and PPP = {Pi, j, 1 ≤ i,j ≤ |G|}, where pi denotes the token holding probabil-ity of node i, Pi, j represents the token passing proba-bility from node i to node j, and |G| is the size of G.Without service differentiation among local data nodes,the values of pi and Pi, j are given by

pi = 1/|G|, i ∈ {1, 2, · · · , |G|} (1)

Pi, j ={

1/(|G| − 1), if i �= j, i, j ∈ {1, 2, · · · , |G|}0, otherwise.

(2)

Specifically, when |G| = 1, the only node will alwayshold the token by letting pi = 1 and Pi,i = 1. If a tokenholder has a data packet to transmit, the token is piggy-backed to the data packet; otherwise, it simply passesthe token to the next node.

3.4 Truncated token passing for the dedicated phase

For the dedicated phase, a truncated token passingpolicy is adopted. It is based on the framework ofprobabilistic token passing as discussed in Section 3.3.However, if the queue of a certain nomadic node isdrained, the token will not be passed on to that nodeanymore within the current superframe. In the ULsubphase, suppose a set GUL of nomadic nodes in a hotspot are contending for channel access. If node m, m ∈GUL, has an empty queue, it broadcasts a DRAINEDmessage to all other nodes and passes the token toone of them with equal probabilities. For the remainingnodes, upon receiving the DRAINED message, theyrecalculate the token passing probability by replacingGUL with GUL \ {m}. The above procedure continuesuntil there is only one nomadic node left in the ULsubphase and its queue is also drained or the boundaryof UL subphase is reached or crossed. In such cases,inter-phase token passing is initiated to pass the tokento the DL subphase, to be discussed in Section 3.5.The truncated token passing policy is virtually adoptedin the DL subphase for downlink delay tolerant dataqueues, where the AP always holds the token. Thetruncated token passing procedures for the UL and DLsubphases are restarted after the beacon period of eachsuperframe. Obviously, when the message arrival rateof delay tolerant data services is low, the probabilityfor a message to arrive during a dedicated phase is lowaccordingly since the duration of each dedicated phaseis short. By using the truncated token passing policy,unnecessary token passing within each dedicated phasecan be reduced when the uplink or downlink delaytolerant data service queue is empty, and the unutilizedproportion of the superframe can be passed on to thelocal data phase for local data services.

3.5 Inter-phase token passing

For inter-phase token passing, each node (includingthe AP) keeps three synchronized timers to track howmuch time has elapsed within the current superframe,UL subphase, and DL subphase. These timers arerestarted accordingly at the beginning of each super-frame, UL subphase, and DL subphase.

After the beacon period, the local voice phase startswith the AP transmitting its aggregated voice packets.

536 Mobile Netw Appl (2011) 16:531–543

At the end of a voice phase, the next token holder ofthe last local voice node is set to a nomadic node (ora queue) belonging to the dedicated phase with thesame probability as the token holding probability of thetarget node.

In the dedicated phase, before each packet transmis-sion, the token holder checks the synchronized timerand compares it with the boundary of each subphase,i.e., TUL(k) and TDL(k) in hot spot k. After the currenttransmission is finished, if the boundary is reached orcrossed, the next token holder is set to a node belongingto the next phase with the same probability as the tokenholding probability of the target node in that phase.

As data transmissions in the beacon period are notallowed, if the time left in the local data phase withinthe superframe is not sufficient for transmitting onemore packet after the current transmission, the currenttoken holder passes the token through the currenttransmission to a node belonging to the next phase orsubphase. A voice node switching from OFF to ONduring a local data phase defers its current transmissionif the time left in the current superframe is not enoughfor the transmission of a voice packet. The new tokenholder will not start its transmissions until the beaconperiod of the new superframe is finished.

3.6 Node (call) arrivals and departures (completions)

When a node arrives at or departs from the WLAN (ora voice/data call arrives or completes), the source nodesends an ARRIVAL or DEPARTURE message aftersensing the channel idle for a short period of time τAD

(τAD < τIV). After receiving the message, the currenttoken holder discards the token, and the AP recal-culates the durations (TUL(k) and TDL(k)) and tokenpassing parameters (ppp and PPP) for all the phases andsubphases. If the new call is a local voice call, the nodeis put at the end of the deterministic token passing list.Then the AP broadcasts a NOTE message to inform allnodes of the updated information as well as the nodeto which a new token is issued. The new token holderstarts transmission τI after the beacon signal fromthe AP.

