On The Design of Opportunistic MAC Protocols - Copy

8/8/2019 On The Design of Opportunistic MAC Protocols - Copy

http://slidepdf.com/reader/full/on-the-design-of-opportunistic-mac-protocols-copy 1/16

1

On The Design of Opportunistic MAC Protocolsfor Multi-hop Wireless Networks with

Beamforming AntennasOsama Bazan, Member, IEEE, and Muhammad Jaseemuddin, Member, IEEE

Abstract —Beamforming antennas promise a significant increase in the spatial reuse of the wireless medium when deployed in multi-

hop wireless networks. However, existing directional Medium Access Control (MAC) protocols with the default binary exponential

backoff mechanism are not capable of fully exploiting the offered potential. In this paper, we discuss various issues involved in the

design of MAC protocols specific for beamforming antennas. Based on our discussion, we argue that the traditional binary exponential

backoff mechanism limits the possible spatial reuse and aggravates some beamforming-related problems such as deafness and head-

of-line blocking. To grasp the transmission opportunities offered by beamforming antennas, we design an Opportunistic Directional MAC

(OPDMAC) protocol for multi-hop wireless networks. The OPDMAC protocol employs a novel backoff mechanism in which the node

is not forced to undergo idle backoff after a transmission failure but can rather take the opportunity of transmitting other outstanding

packets in other directions. This mechanism minimizes the idle waiting time and increases the channel utilization significantly and

thereby enables OPDMAC to enhance the spatial reusability of the wireless medium and reduce the impact of the deafness problem

without additional overhead. Through extensive simulations, we demonstrate that OPDMAC enhances the performance in terms of

throughput, delay, packet delivery ratio and fairness. To further improve its performance, we discuss and evaluate the benefits ofcarefully choosing some protocol parameters instead of using the default values commonly used by other directional MAC protocols.

Index Terms —Beamforming antennas, Medium Access Control, Multi-hop Wireless Networks, Spatial reuse.

!

1 INTRODUCTION

THE increasing use of multi-hop wireless networksand the growing demand of bandwidth-intensive

network applications are the driving force to explore

innovative techniques that can enhance the networkcapacity. The commonly used omni-directional antennagenerates interference in all directions that can sig-nificantly limit the spatial reusability of the wirelessmedium. In this context, the beamforming antenna isa promising technology that enables directional trans-mission and reception which can enhance the spatialreusability and consequently the overall capacity of thewireless network [1]. However, traditional multi-hopwireless network protocols are not capable of exploitingthe benefits offered by beamforming antennas since theyare originally designed assuming that the nodes areequipped with omni-directional antennas.

Over the last few years, the research community has been working on developing Medium Access Control(MAC) schemes for multi-hop wireless networks with

beamforming (directional) antennas. A substantial number of these research efforts focus on adapting IEEE802.11 MAC to appropriately work with beamformingantennas. However, the scope of revisiting IEEE 802.11was primarily limited to choose whether to perform

• The authors are with the Department of Electrical and Computer Engi-neering, Ryerson University, Toronto, ON, Canada.E-mail: obazan, jaseem @ee.ryerson.ca.

the protocol operations with the antenna in a direc-tional or an omni-directional mode. For example, adirectional version of IEEE 802.11, known as the BasicDMAC protocol [2], has been proposed that performscarrier sensing, backoff and RTS/CTS handshake in a

directional mode. In the context of DMAC, some un-precedented beamforming-related challenges have beenidentified including deafness and directional hiddenterminal problems. Deafness occurs when a transmitterfails to communicate with its intended receiver becausethe receiver is beamformed away from the transmitter[3]. The transmitter interprets such failure as collisionand invokes the binary exponential backoff procedurewhich results in channel under-utilization, packet dropsand unfairness. The directional hidden terminal prob-lem occurs when a node is unable to hear RTS/CTSexchanged by a pair of communicating nodes and theninitiates a transmission causing collision to that ongoing

communication. Recently, researchers have focused onproposing directional MAC protocols that incorporateapproaches that resolve these beamforming-related prob-lems specially the deafness problem since it is the mostcritical problem [4]. The majority of the approacheshandle the possibility of deafness proactively by sendingmultiple directional control packets sequentially on other

beams to inform neighbors about the imminent dialog[4]–[7]. Although this strategy can reduce deafness, theoverhead could be high enough to offset the benefit of spatial reuse. Only few solutions have been proposed toreduce the impact of deafness reactively either by tone-

Digital Object Indentifier 10.1109/TMC.2010.68 1536-1233/10/$26.00 © 2010 IEEE

IEEE TRANSACTIONS ON MOBILE COMPUTING

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Authorized licensed use limited to: Vidya Academy of Science & Technology. Downloaded on June 30,2010 at 06:25:49 UTC from IEEE Xplore. Restrictions apply.



2

based notification [3] which is complex to implement or by a polling mechanism that relies on explicit next packetnotification [8].

Although the Binary Exponential Backoff (BEB) algo-rithm is a major component of the IEEE 802.11 MAC,it has rarely been revisited in the context of directionalMAC protocols. In this paper, we investigate this issueand show that the BEB algorithm is over-conservativeand should not be used in the presence of beamformingantennas. In addition to limiting the possible spatialreuse, this backoff algorithm contributes to the deafnessand the head-of-line blocking [9] problems. With the ob-

jective of enhancing the spatial reusability of the wirelessmedium, we design an Opportunistic Directional MAC(OPDMAC) protocol for multi-hop wireless networkswith beamforming antennas. OPDMAC aims to graspthe transmission opportunities offered by beamformingantennas while dealing with the beamforming-relatedchallenges, such as deafness, without the need for ad-ditional overhead. The OPDMAC protocol employs a

novel backoff mechanism in which the node is not forcedto undergo idle backoff after a transmission failure butcan rather take the opportunity of transmitting otheroutstanding packets in other directions. This backoff mechanism minimizes the idle waiting time, increasesthe channel utilization, reduces the impact of the deaf-ness and prevents the head-of-line blocking.

The paper is organized as follows. In Section 2, weprovide an overview of the related work. We investigatethe shortcomings of the directional MAC protocols thatuse the BEB mechanism in Section 3. In Section 4, wediscuss various issues involved in the design of MACprotocols to be used with beamforming antennas. Section

5 describes the operation of the proposed OPDMACprotocol. In Section 6, we investigate the careful choice of some protocol parameters rather than using the defaultvalues specified in the IEEE 802.11 standard. We evaluatethe performance of our OPDMAC protocol in Section 7.We discuss some miscellaneous issues in Section 8 andconclude the paper in Section 9.