3.7 Failure recovery

The current token holder monitors the current tokenpassing procedure. If a new transmission does not beginfrom the next token holder i after τF , where τF isslightly larger than τI, a new token will be issued andpassed to node i again. This procedure will continue forat most NF times. After that, if there is still no trans-

mission initiated by node i, the node is consideredto be inactive in the system due to battery depletionor hardware impairments. A NOTE message is thenbroadcasted by the current token holder, treating nodei as a leaving node. (Note that the node mobility isnot considered as a cause of failure since the nodedeparture procedure can be applied when a node tendsto move out of a hot spot, as discussed in Section 3.6.)During the recovery procedure, the passing of a newtoken by the current token holder is considered thesame as a normal token passing. If the inter-phasetoken passing condition is satisfied according to thediscussion given in Section 3.5, the current token holderstops the recovery procedure and, at the same time,generates a new token and passes it to a node belongingto the new phase accordingly.

4 Performance analysis

For analysis tractability, the following assumptions aremade:

1. All mobile nodes and APs are equipped with thesame radio transceivers with a transmission rangeRTR. The transmission range is much smaller thanthe dimension of network coverage area ψ . Eachhot spot covers a circular region with diameterφ (φ2 � ψ). There is no node failure.

2. The movement of nomadic nodes follows a randomdirection (RD) mobility model [17], and the densityof nomadic nodes is low.

3. Packet arrivals of local data service and message ar-rivals of delay tolerant data service are independentPoisson processes. Each message is composed of Kpackets.

For a nomadic node, define inter-visiting time as theduration between its two consecutive visits to a hotspot, and sojourn duration as the period that it stayswithin a hot spot in each visit. Based on assumptions(1) and (2), the inter-visiting time is approximatelyexponentially distributed [17]. We denote the rate thata nomadic node visits a hot spot (inter-visiting rate)by γV, which is the reciprocal of inter-visiting time.Although the distribution of sojourn duration is notobtainable, its expectation TSJ can be derived as a spe-cial case of the average contact duration between twomobile nodes by letting one of them stationary [17]. Theprobability that there are two or more nomadic nodessimultaneously within a hot spot is negligible with a lownomadic node density and φ2 � ψ .

The average packet arrival rate of local data node iwithin hot spot Hk is denoted by λi,k (0 ≤ i ≤ MLD,k,

Mobile Netw Appl (2011) 16:531–543 537

1 ≤ k ≤ MH), where MLD,k is the number of local datanodes within hot spot Hk, MH is the number of hotspots, and λ0,k represents the packet arrival rate at theAP of hot spot Hk and destined for the local data nodes.For nomadic nodes, the average message arrival ratesof uplink and downlink delay tolerant data services aredenoted by λU,i and λD,i (1 ≤ i ≤ MN), respectively,where MN is the number of nomadic nodes. The buffersize for uplink delay tolerant data service at nomadicnode i is BU,i, and the buffer size for downlink delaytolerant data service at the AP for nomadic node iis BD,i.

Let MLV,k denote the number of local voice nodeswithin hot spot k. The ON and OFF periods of localvoice traffic flows are exponentially distributed withaverage value of 1/α and 1/β, respectively. At the ON

state, voice packets arrive with a constant inter-arrivaltime τV.

4.1 Fraction of channel time for local voice services

As discussed in Section 3.2, the MAC scheme is de-signed to guarantee a deterministic delay bound forlocal voice services. Therefore, we focus on the channeltime occupancy by local voice services in this subsec-tion. At the beginning of each superframe, a voicesource is at the OFF state with probability α

α+βand

at the ON state with probability β

α+β[11]. The joint

probability mass functions (PMFs) of the voice sourcesat the OFF/ON states at the beginning of a superframeand the number of voice packet arrivals during thesuperframe are given respectively by