2 BACKGROUND AND RELATED WOR K

Recently, several directional MAC protocols have beenproposed to exploit the benefits of smart beamformingantennas in multi-hop wireless networks. In this section,

we present a brief overview of IEEE 802.11 MAC as wellas the existing directional MAC protocols.

2.1 IEEE 802.11 DCF

The IEEE 802.11 [10] is the most widely used MAC pro-tocol in wireless networks. It is based on the concept of Carrier Sense Multiple Access with Collision Avoidance(CSMA/CA). A node that needs to access the wirelessmedium should perform physical carrier sensing forDIFS period before initiating transmission. To overcomethe hidden terminal problem, collision avoidance isimplemented by a handshaking mechanism (RTS/CTS)

before data transmission [11]. Both RTS and CTS packetscontain the proposed duration in which neighboringnodes must defer transmission. This is called virtual car-rier sensing and is implemented through a mechanismcalled the Network Allocation Vector (NAV).

The IEEE 802.11 MAC protocol uses a backoff mech-anism to resolve channel contention. If the channel isfound busy during the physical carrier sensing, the nodechooses a random backoff interval from [0, CW ], whereCW is called the contention window. After every idleslot time, the node decrements the backoff counter byone. When it reaches zero, the node can transmit itspacket. In case a CTS or ACK packet is not received back,the node assumes a collision has occurred with someother transmission and it invokes the binary exponential

backoff algorithm. In this algorithm, the node doubles itsCW after each collision, chooses a new backoff intervaland tries retransmission again once the backoff timerexpires. Once a packet is successfully transmitted, CW

is initialized to its minimum value.

2.2 MAC Using Beamforming Antennas

Ko et al. [12] propose a directional MAC in which RTS issent directionally (DRTS) if one of its beams is blocked,to avoid unnecessary waiting time. Nasipuri et al. [13]consider the case where the location information maynot be available and propose to send both RTS andCTS omni-directionally. The data and its acknowledge-ment are exchanged directionally in order to reduceinterference. However, the spatial reuse is limited bythis conservative channel reservation. Elbatt et al. [14]propose adding new fields in RTS/CTS to be used for theimminent communication. Upon receiving RTS/CTS, the

neighbor can take an appropriate antenna blocking deci-sion. Wang and Garcia-Luna-Aceves [15] investigate theinteraction between spatial reuse and collision avoidanceand conclude that the DRTS/DCTS scheme outperformsconservative collision-avoidance schemes. Bandyopad-hyay et al. [16] present a MAC protocol that employsadditional messages to inform the neighborhood nodesabout ongoing communications.

Takai et al. propose the concept of Directional Vir-tual Carrier Sensing (DVCS) in [17]. If a node receivesRTS/CTS from a certain direction, it needs to defer trans-missions only in that direction in which other commu-nication is in progress. The DVCS is implemented using

a Directional NAV (DNAV) mechanism. Along the sameline, Choudhury et al. in [2] propose the Basic DMACprotocol which is commonly used as the benchmarkfor directional MAC protocols. It employs the DNAVmechanism and performs carrier sensing, back-off, andthe four-way handshake in a directional mode.

Dealing with the problems exclusive to the use of beamforming antennas [18] like deafness and the di-rectional hidden terminals, researchers have started de-signing directional MAC protocols that address theseissues. Choudhury and Vaidya in [3] propose ToneD-MAC to handle the deafness problem. ToneDMAC uses






3

a tone-based notification mechanism that allows theneighbors of a node to distinguish collision from deaf-ness. However, the implementation of the protocol iscomplex. Korakis et al. in [5] propose sending Circu-lar Directional RTS (CDR) sequentially over all beams.By informing neighbors in other directions about theupcoming transmission, the problems of deafness anddirectional hidden terminals can be reduced. However,this approach tends to increase the control overheaddrastically. Jakllari et al. in [6] propose sending circularRTS and circular CTS packets prior to data transmission.Gossain et al. in [7] propose that the sender and thereceiver transmit the circular redundant RTS and CTSpackets simultaneously after they successfully exchangethe single directional RTS/CTS. This ensures the circularoverhead packets are only transmitted after the originalRTS is successfully received. In [4], Takata et al. proposelimiting the transmission of the circular control packetsonly to potential transmitters (potentially suffering fromdeafness) to reduce the MAC overhead. The potential

transmitter is selected either based on the history of previous communications or by means of explicit nextpacket notification. In [8], Takata et al. propose a polling-

based directional MAC protocol in which the node pollsone of its neighbors that were possibly suffering fromdeafness. RamMohan et al. in [19] address the problemof hidden terminals due to unheard RTS/CTS. They pro-pose a protocol in which the data packet is fragmentedand tones are sent by the receiving node in the periods

between the fragments. Their results show a significantdecrease in the number of collisions but a marginalimprovement in the throughput and delay performancesince their protocol aggravates the deafness problem. In

[20], Subramanian and Das address deafness and hid-den terminal problems by separating the transmissionof control and data packets in time. Several RTS andCTS are exchanged omni-directionally within a controlwindow duration, followed by concurrent directionaltransmission of the DATA packets.

Kolar et al. [9] identify the Head of Line (HoL) block-ing problem when beamforming antennas are used. Theexisting link layer implementations, that is based on FirstInput First Output (FIFO) queuing, lends itself to HoL

blocking if the medium is sensed busy in the directionof the packet at the top of the queue but is available inother directions. Based on the DNAV table, the authors

propose using the minimum waiting time to select thefirst packet for transmission. The proposed scheme doesnot consider the effect of deafness, which may cause theDNAV entries to be invalid.

While the binary exponential backoff may not be the best choice to be used with directional antennas, thisissue has rarely been addressed. In [21], Ramanathanet al. propose a new backoff algorithm (called forcedidle) in which the duration and the window adjustmentmechanism depend on the type of event causing the

backoff, for example whether the event is busy channel,missing CTS, or missing ACK. If the channel is sensed

busy, the contention window remains constant. If CTSis found missing, the value of the contention windowis increased linearly. When ACK is absent, the increaseof contention window is exponential. Upon receivingan ACK, the value of the contention window is de-creased exponentially. The rationale behind this backoff mechanism is not discussed and the evaluation doesnot provide any insights about its effectiveness. In [22],we have presented a basic description of the OPDMACprotocol with its novel active backoff mechanism. Inthis paper, we explain the rationale behind our backoff mechanism, discuss various design issues involved inthe design of the OPDMAC protocol and examine someimplementation choices and their tradeoffs. In addition,we evaluate the protocol performance using extensivesimulations and several performance metrics.