POFF,i =

⎧⎪⎨⎪⎩

αα+β

[exp(−β(T − iτV)) − exp(−β(T − (i − 1)τV))], if 1 ≤ i ≤ NV − 1α

α+β[1 − exp(−βτV)], if i = NV

αα+β

− ∑NVi=1 POFF,i, if i = 0

(3)

PON,i =

⎧⎪⎨⎪⎩

β

α+β[exp(−α(i − 1)τV)) − exp(−αiτV)], if 1 ≤ i ≤ NV − 1

β

α+β[exp(−α(T − τV))], if i = NV

β

α+β− ∑NV

i=1 PON,i, if i = 0

(4)

where NV = T/τV is the maximum number of voicepackets that can be generated by a voice source duringeach superframe. POFF,i and PON,i are the probabilitythat there are i voice packet arrivals during a super-frame, and at the beginning of this superframe the voicesource is at OFF and ON states, respectively.

Since under the DFMAC scheme, a local voice nodeswitching from the OFF state to the ON state transmitsthe first voice packet with waiting time τIV for channelidle, the average channel time occupied by local voiceservices in a superframe is given by

TV(k) =NV∑i=1

[POFF,i · (2TO − τI + τIV + i · TPL,V)

+PON,i · (TO + i · TPL,V)] · MLV,k

+NV∑i=1

(POFF,i + PON,i) · (TO + i · TPL,V) · MLV,k

(5)

where the first and second summations correspond tothe fractions of channel time occupied by uplink anddownlink voice services, respectively; TO is the dura-tion for transmitting the overhead of a voice packet,including τI, PHY preamble, and RTP/UDP/IP headers

[12]; TPL,V is the duration for the payload transmissionof a voice packet; k is the index of a hot spot. Then, thefraction of channel time for local voice services can becalculated as ηV(k) = TV(k)/T.

4.2 Throughput of local data services

The main system performance measure for local dataservices is normalized system throughput, which isdefined as the fraction of channel time used forpacket payload transmissions. Consider the worst-casethroughput for local data services when the dedicatedphase is fully occupied by nomadic nodes. Note thatunder the truncated token passing policy described inSection 3.4, the actual normalized system throughputcan be better than the worst-case value if the messagearrival rate of delay tolerant data service is low.

Let TR,i,k be the token rotation time of local datanode i within hot spot Hk, which is the time durationbetween two consecutive token holdings of the node.Then, the queue utilization of node i within hot spotHk can be calculated by

ρi,k = λi,k E[TR,i,k

]. (6)

When nomadic nodes are present in hot spot Hk,define a random variable Di,k, where Di,k = 1 when

538 Mobile Netw Appl (2011) 16:531–543

the local voice phase and dedicated phase are includedin the token rotation cycle of node i within hot spotHk, and Di,k = 0 otherwise. We have P(Di,k = 0) =1 − P(Di,k = 1). Therefore, the average token rotationtime is given by

E[TR,i,k] = E[TR,i,k|Di,k = 0]P(Di,k = 0)

+ E[TR,i,k|Di,k = 1]P(Di,k = 1),

0 ≤ i ≤ MLD,k, 1 ≤ k ≤ MH (7)

where

E[TR,i,k|Di,k = 0] =MLD,k∑n=0

[TPρn,k + TT(1 − ρn,k)] (8)

E[TR,i,k|Di,k = 1] = E[TR,i,k|Di,k = 0] + TB

+ TV(k) + TD(k) (9)

TP and TT being the time needed for data packettransmission and token passing, respectively. Note thatfor analysis tractability, an approximation is made in(9) by using the result of (5) which includes the packettransmission time of local voice nodes switching fromOFF to ON. The accuracy of this approximation will beshown in Section 5.2. Since there is only one token rota-tion cycle with Di,k = 1 in each superframe, the averagetime occupied by token rotation cycles with Di,k = 0 ina superframe is T − E[TR,i,k|Di,k = 1]. Then, we have

P(Di,k = 1) = 1

1 + (T−E[TR,i,k|Di,k=1])E[TR,i,k|Di,k=0]

= E[TR,i,k|Di,k = 0]T − TB − TV(k) − TD(k)

. (10)

From (6)–(10), we can find the queue utilizationsρi,k (0 ≤ i ≤ MLD,k, 1 ≤ k ≤ MH). Then the normalizedsystem throughput of local data services within hot spotHk in the presence of nomadic nodes can be derived as

Thk = T − TB − TV(k) − TD(k)

T

·∑MLD,k

i=0 TPL,Dρi,k∑MLD,k

i=0 [TPρi,k + TT(1 − ρi,k)](11)

where TPL,D is the time used for data packet payloadtransmission.