To exploit the benefit of higher communication rangeat the MAC layer, Choudhury et al. in [18] proposeMMAC protocol that aims to transmit a data packetover the longest possible hops by relaying the RTS over

multiple hops. Other researchers propose the use of directional idle listening [23], [24] in which the receivingantenna is always in a directional mode but continuouslysweeps in all directions sequentially. Throughout theliterature, there have been few MAC protocols that relyon the use of busy tones [25], [26]. However, they requirean additional control channel that adds to the transceivercomplexity. Other researchers looked into using moresophisticated antenna model known as multi-beam an-tennas to further improve the performance [27]–[29]. As-suming the availability of system-wide synchronization,few synchronous directional MAC protocols have beenproposed. In [29], Bao et al. propose a TDMA protocol

that is capable of exploiting the multi-beam capabilityin both transmission and reception. Wang et al. in [30]propose a synchronous directional MAC with three timephases: random access, DATA and ACK. In [23], Jakllariet al. propose a synchronous polling-based MAC inwhich time is divided into contiguous frames. Eachframe is divided into three segments: search, polling anddata transfer. Although these protocols can reduce theeffect of both the deafness and hidden terminal problem,achieving network wide synchronization is consideredimpractical in multi-hop wireless networks.

3 PROBLEM FORMULATIONIn contrast to omni-directional antennas, beamformingantennas can allow multiple concurrent transmissionswithin the same neighborhood. However, the MAC layershould be able to exploit this potential benefit. In thissection, we discuss some limitations of the existing direc-tional MAC protocols by considering three different sce-narios. Our discussions are in the context of the DMACprotocol with the default binary exponential backoff procedure which is the basis of most of the proposedprotocols. Consider the directional hidden terminal sce-nario of Fig. 1, where node S 1 communicates with both






4

1.pdf

D1

D

D2

S1

S2

Fig. 1. A directional hidden terminal scenario.

2.pdf

D1

D

D2

S1

S2

Fig. 2. A deafness scenario.

node D and node D1 while node S 2 communicates with both node D and node D2. The use of IEEE 802.11MAC in an omni-directional mode limits the spatialreuse as it permits only one transmission at a time.However, beamforming antennas can allow two trans-missions concurrently provided that both transmissionsare not targeted to the same receiver D. If DMAC is inoperation, the transmission scenario is as follows: if bothnodes have a packet directed to node D at the head of their respective queues, a collision is likely to happen atnode D. As a result, both nodes backoff exponentially

before contending again for the channel. Due to thedirectional transmission, each node is a hidden terminalwith respect to the other although they are within thetransmission range of each other. Eventually, one node(say node S 1) succeeds in transmitting DRTS that nodeD responds to with DCTS. Upon hearing DCTS, nodeS 2 freezes its backoff counter and waits in an idle stateuntil the other transmission finishes. With IEEE 802.11,when a node overhears the RTS sent by the other node, itdoes not initiate its transmission. However with DMAC,the node cannot hear DRTS but halts its transmissionwhen it receives the receiver’s DCTS. It is obvious thatsuccessive retries in the same direction is not a good

approach especially if the node has other packets in thequeue outstanding for transmission in other directions.This may limit the ability of DMAC to gain from thespatial reuse benefit. On the other hand, forcing the nodeto keep silent until the other transmission ends shouldnot be a mandatory requirement in directional MACprotocols because transmitting in another direction maynot affect the ongoing transmission.

Fig. 2 shows a typical deafness scenario. Similar tothe previous scenario, node S 1 communicates with bothnode D and node D1 while node S 2 communicateswith both node D and node D2. However, node D

3.pdf

B

D

E

C

F

A

Fig. 3. A scenario illustrating the trade-off between direc-tional backoff and omni-directional backoff.

communicates with node S 1 and node S 2 using twodifferent beams. If node D is engaged with node S 1, it

becomes deaf to node S 2. According to DMAC, node S 2continues attempting transmission towards node D, andas a result experiences repeated backoff. Since retrans-mission attempts increase the subsequent backoff peri-ods exponentially, it is likely that node S 2 lies in waitingfor a long backoff period when node S 1 completes itstransmission to node D. As a consequence of exponential

backoff, nodeS 1

may succeed in transmitting subsequentpackets to node D, if they are ready in its queue, beforenode S 2 is even able to contend for the channel towardsnode D after finishing its backoff. In an extreme situationof continuous backoff, node S 2 may eventually drop thepacket if it reaches the maximum retry limit. This exam-ple shows why deafness is a major drawback of BasicDMAC as it causes large delays and potentially packetlosses. Although some recent protocols have addressedthe deafness problem by informing other nodes aboutthe deafness duration, the long failure recovery time isone of the disadvantages common to all such protocolsthat occurs as a result of wasting the time insisting on

communicating with a deaf node first.Consider the scenario shown in Fig. 3, where node

A has packets to send to each of node B and nodeC . Also, each of nodes B and C has its own flowto node D while node E communicates with node F .When omni-directional antennas are used, node A hasto contend for the channel access with nodes B, C

and E . On the other hand, if the nodes are equippedwith beamforming antennas, the spatial reusability can

be enhanced. Ideally speaking, when the transmissionand reception occur directionally, flow E -F should notinterfere with node A’s transmission to either node B

or node C . Moreover, flows A-B and C -D (or flows A-

C and B-D) can be active simultaneously. The actualperformance of the network with directional antennaslies in the operation of the MAC protocol. In partic-ular, the antenna mode (directional/omni-directional)during the backoff phase is a critical trade-off. In theinterest of higher spatial reuse, Basic DMAC performsthe directional backoff. Remaining in the directionalmode can prevent a node from getting unnecessarilycaptured by surrounding communications. In this case,node A is not affected by the transmission of node E

to node F as long as node A has packets in its queue.However, both nodes B and C appear deaf to node






5

A even though only one of them is busy with nodeD because the other nodes experiences backoff at thattime. Using directional backoff (i.e. spending the backoff phase in a directional mode), the drawbacks of persistentdeafness overcome the benefits of interference reduction.Hence, several directional MAC protocols perform the

backoff phase in an omni-directional mode [2]–[4], [7],[21]. Although persistent deafness of nodes B and C can

be alleviated by employing the omni-directional backoff,node A may be unnecessarily captured by flow E -F every time it performs omni-directional backoff. With thedefault binary exponential backoff mechanism, node A

remains susceptible to flow E -F for a longer period of time, which eventually reduces the spatial reuse. Usingthe existing directional MAC protocols, if node A fails toestablish communication with node B because of nodeB’s communication with node D, node A begins theidle omni-directional backoff phase instead of benefitingfrom sending to node C which is in the omni-directional

backoff phase.