When nomadic nodes are not present in hotspot Hk, we have TD(k) = 0 and P(Di,k = 1) = 0.Since E[TR,i,k|Di,k = 0] = E[TR,i,k] with P(Di,k = 0) =1, from (6) and (8) we can obtain another set of queue

utilizations ρ∗i,k (0 ≤ i ≤ MLD,k, 1 ≤ k ≤ MH). Then, the

normalized system throughput is given by

Th∗k = T − TB − TV(k)

T

·∑MLD,k

i=0 TPL,Dρ∗i,k∑MLD,k

i=0

[TPρ∗

i,k + TT(1 − ρ∗

i,k

)] . (12)

4.3 Delivery probability of delay tolerant data services

Since the inter-visiting time is exponentially distributedand is expected to be much larger than the sojournduration under assumptions (1) and (2), the queueingbehaviors of both uplink and downlink delay toler-ant data services can be described by continuous timeMarkov chains (CTMCs). Without loss of generality,consider the uplink delay tolerant data service. Fora tagged nomadic node with message arrival rate λU

and mobility parameters γV and TSJ, define a CTMCXU (t), where XU (t) ∈ [0, BU] denotes the number ofdelay tolerant data messages in the uplink queue andBU is the maximum queue length. The transition rateqU

i, j (0 ≤ i, j ≤ BU) from state i to state j of the uplinkqueue can be calculated as follows.

With the message arrival rate λU, we have

qUi,i+1 = λU, 0 ≤ i ≤ BU − 1. (13)

When the tagged nomadic node visits one of the hotspots, messages can be transmitted to the destinationvia the AP. If after the current visit, the uplink delaytolerant data service queue is still not empty, the tran-sition rate is given by

qUi, j =

MH∑k=1

γV I(SUL(k), i − j), 1 < i ≤ BU, 1 ≤ j < i

(14)

where I(i, j) is an indication function which equals1 if i is equal to j, and 0 otherwise. The functionSUL(k) denotes the maximum number of uplink delaytolerant data messages that can be transmitted duringthe sojourn duration when hot spot k is visited, and isgiven by

SUL(k) =⌊

round(

TSJ

T

)TUL(k)

TP K

⌋. (15)

Note that we make an approximation on the fractionalsuperframe in the calculation of SUL(k) by consideringthe fact that the duration of a superframe is muchshorter than the sojourn duration. For the same reason,the time spent on the arrival and departure proceduresof a nomadic node is ignored.

Mobile Netw Appl (2011) 16:531–543 539

After the current visit, if the uplink delay tolerantservice queue of the node is empty, the transition rateis given by

qUi,0 =

MH∑k=1

γVC(SUL(k), i), 1 ≤ i ≤ BU (16)

where C(i, j) is a comparison function which equals 1if i ≥ j and 0 otherwise. For any other state transitionsnot mentioned precedingly, the transition rate is equalto 0.

Obviously, the CTMC is a finite state irreducibleMarkov chain; therefore it is ergodic, and the stationaryprobability exists. Denote the stationary probability byπππU = {

πUi , 0 ≤ i ≤ BU

}. It can be calculated by solving

a set of balance equations [18].The delivery probability of uplink delay tolerant

data service is given by PD,U L = 1 − πUBU

. Similarly,we can obtain the stationary probability πππ D = {

π Di ,

0 ≤ i ≤ BD}

and the delivery probability PD,DL forthe downlink.

5 Numerical and simulation results

In this section, numerical and simulation results arepresented to demonstrate the performance of the DF-MAC. Performance comparisons with the modifiedHCF scheme [11] and probabilistic token-based scheme[12] are also provided to show the efficiency of ourproposed scheme. Note that a comparison betweenthe DFMAC and the IEEE 802.11e MAC is not pro-vided since the advantage of the modified HCF schemeover IEEE 802.11e MAC is already demonstratedin [11].