It is obvious from the previous discussion that the binary exponential backoff algorithm is not adequatein the presence of beamforming antennas. This raisesseveral questions: What is the backoff mechanism thatcan enhance the spatial reuse? Can an antenna-aware

backoff algorithm alleviate the impact of deafness? Whatis the relation between the backoff state and the idlestate? What is the suitable packet scheduling policy? Inthe following sections, we address these issues with thegoal of designing a directional MAC protocol for multi-hop wireless networks with beamforming antennas.

4 MAI N DESIGN CONSIDERATIONS

In this section, we present a set of observations thatprovides the basis for our proposed OPDMAC protocol,which we define in the next section.

In the case of omni-directional antennas, collisionis the major reason for transmission failures. Hence,the binary exponential backoff algorithm is needed forcontention resolution. Moreover, since a transmissionreaches all the receivers in the sender neighborhood, theidle backoff phase is mandatory when a transmissionfailure occurs. On the other hand, when beamformingantennas are used, the channel is spatially divided anda transmission in one direction is not sensed in other

directions. This major benefit of beamforming antennasshould be fully utilized. A missing acknowledgmentindicates a transmission failure that could be due tocollision or deafness. In either case, the receiver is notcurrently ready to receive the packet and the sendershould halt the packet retransmission for a certain periodof time, as it happens in IEEE 802.11 backoff process thatis typically employed by all directional MAC protocols.However, the directional MAC protocol should not forcethe sender to remain idle during this backoff period asimplicitly assumed by the existing protocols. Remainingidle during the backoff period introduces unnecessary

blocking time that results in channel underutilizationand a significant increase in the delay. Instead, duringthe period the node is forced to backoff from transmit-ting in one direction as a result of transmission failure,it can take the opportunity of attempting transmissionof other outstanding packets in other directions. Inother words, the need to backoff for a random periodof time before retransmission in one direction shouldnot block packet transmissions in other directions. Thisactive backoff procedure helps in enhancing the spatialreusability of the wireless channel to a great extent.

Observation #1: In case of a missing acknowledgment,the node should not be forced to remain idle between theretransmission attempts as long as it has other packets totransmit in other directions.

In IEEE 802.11 MAC, each node should go into anidle backoff phase after each successful transmission toensure that backlogged nodes do not take control of themedium for long periods of time. This seems to be un-necessary in the case of beamforming antennas. The timethe node spends in transmitting a packet in one directioncan serve as a backoff duration for channel contentionin another direction. Hence, contention resolution can beachieved by avoiding successive packet transmission inone direction rather than forcing the node to remain inan idle state. However, a critical deafness problem couldarise. If a node has a backlog of packets to differentneighbors residing in few directions, it would appeardeaf to all the nodes in the other directions. This problemis similar to the original deafness problem in the contextof DMAC [3]. The commonly adopted solution by mostexisting directional MAC protocols is to perform the

backoff phase in an omni-directional idle state, whichseverely reduces the spatial reuse. The idle backoff,though seems beneficial, is prohibitive of taking advan-tage of observation #1, therefore, we suggest that allevi-ating the deafness chains should be decoupled from the

backoff algorithm. To alleviate the persistent deafness,each node should regularly listen to the medium omni-directionally. Although this listening phase resembles theIEEE 802.11 backoff phase, its rationale and overheadare substantially different. The listening phase is neededwith beamforming antennas to reduce the transmissionfailures due to deafness and to allow each node toupdate its channel state information, which is signifi-

cantly different from the idle backoff algorithm that isoriginally designed for contention resolution. Moreover,the frequency of its occurrence is also different since itis not essential to enter the listening phase after eachtransmission failure as in the case of the IEEE 802.11

backoff phase.

Observation #2: Each node should regularly visit an omni-directional idle state to prevent persistent deafness.

Observation #3: In the case of beamforming antennas, thebackoff phase and the listening phase should be decoupled.

The deafness problem is the most critical challenge






6

facing multi-hop wireless networks with beamformingantennas. Although deafness occurs as a result of atransmission failure when the receiver is beamformedtowards another direction, the sender’s reaction escalatesthe problem. Upon detection of the failure, the binaryexponential backoff algorithm is invoked resulting inchannel underutilization, degradation in the networkcapacity, increase in the packet drops and unfairness inchannel access. Most of the solutions proposed in theliterature focus on reducing the occurrence of deafness

by informing neighboring nodes about the upcomingtransmission that may lead to deafness. This includeseither omni-directional RTS/CTS transmission [13] orsequential directional RTS/CTS transmission over beamsother than the receiver’s direction [4]–[7]. Although theseapproaches may reduce the occurrence of deafness, theycannot completely eliminate it as the overhead controlpackets may suffer from deafness themselves in additionto possible collisions. However, the main drawback of these techniques is the additional overhead that reduces

the network capacity and increases the delay [4], whichessentially offsets the benefits of spatial reuse that theytry to exploit. Therefore, we need to address the deafnessproblem with either no or substantially reduced addi-tional overhead so that the gain due to spatial reuse isnot offset. Without the need of a deafness notification,we suggest the node that detects a packet failure shouldreact in a way that alleviates the negative impact of deafness. The ideal behavior should minimize the block-ing time, avoid the channel underutilization, reducethe correlation between the retransmissions, and avoidinvolving in an unfair backoff. The rationale behindour approach of having each node relieves the deafness

problem on its own is the fact that deafness has noharmful impact on any other ongoing communication.In contrary, the hidden terminal problem may becomemore destructive by harming the ongoing transmission;therefore, a node is required to inform the neighborhooda priori to protect its transmission.

Observation #4: Each node should react to transmission failures in a way that mitigates the impact of deafness.

Observation #5: To leverage the benefit of spatial reuse,overhead should be minimized.

The use of beamforming antennas introduce new hid-den terminal problems in which the regular RTS/CTS

fails to inform the hidden nodes about the ongoingcommunication. The hidden terminal problem due toasymmetry in gain is shown to be very rare [18], whilecollisions due to unheard RTS/CTS can occur morefrequently. In this case, the virtual carrier sensing fails

because it was performed while some nodes are beam-formed towards other directions. This results in a lossof the channel state information whenever the node is

beamformed. To reduce the effect of the hidden terminalproblem, the node should try to retrieve its channel stateinformation before each transmission attempt.

Observation #6: To reduce the hidden terminal problem

due to unheard RTS/CTS, the node should try to retrieve itschannel state information before transmission.

Considering no additional QoS requirement, the FIFOqueuing policy works fine in the case of omni-directionalantennas since all outstanding packets use the samemedium. If the medium is busy, no packet can betransmitted. However, in case of beamforming antennas,

FIFO leads to the HoL blocking problem [9]. In order toimprove the spatial reuse, the packet scheduling policyshould not block the transmission of any ready packet.