The system parameters used in analysis and simula-tion are shown in Table 2 and are in accordance with

Table 2 System parametersused in analysis andsimulation

Parameter Value

τI 40 μsτIV 20 μsτAD 10 μsτV 20 msT 100 msTB 0.3920 msTP 0.9793 msTT 0.3960 msTPL,D 0.7273 msTPL,V 0.024 msTO 0.2858 ms1/α 352 ms1/β 650 ms

previous work [11, 12]. There are two non-overlappinghot spots (MH = 2) with arbitrary deployment loca-tions in the network region. For illustration clarity, weconsider the two hot spots are equivalent with βUL(1) =βUL(2) = βUL and βDL(1) = βDL(2) = βDL. The taggednomadic node is considered as a pedestrian withRD mobility speed 1.5 m/s and radio transmissionrange RTR = 250 m. The diameter of each hot spotis φ = RTR. Using typical RD mobility parameters[17], for network coverage area ψ = 2,000 × 2,000 m2,we have γV = 9.1912 × 10−5 s−1, TSJ = 107.1038 s; forψ = 2,500 × 2,500 m2, we have γV = 5.8824 × 10−5 s−1,TSJ = 107.1038 s. Note that the same sojourn durationis obtained since it only depends on the RD mobilityparameters and is not affected by the size of network.For delay tolerant data services, each message consistsof K = 100 packets.

5.1 Fraction of channel time for local voice services

The fraction of channel time used by local voice ser-vices versus the number of local voice nodes is shownin Fig. 3 for the three MAC schemes. We can see that,the analytical and simulation results match well witheach other for our proposed DFMAC scheme. As thenumber of local voice nodes increases, the fraction ofchannel time used by the local voice services increasessince more voice packets per unit time need to betransmitted. However, for different number of localvoice nodes, the DFMAC scheme requires a minimalchannel time. Compared with the probabilistic token-based scheme, the DFMAC scheme incorporates a

10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Number of local voice nodes

Fra

ctio

n of

cha

nnel

tim

e

Probabilistic tokenModified HCFDFMAC (simulation)DFMAC (analysis)

Fig. 3 The fraction of channel time for the local voice servicesversus the number of local voice nodes

540 Mobile Netw Appl (2011) 16:531–543

superframe structure and exploits the voice packetaggregation to reduce the packetization overhead.Although the delay experienced by each voice packetmay be slightly larger under the DFMAC scheme thanunder the probabilistic token-based scheme, the delayis bounded and can be tuned by selecting differentsuperframe duration T [11, 16]. Compared with themodified HCF scheme, the DFMAC scheme has alower channel occupancy time for local voice servicessince the polling frames for local voice nodes at the ONstate are eliminated.

5.2 Throughput of local data services

To evaluate the throughput of local data services, wetake hot spot H1 as an example. For illustration clarity,the same value of packet arrival rate is chosen for alllocal data nodes and the AP. Since the influence ofβUL and βDL on the local throughput is equivalent, weset βDL to 0.1 and tune the value of βUL to show itseffect. Figures 4, 5 and 6 show the normalized systemthroughputs for 10 local voice nodes (MLV,1 = 10) and10 local data nodes (MLD,1 = 10), 10 local voice nodesand 30 local data nodes, and 30 local voice nodes and30 local data nodes, respectively. As the packet arrivalrate of each node increases, the normalized systemthroughput increases almost linearly before saturation,and the speed of increment is much higher for the caseof 30 local nodes. For the DFMAC with a larger valueof βUL, the saturation occurs at a smaller packet arrivalrate. But the normalized saturated throughput does not

20 40 60 80 100 120 140

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Packet arrival rate (packets per second)

Nor

mal

ized

sys

tem

thro

ughp

ut

Without nomadic nodes (analysis)Without nomadic nodes (simulation)With nomadic nodes, βUL=0.1 (analysis)

With nomadic nodes, βUL=0.1 (simulation)

With nomadic nodes, βUL=0.2 (analysis)

With nomadic nodes, βUL=0.2 (simulation)

Fig. 4 The normalized system throughput of local data servicesversus packet arrival rate (10 local voice nodes, 10 local datanodes)