Observation #7: The packet scheduling policy shouldenable the transmission of any ready packet, thus eliminatesthe HoL blocking.

5 PROTOCOL DESCRIPTION

Based on the set of observations presented in the pre-vious section, we propose an opportunistic directionalMAC protocol called OPDMAC for multi-hop wirelessnetworks with beamforming antennas. The OPDMAC isa contention-based directional MAC protocol that aimsto maximally harness the benefits of spatial reuse byminimizing the idle waiting time and exploring thenew transmission opportunities. Although not manda-tory, OPDMAC employs RTS/CTS exchange before datatransmission. All messages are sent directionally whilethe idle node listens to the medium in an omni-directional mode. The OPDMAC uses the DNAV mech-anism [17] for the directional virtual carrier sensing. Weassume that an upper layer (e.g. routing layer) is capableof providing OPDMAC with neighbors’ directions. Thisassumption is common among various directional MAC

protocols [2]–[4], [7], [8], [12], which is justified becauserouting protocols usually learn the direction of neighborsthrough the reception of control packets such as routerequest and route reply packets during route discoveryor periodic HELLO packets.

5.1 RTS Transmission

When a node ends its listening period, which is de-scribed in Section 5.5, it scans the packets in its non-empty link layer queue sequentially in the order of their arrival time to pick the first unblocked packet fortransmission. The node attempts to transmit the packet

by beamforming in the direction of the intended receiverand starting directional carrier sensing. If the mediumis sensed idle for a DIFS period, the node transmitsRTS packet. If the medium is sensed busy during thecarrier sensing, the node has to defer transmission onthis beam, however, it can still transmit over other

beams. Accordingly, OPDMAC allows the node to rescanits queue and chooses the next unblocked packet andattempts transmitting it. The node beamforms in the newdirection and starts the carrier sensing again. When thenode succeeds in sending RTS, it initiates a wait-for-CTStimer and remains beamformed in the same direction.






7

5.2 RTS Reception and CTS Transmission

When a node receives RTS intended for the node itself, itperforms a directional carrier sensing for a SIFS period.If the medium is sensed idle, it sends CTS in response.All other nodes that receive RTS not destined to them,update their DNAV table.

5.3 CTS Reception and DATA/ACK Exchange

Similar to most directional MAC protocols, when thesender receives CTS within the CTS-timeout duration, itsends the DATA packet after SIFS period. Upon receivingDATA, the receiver responds with the ACK packet indi-cating successful reception of the DATA packet. All othernodes that hear CTS, DATA and ACK packets, updatetheir DNAV tables accordingly.

5.4 Missing CTS

If the sender does not receive CTS within the CTS-

timeout duration, this means that the receiver is notcurrently ready for receiving the DATA packet. Sincethe sender could not distinguish between deafness andcollision, it should not continue contending for thechannel in this direction for a certain period of timesimilar to the backoff process of IEEE 802.11 generallyemployed by all directional MAC protocols. But, insteadof forcing the sender to remain idle potentially goingthrough exponential rounds of backoff, the OPDMACallows the node to recheck its queue and try sendinganother packet in a different unblocked direction. The

period the node spends to transmit a packet in the seconddirection serves as a backoff period for the first direction.

This novel mechanism allows the node to be activeduring the backoff state and hence is able to minimizethe delay and enhance the spatial reuse significantly.Figure 4 shows an illustrative example. When node A

fails to communicate with node B, it opportunisticallyreplaces the traditional idle backoff time by a usefultransmission of the packet destined to node D. Afternode A completes its transmission attempt to node D,it retransmits the packet destined to node B.

Occasionally, the node may not find any unblockedpacket so it is forced to enter an idle backoff state. InOPDMAC, the node backs off in an omni-directionalmode for a random time derived from a constant con-

tention window. Thus, it does not exponentially increasecontention window with every round of backoff. The ra-tionales for keeping the contention window constant areas follows. First, if the RTS failure is due to collision, thecontention will likely dissipate because other contendingnodes may contend for the channel in other directionsas a result of finding unblocked packets after rescanningtheir queues. Second, if the CTS is not returned due todeafness of the receiver, the binary exponential backoff mechanism usually prolongs the deafness-related delay.Using the same rationales, constant backoff mechanismis also employed in case of retransmission caused by

missing ACK. Thus, the idle backoff in OPDMAC issubstantially different from the backoff of IEEE 802.11 intwo ways: (a) it occurs rarely after finding no unblockedpacket instead of transmission failure, and (b) it employsconstant backoff time instead of rounds of exponen-tial backoff periods. This is another novel feature of OPDMAC, which is introduced to minimize the impactof backoff period (control overhead) on the throughputand delay of the node.

5.5 The Listening Period

After each successful transmission, the node is forcedto remain idle for a certain period of time called theListening Period (LP) even if it has packets outstand-ing for transmission. During the LP, the node listensin an omni-directional mode. The LP is essential tomitigate persistent deafness by allowing other nodes tocommunicate with the deaf node. Also, overhearing themedium is beneficial because the node needs to collectuseful information about its neighborhood to retrieve

the channel state information. For example, it has toupdate its DNAV table which is likely to be outdatedas a result of previous beamforming. Although thisidle period trades off the spatial reuse, it is necessaryto eliminate persistent deafness. In contrast to otherdirectional MAC protocols that employ omni-directionalexponential backoff after each transmission failure, theLP, that is derived from a constant window, is neededafter each successful transmission. At the end of the LP,the FIFO policy is reinforced. The node scans the packetsin the order of their arrival time and transmits the firstunblocked packet as mentioned in Section 5.1.

6 SOM E IMPLEMENTATION DETAILS

Since the IEEE 802.11 default parameters and timingsare originally defined considering omni-directional an-tennas, they must be reviewed for beamforming anten-nas. Although OPDMAC can perform correctly usingthe default values, performance improvement can beachieved if those values are adjusted to address thenew challenges. In this section, we discuss the impactof the OPDMAC parameters on its operation and thusthe overall performance.

A node can transmit if it senses the medium idlefor DIFS period which is 50 μs in the case of IEEE

802.11 Direct Sequence Spread Spectrum (DSSS) [10].When omni-directional antennas are used, this periodis more than enough to ensure there is no ongoingcommunication on the channel. In the case of directionaltransmission and reception, the virtual carrier sensingmay fail in informing all the neighbors about the on-going communication due to the possibility of unheardRTS/CTS if any of the neighbors is beamformed towardsanother directions at that time. To reduce the impact of this kind of hidden terminal problem, the DIFS period

before each transmission attempt should be prolonged toavoid colliding with a possible ongoing communication.