20 40 60 80 100 120 140

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Packet arrival rate (packets per second)

Nor

mal

ized

sys

tem

thro

ughp

ut

Without nomadic nodes (analysis)Without nomadic nodes (simulation)With nomadic nodes, βUL=0.1 (analysis)

With nomadic nodes, βUL=0.1 (simulation)

With nomadic nodes, βUL=0.2 (analysis)

With nomadic nodes, βUL=0.2 (simulation)

Fig. 5 The normalized system throughput of local data servicesversus packet arrival rate (10 local voice nodes, 30 local datanodes)

change much with the number of local nodes for a givenβUL value, since the local data phases are already fullyoccupied by packet transmissions, no matter how manylocal nodes are present. This phenomenon can alsobe explained by (11): As all ρi,k approaches one, thesecond factor on the right hand side remains unchangedeven if the value of MLD,k increases. As a result, thenormalized system throughput depends only on thefirst factor which is a non-increasing function of βUL.From these three figures we can also observe that, fordifferent system configurations, our analytical model

20 40 60 80 100 120 140

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Packet arrival rate (packets per second)

Nor

mal

ized

sys

tem

thro

ughp

ut

Without nomadic nodes (analysis)Without nomadic nodes (simulation)With nomadic nodes, βUL=0.1 (analysis)

With nomadic nodes, βUL=0.1 (simulation)

With nomadic nodes, βUL=0.2 (analysis)

With nomadic nodes, βUL=0.2 (simulation)

Fig. 6 The normalized system throughput of local data servicesversus packet arrival rate (30 local voice nodes, 30 local datanodes)

Mobile Netw Appl (2011) 16:531–543 541

provides a good estimation of the normalized systemthroughput of local data services.

The relation between the fraction of channel timeoccupied by delay tolerant data services and the nor-malized system throughput of local data services isdemonstrated in Fig. 7 for 10 local voice nodes and30 local data nodes case and in Fig. 8 for 30 local voicenodes and 30 local data nodes case. Using the DFMACscheme, the fraction of channel time occupied by delaytolerant data services is equivalent to βUL + βDL. Thepacket arrival rates of local data node and the AP arefixed to 15 packets per second. Note that this packetarrival rate does not result in throughput saturationwhen there is no nomadic node in a hot spot, as shownin Figs. 5 and 6. From Figs. 7 and 8, we observe that,as the fraction of channel time increases, the normal-ized system throughput under the DFMAC scheme isalmost unchanged when the fraction is small. This isbecause, when the queue utilization of local data nodesis low, the probability for local data queues to be emptyis high, and a considerable portion of channel time isoccupied by pure token passing. On the other hand,as the fraction of channel time increases, the queueutilization of local data nodes increases and the over-head on the pure token passing decreases. However,when the fraction of channel time further increases, thenormalized system throughput degrades dramaticallydue to throughput saturation in the local data phase.In a saturation condition, the local data phases arefully occupied by packet transmissions, therefore thenormalized system throughput only depends on theduration of local data phases. With nomadic nodes in

0 0.1 0.2 0.3 0.4 0.5 0.6 0.70.15

0.2

0.25

0.3

0.35

Fraction of channel time

Nor

mal

ized

sys

tem

thro

ughp

ut

DFMACProbabilistic tokenModified HCF

Fig. 7 The normalized system throughput of local data servicesversus the fraction of channel time occupied by delay tolerantdata services (10 local voice nodes, 30 local data nodes)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.70.15

0.2

0.25

0.3

0.35

Fraction of channel time

Nor

mal

ized

sys

tem

thro

ughp

ut

DFMACProbabilistic tokenModified HCF

Fig. 8 The normalized system throughput of local data servicesversus the fraction of channel time occupied by delay tolerantdata services (30 local voice nodes, 30 local data nodes)

hot spots, the only possibility to facilitate delay tolerantdata services is to sacrifice the throughput of local dataservices. A theoretical explanation can also be givenfrom (11), since the normalized system throughputdepends only on the first factor on the right hand sideas all ρi,k approaches one.

A comparison among the DFMAC scheme, theprobabilistic token-based scheme, and the modifiedHCF scheme is also demonstrated in Figs. 7 and 8.Before throughput saturation, the three schemesachieve similar normalized system throughput of localdata services. However, in a saturation condition, theDFMAC scheme outperforms the other two schemesand thus can achieve a better tradeoff between thedelay tolerant data services and local data services. Forinstance, for a given normalized system throughput oflocal data services, a larger fraction of channel time canbe occupied by the delay tolerant data services underthe DFMAC scheme. Compared with the probabilis-tic token-based scheme, the performance improvementobtained by utilizing the DFMAC scheme increaseswith the increase of the number of local voice nodes,since the performance gain of the DFMAC schemeover the probabilistic token-based scheme is obtainedby reducing the channel occupancy of local voice ser-vices. On the other hand, although the modified HCFscheme can provide a relatively efficient resource allo-cation for local voice services as shown in Fig. 3, itcannot achieve an efficient tradeoff between the delaytolerant data services and local data services comparedwith the DFMAC scheme, since the contention-basedMAC of the modified HCF scheme for local dataservices is not efficient when the channel contention

542 Mobile Netw Appl (2011) 16:531–543

among local data nodes is high (i.e., in a saturationcondition).

5.3 Delivery probability of delay tolerant data services

Consider the uplink transmission. The delivery proba-bility is evaluated in four cases with different configu-rations of network coverage area (ψ), uplink delaytolerant data service arrival rate (λU), and buffer size(BU), given as follows:

Case 1 ψ = 2,000 × 2,000 m2, λU = 0.03 messagesper second, BU = 800 messages;

Case 2 ψ = 2,000 × 2,000 m2, λU = 0.03 messagesper second, BU = 300 messages;

Case 3 ψ = 2,500 × 2,500 m2, λU = 0.03 messagesper second, BU = 300 messages;

Case 4 ψ = 2,500 × 2,500 m2, λU = 0.06 messagesper second, BU = 300 messages.

Figure 9 shows the message delivery probability ofuplink delay tolerant data service versus βUL. It isobserved that, the delivery probability increases as βUL

increases, since more radio resources are provided toa nomadic node when it comes into a hot spot. Thedelivery probability can also be improved with a largerbuffer size or a lower message arrival rate because ofthe lower chances for the buffer to be fully occupied.On the other hand, the network area has a negativeeffect on the delivery probability. For a given set ofRD mobility parameters, the sojourn duration is inde-pendent of the network area, but the inter-visiting timebecomes longer for a network with a larger area, result-ing in a lower delivery probability. It is observed that

0.05 0.1 0.15 0.2 0.25 0.30

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

βUL

Del

iver

y pr

obab

ility

Case 1 (analysis)Case 2 (analysis)Case 3 (analysis)Case 4 (analysis)

Case 1 (simulation)Case 2 (simulation)Case 3 (simulation)Case 4 (simulation)

Fig. 9 The delivery probability of uplink delay tolerant dataservice versus βUL

0 0.02 0.04 0.06 0.08 0.10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Message arrival rate (messages per second)

Del

iver

y pr

obab

ility

βUL

= 0.3

βUL

= 0.25

βUL

= 0.20

βUL

= 0.15

βUL

= 0.10

βUL

= 0.05

Fig. 10 The delivery probability of uplink delay tolerant dataservice versus message arrival rate

the analytical and simulation results closely match witheach other.

In order to evaluate how the effectiveness ofDFMAC changes with the uplink delay tolerant dataservice arrival rate, we fix the values of ψ and BU to2,500 × 2,500 m2 and 300 messages, respectively. Asshown in Fig. 10, with an increase of the message arrivalrate, the delivery probability of uplink delay tolerantdata service decreases. As expected, the delivery prob-ability can be increased by choosing a larger βUL value.Nonetheless, for a fixed message arrival rate, the degreeof the message delivery improvement dwindles whenthe value of βUL increases. The rationale is that, as βUL

increases, the limited buffer size gradually becomes thebottleneck of message delivery.