8

4.pdf

RTS RTS RTS RTS CTS ACK DATA RTS CTS ACKDATA

RTS CTS ACK DATA

Directional MACwith traditional

backoff

RTS RTS RTS CTS ACK DATA

RTS CTS ACK DATA

CTS ACK DATA

B

A

B

A

OPDMAC with

opportunisticbackoff

Communication with C

Communication with C

Communication with D

Communication with D

Communication with B

Communication w ith BFailed

to B

Reduced overall delay

T

T

T

T

A

B

C

D

C

B

D

Fig. 4. A scenario illustrating the benefits of the opportunistic backoff employed by OPDMAC

To completely solve this hidden terminal problem, theDIFS period should be at least equal to the transmissiontime of a DATA packet but this is a very long periodfor a node to wait. A large DIFS duration incurs asignificant delay that may be unnecessary when thereis no ongoing communication to avoid, which reducesthe channel utilization and increases deafness. Choosingthe value of DIFS period is a trade-off between reducingthe probability of collisions due to unheard RTS/CTSand increasing the delay. In [19], the authors show thatpreceding any transmission with a pause period that

is equal to the transmission time of a CTS packet (304μs) yields the best performance. Hence, we use a DIFSperiod of 300 μs in our simulations.

When a node does not receive CTS within CTS-timeout duration, the RTS packet is considered lost. Thevalue of CTS-timeout is not specified in the IEEE 802.11standard but it is usually defined as SIFS + T ACK (whereT ACK is the duration of the ACK packet), which is equalto 314 μs in the case of IEEE 802.11b [31]. This value isconsidered too large duration since the CTS is expectedto be received after SIFS + Round Trip Time (RTT). Inthe case of omni-directional antennas, such unnecessary

waiting time does not have a significant effect since thenode experiences idle backoff for a random period of time before initiating a retransmission. On the contrary,with beamforming antennas, a large CTS-timeout couldresult in unnecessary idle waiting time that could un-derutilize the spatial reuse. In OPDMAC operation, if a CTS-timeout counter expires, the node is expected toseek another transmission in another direction. Hence,a longer than necessary CTS-timeout duration degradesthe performance of OPDMAC. The reduction in the CTS-timeout duration can compensate for the increase in theDIFS period proposed to reduce the hidden terminal

problem. In this work, we use a CTS-timeout equal toone slot time which is 20 μs.

Although OPDMAC eliminates the need for an idle backoff after transmission failure, it may sometimes benecessary to enter an idle backoff state if the nodedoes not have outstanding packets for transmission inother directions. In this case, the node has to backoff in an omni-directional mode for a random time derivedfrom a constant Contention Window (CW). However,one important difference from the IEEE 802.11 backoff is that there is no need to freeze the backoff counter if a

carrier is sensed in other directions. If the counter expiresduring a packet reception, the node should wait until itcompletely receives the incoming packet. If the packetis destined to the node, it will respond. Otherwise, thenode just updates its DNAV and then beamforms to startthe directional carrier sensing preceding the retransmis-sion. The initial CW specified in the IEEE 802.11 standardis [0, 31]. In OPDMAC, a very small backoff intervalmight be counterproductive because it might expire even

before the contention is dissipated or the receiver comesout of deafness. Hence, we propose using a lower boundof the backoff CW. Based on extensive simulations, weadopt a backoff CW of [16, 31].

The OPDMAC requires a node to remain idle for aListening Period (LP) after each successful transmission.Similar to the backoff in IEEE 802.11, the node freezesits counter if the medium is sensed busy. A longer LPreduces the probability of deafness at the expense of additional delay that may decrease the spatial reuse.The trade-off in choosing the value of LP window isevaluated in the next section.

7 PERFORMANCE EVALUATION

In this section, we evaluate the performance of ourOPDMAC protocol. We use OPNET 12.1 [32] as our






9

5.pdf

2 3 4 5 6 7 8 9 100

1

2

3

4

5

6

Total Offered Load (Mbps)

A g g

r e g a t e T h r o u g h p u t ( M b p s )

OPDMAC

DMAC

802.11

Fig. 5. Aggregate throughput for the scenario in Fig. 1.

6.pdf

2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2


A v e r a g e D e l a y ( S e c )

OPDMAC

DMAC

802.11

Fig. 6. Average delay for the scenario in Fig. 1.

network simulator. We implemented a smart antennawith directional gain of 10 dB and beamwidth 60

o inOPNET using its powerful antenna pattern editor. Wealso implemented several directional MAC protocolsusing OPNET Modeler. To focus on the benefits of thespatial reuse, we set the communication range for bothdirectional and omni-directional protocols to 250 m. Thepacket size is 1024 bytes and the data rate is 11 Mbps.We do not consider node mobility in our simulations.In the first set of experiments, we show the benefitsof OPDMAC using the default values and timing asspecified in the IEEE 802.11 standard. Next, we evaluatethe impact of changing these values on the performanceof OPDMAC, which is discussed in Section 6.

7.1 Simple Topologies

The use of beamforming antennas introduces new chal-lenges such as deafness and directional hidden terminalproblems. The impacts of these problems are highlydependent on the topology and the traffic flows. In orderto evaluate the performance of OPDMAC in the presenceof those challenges, we first simulate few simple sce-narios to illustrate the issues underlying each problemseparately and then we evaluate the overall performancein larger scenarios.

First, we simulated the scenario shown in Fig. 1. In thisscenario, we consider the directional hidden terminalproblem due to unheard RTS/CTS as well as the HoL

7.pdf

2 3 4 5 6 7 8 9 100

1

2

3

4

5

6

7


A g g

r e g a t e T h r o u g h p u t ( M b p s )

OPDMAC

DMAC

802.11

Fig. 7. Aggregate throughput for the scenario in Fig. 2.

8.pdf 2 3 4 5 6 7 8 9 10

0

0.2

0.4

0.6

0.8

1

1.2



OPDMAC

DMAC

802.11

Fig. 8. Average delay for the scenario in Fig. 2.

blocking. Node S 1 communicates with both node D andnode D1 through two different flows. Also node S 2establishes separate flows with node D and node D2.The traffic of each flow follows Poisson distribution. Theaverage sending rate of the flow (S 1 →D) is four timesthe average sending rate of the flow (S 1 →D1) to model

bursty traffic in the direction of contention. As for nodeS 2, the average sending rate of the flow (S 2 → D2) isfour times the average sending rate of the flow (S 2 →D)to model bursty traffic in the contention-free direction.The total offered load is the sum of the rates of the fourflows.