6 Conclusions

In this paper, a DTN/WLAN interworking scenario isconsidered. A DTN-friendly MAC (DFMAC) schemeis proposed to facilitate link-layer resource allocationfor local nodes and nomadic nodes in a hot spot. Bytuning the phase duration parameters of the DFMAC,the performance balance between local services anddelay tolerant data services can be achieved. The per-formance of the proposed MAC scheme is evaluated bymathematical analysis and further validated by simula-tions. Our results show that, by properly choosing thevalue of phase duration parameters, we can achieve sat-isfactory system performance for both nomadic nodesand local nodes. Further work includes the extensionof the DFMAC scheme to joint design with different

Mobile Netw Appl (2011) 16:531–543 543

store-carry-forward routing protocols and the investi-gation of wireless channel impairments.

Acknowledgements The authors wish to thank Mr. Ho TingCheng for his helpful suggestions which improved the quality ofthis paper. This work was supported by a research grant from theNatural Science and Engineering Research Council (NSERC) ofCanada.

References

1. Ali RB, Pierre S (2008) Optimal voice admission controlperformance under soft vertical handoff in loosely coupled3G/WLAN networks. In: Proc. IEEE WCNC’08. Las Vegas,pp 2980–2985

2. IEEE 802.11 WG. (2004) IEEE 802.11e/D11. IEEE Stan-dard for information technology-Telecommunications andinformation exchange between systems-Local and metropol-itan area networks-Specific requirements-Part 11: WirelessMedium Access Control (MAC) and Physical Layer (PHY)specifications: Amendment 7: Medium Access Control (MAC)Quality of Service (QoS) Enhancements.

3. Zhu H, Lin X, Shi M, Ho P, Shen X (2009) PPAB: aprivacy preserving authentication and billing architecturefor metropolitan area sharing networks. IEEE Trans VehTechnol 58(5):2529–2543

4. FON. http://www.fon.com5. Chen BB, Chan MC (2009) MobTorrent: a framework

for mobile Internet access from vehicles. In: Proc. IEEEINFOCOM’09. Rio de Janeiro, pp 1404–1412

6. Messier A, Robinson J, Pahlavan K (1997) Performancemonitoring of a wireless campus area network. In: Proc.IEEE LCN’97. Minneapolis, pp 232–238

7. Kachienga MO (2008) Technological innovations in healthcare: challenges of managing telehealth technology in SouthAfrica. In: Proc. IEEE IEMC’08. Estoril, pp 1–5

8. Fall K (2003) A delay-tolerant network architecture for chal-lenged internets. In: Proc. ACM SIGCOMM’03. Karlsruhe,pp 27–34

9. Zhang Z (2006) Routing in intermittently connected mobilead hoc networks and delay tolerant networks: overview andchallenges. IEEE Commun Surveys Tut 8(1):24–37

10. Jindal A, Psounis K (2009) Contention-aware performanceanalysis of mobility-assisted routing. IEEE Trans MobComput 8(2):145–161

11. Wang P, Jiang H, Zhuang W (2007) Capacity inprovementand analysis for voice/data traffic over WLAN. IEEE TransWirel Commun 6(4):1530–1541

12. Wang P, Zhuang W (2008) A token-based scheduling schemefor WLANs supporting voice/data traffic and its performanceanalysis. IEEE Trans Wirel Commun 7(5):1708–1718

13. Liang H, Zhuang W (2009) DFMAC: DTN-friendly mediumaccess control for wireless local area networks. In: Proc.IEEE CHINACOM’09. Xi’an, pp 1–7

14. Song W, Zhuang W (2009) Multi-service load sharing forresource management in the cellular/WLAN integrated net-work. IEEE Trans Wirel Commun 8(2):725–735

15. Cheng HT, Zhuang W (2009) Novel packet-level resourceallocation with effective QoS provisioning for wireless meshnetworks. IEEE Trans Wirel Commun 8(2):694–700

16. Li T, Logothetis D, Veeraraghavan M (2006) Analysis ofa polling system for telephony traffic with application towireless LANs. IEEE Trans Wirel Commun 5(6):1284–1293

17. Spyropoulos T, Jindal A, Psounis K (2008) An analyti-cal study of fundamental mobility properties for encounter-based protocols. Int’l J Auton Adapt Commun Syst(IJAACS) 1(1–2):4–40

18. Kleinrock L (1975) Queueing systems. Wiley, New York