Fig. 5 shows the aggregate throughput versus theoffered load. As expected, IEEE 802.11 performs theworst because there is no possible spatial reuse. Atlow loads, the performance of OPDMAC is similar to

DMAC because there are no available packets to offerthe opportunity exploited by OPDMAC. However, asthe load increases, the OPDMAC outperforms DMACsince it does not force the nodes to resort to idle backoff when they experience contention in one direction. Thisis beneficial not only because node S 2 is able to sendin the contention free direction (i.e. to node D2) butalso it withdraws from contending with node S 1 for thesame target (node D) allowing node S 1 to communicatecontention free with node D. Fig. 6 shows the averagedelay versus the offered load. It is clear that OPDMACachieves the minimum delay since it tries to minimize






10

the channel idle time.

To evaluate the performance of OPDMAC in a deaf-ness scenario, we simulated the scenario shown in Fig.2. The traffic configuration is similar to the previousscenario. Fig. 7 shows the aggregate throughput versusthe offered load in the deafness scenario. The OPDMACoutperforms the other two protocols because it is more

effective in exploiting the available spatial reuse. Fig. 8shows the average delay versus the offered load for thesame deafness scenario. The DMAC experiences highdelay due to the consecutive failures experienced whiletrying to communicate with a deaf node resulting ina large waiting time between packet transmissions. Incontrast, the OPDMAC protocol avoids being locked intrying to establish communication with a deaf node,which results in achieving a lower delay.

Next, we simulated the scenario shown in Fig. 3.Our objective is to evaluate the trade-off between thedirectional backoff as in the Basic DMAC [2] and theomni-directional backoff represented by DMAC-OM-BO.We consider node A with two backlog flows destinedto node B and node C . Flow E -F is considered aninterfering flow that could affect the spatial reuse gain.Flows B-D and C -D are deafness flows with respectto flows A-B and A-C respectively. To avoid beingdistracted by the interfering flow, node A should backoff in a directional mode. However, to reduce the impact of deafness on flows A-B and A-C , nodes B and C should

backoff omni-directionally. Since nodes A, B and C arerunning the same MAC protocol, there should be a trade-off in choosing the antenna mode during the backoff phase. Our OPDMAC protocol addresses this trade-

off by decoupling the listening period and the backoff phase. Table 1 shows the aggregate throughput of flowsA-B and A-C under different offered load for the in-terference and deafness flows. In the case of DMAC,although the interference flow E -F has no effect sincenode A backs-off directionally, the aggregate throughput

becomes zero when the load of deafness flows is high.This is the consequence of the directional backoff of nodes B and C . For the case of DMAC-OM-BO, nodeA can communicate with both node B and node C

even under highly loaded deafness flows because of theomni-directional backoff. However, the flow E -F has a

TABLE 1Aggregate throughput of flows A-B and A-C for the

scenario in Fig. 3

Interferenceflow E -F

load

Deafnessflows B-Dand C -D

load

DMAC DMAC-OM-BO OPDMAC

(Mbps) (Mbps) (Mbps) (Mbps) (Mbps)

0.4 0.4 0.65 0.69 3.19

0.4 4 0 0.58 1.8

4 0.4 0.65 0.33 2.1

4 4 0 0.3 1.27

significant effect on node A’s transmission since node A

is susceptible to the interfering flow during the backoff periods. On the contrary, OPDMAC outperforms bothprotocols as a result of its novel backoff mechanism.For the case of nodes B and C , when one of themsuffers from transmission failure, it is forced to backoff omni-directionally since it has no other packets to send.On the other hand, when node A fails to transmit apacket to one of the nodes, it does not go to an idle

backoff but explores the opportunity of transmitting apacket to another node in a different direction. Thisincreases the chance of having flows A-B and C -D (andflows A-C and B-D) active simultaneously. Moreover,the OPDMAC limits the period in which it is affected byflow E -F to the Listening Period (LP) instead of the long

backoff periods of DMAC-OM-BO. Hence, OPDMACexploits the spatial reuse more effectively by reducingthe impact of both interference and deafness.

7.2 Random Topologies

In the next set of experiments, we evaluate the per-formance of the OPDMAC protocol in a large multi-hop network. We compare OPDMAC with Basic DMACprotocol, DMAC protocol with omni-directional backoff (DMAC-OM-BO), Circular Directional RTS (CDR) MACprotocol [5] and the IEEE 802.11 standard. In a randomnetwork, the challenges are more complex but the ad-ditional transmission opportunities can provide morespatial reuse gain. We simulated a network with 30nodes randomly placed in an area of 1000 m X 1000m. The results are averaged over 10 different simulationruns. We evaluate the performance for both one-hop

flows and multi-hop flows.

7.2.1 One-hop flows

In each simulation run, 10 out of the 30 nodes arerandomly chosen as sources. Each source generates Con-stant Bit Rate (CBR) traffic and the destination of eachpacket is chosen randomly from the set of the node’sone-hop neighbors. We consider the aggregate through-put, the average delay and the packet delivery ratio asour performance metrics.

Fig. 9 shows the aggregate throughput as the totaloffered load increases. We can see that OPDMAC outper-

forms all other protocols due to its ability in exploitingthe offered spatial reuse. The results also show thatDMAC-OM-BO outperforms DMAC since it alleviatesdeafness chains and deadlocks. We can also see thatCDR-MAC achieves the least throughput as a result of the large control overhead associated with the protocolthat significantly offsets the benefits of spatial reuse.

Fig. 10 illustrates the average delay in the samenetwork. The figure shows that the average delay of OPDMAC is in terms of milliseconds even at veryhigh loads. In contrast, other protocols experience muchhigher average delay in terms of seconds at high loads.






11

9.pdf 2 4 6 8 10 12 14 16 18 20

0

5

10

15


A g g r e g a t e T h r o u g h p u t ( M b p s )

OPDMAC

DMAC−OM−BO

DMAC

802.11

CDR

Fig. 9. Aggregate throughput for multihop random net-work with one-hop flows.

10.pdf 2 4 6 8 10 12 14 16 18 20

0

0.5

1

1.5

2

2.5

3

3.5


A v e r a g e D e

l a y ( S e c )

OPDMAC

DMAC−OM−BO

DMAC

802.11

CDR

Fig. 10. Average delay for multihop random network withone-hop flows.

Although the delay experienced by IEEE 802.11 is nec-essary to resolve contention, the other directional MAC

protocols fail to fully exploit the benefits of beamform-ing antennas. The significant improvement achieved byOPDMAC is mainly because it is very effective in ex-ploiting transmission opportunity offered in this casewhen multiple flows at each node are ready for transmis-sion in different directions. The proposed scheme pre-vents the node from undergoing unnecessary idle waittime and minimizes the queuing delay by transmitting apacket in one direction during the backoff period needed

before transmitting another packet in another direction.On the other end, CDR-MAC suffers from very highdelay due to the time consumed in transmitting severalRTS packets before each data transmission.

In Fig. 11, we plot the Packet Delivery Ratio (PDR)versus the total offered load. We can see that CDR-MAC performs the best since it is a conservative protocolthat aims to performs collision and deafness avoidance.DMAC-OM-BO performs similar to IEEE 802.11 becauseof the prolonged omni-directional backoff periods. Atlow loads, DMAC starts to suffer from a low PDRdue to the successive failures resulting from deafnesswhile OPDMAC has a higher PDR since it minimizesthe correlation between successive retransmission at-tempts. At high loads, the PDR for OPDMAC decreasesdramatically. This is mainly due to the IEEE 802.11

11.pdf 2 4 6 8 10 12 14 16 18 20

0.86

0.88

0.9

0.92

0.94

0.96

0.98

1


P a c k e t D e l i v e r y R a t i o

OPDMAC

DMAC−OM−BO

DMAC

802.11

CDR

Fig. 11. Packet delivery ratio for multihop random networkwith one-hop flows.

default parameters and timing used with OPDMAC.As discussed in section 6, those values are not suitablefor the case of beamforming antennas. In contrary tothe conservative CDR-MAC, OPDMAC is an aggressiveprotocol that aims to exploit the spatial reusability of the

wireless channel. Hence, its implementation parametersshould be carefully chosen to deal with the possibilityof transmission failures especially at high traffic loads.

In the next experiment, we evaluate the performanceof OPDMAC by changing the default values for DIFS,CTS-timeout and the backoff CW to those discussedin Section 6. The new and the standard values areshown in Table 2. We plot the aggregate throughput,average delay and the packet delivery ratio in Figs. 12,13 and 14 respectively. As we can see, the aggregatethroughput for OPDMAC obtained with the new valuesis almost identical to that obtained using the standardones. However, the new parameters improve the packet

delivery ratio. At high offered loads, the PDR increasesfrom 86.9% to 90.1%. This is mainly due to the reductionin the number of hidden terminals when a longer DIFSis performed before transmission. As shown in Fig. 13,the gain in the PDR comes at the expense of a slightincrease in delay which is due to using a higher valuefor DIFS. However, the average delay is still far belowthat achieved by the other protocols.

In the previous experiments, we used an LP windowsimilar to the contention window of IEEE 802.11b whichis equal to [0, 31]. In this experiment, we evaluatethe performance of OPDMAC when the LP window is

changed while keeping the new values for DIFS, CTS-timeout and the backoff CW shown in Table 2. Figs. 15,16 and 17 show the aggregate throughput, average delayand the packet delivery ratio respectively. By increasing

TABLE 2New OPDMAC parameters

DIFS CTS-timeout Backoff CW

New 300 μs 20 μs [16,31]

802.11 Standard 50 μs 314 μs [0,31]






12

12.pdf

2 4 6 8 10 12 14 16 18 200

5

10

15



Standard

New

Fig. 12. Aggregate throughput with the new parameters.

13.pdf

2 4 6 8 10 12 14 16 18 200

0.05

0.1

0.15

0.2

0.25

0.3

0.35



Standard

New

Fig. 13. Average delay with the new parameters.

14.pdf

2 4 6 8 10 12 14 16 18 200.86

0.88

0.9

0.92

0.94

0.96

0.98

1


P a c k e t D e l i v e r y R a

t i o

Standard

New

Fig. 14. Packet delivery ratio with the new parameters.

the LP window, the transmission failures due to deafness

decrease since the nodes are likely to spend more time inomni-directional mode listening for the medium. Fig. 17depicts the benefits of increasing the LP window on thepacket delivery ratio. However, the throughput curvesshown in Fig. 15, shows that the largest LP window[0,127] results in a decrease in the aggregate throughput

by 3%. This is mainly due to the increase in the idlewaiting time accompanied by the large LP that coulddecrease the spatial reuse gain. With respect to the aver-age delay, Fig. 16 shows that the moderate LP window[16, 63] achieves a delay equal to the delay achieved by asmaller LP window at very high loads. This shows that

15.pdf

2 4 6 8 10 12 14 16 18 200

5

10

15



LP ∈ [0,31]

LP ∈ [16,63]

LP ∈ [0,127]

Fig. 15. Aggregate throughput with different LP windows.

16.pdf

2 4 6 8 10 12 14 16 18 200

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4



LP ∈ [0,31]

LP ∈ [16,63]

LP ∈ [0,127]

Fig. 16. Average delay with different LP windows.

17.pdf

2 4 6 8 10 12 14 16 18 200.86

0.88

0.9

0.92

0.94

0.96

0.98

1


P a c k e t D e l i v e r y R a

t i o

LP ∈ [0,31]

LP ∈ [16,63]

LP ∈ [0,127]

Fig. 17. Packet delivery ratio with different LP windows.

the LP window of [16, 63] achieves a trade-off between

the probability of deafness and the unnecessary idlewaiting time in the considered scenarios.

7.2.2 Multi-hop flows

In this subsection, we evaluate the performance of theMAC protocols in the presence of multi-hop flows. Weconsider five CBR flows with random source-destinationpairs. The flows are routed over minimum hop routesthat are statically assigned. We consider four perfor-mance metrics which are the aggregate end-to-endthroughput, the average end-to-end delay, the controloverhead and Jain’s fairness index [33].






13

18.pdf 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0

0.5

1

1.5

2

2.5

3

3.5

Per −flow Load (Mbps)

A g g r e g a t e

E n d − t o − E n d T h r o u g h p u t ( M b p s ) OPDMAC

DMAC−OM−BO

DMAC

802.11

CDR

Fig. 18. Aggregate end-to-end throughput for randommulti-hop topologies with five multi-hop flows.

19.pdf 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5


A v e r a g e E n d − t o − E

n d D e l a y ( S e c )

OPDMAC

DMAC−OM−BO

DMAC

802.11

CDR

Fig. 19. Average end-to-end delay for random multi-hoptopologies with five multi-hop flows.

Fairness Index =

l

i=1

xi2

l

li=1

x2

i

, (1)

where l is the number of flows and xi is the end-to-endthroughput of flow i.

In Fig. 18, we plot the aggregate end-to-end through-put versus the per-flow offered load. As we can see, theOPDMAC protocol significantly outperforms the otherprotocols since it fully exploits the benefits of spatialreuse introduced by the beamforming antennas. We alsonotice DMAC and DMAC-OM-BO performs better thanIEEE 802.11 since they benefit from the spatial reuse

although they suffer from deafness while CDR-MACfails to exploit the benefits of beamforming antennas dueto the large overhead used to address their challenges.

Fig. 19 shows the average end-to-end delay versus theoffered load. As expected, CDR-MAC and IEEE 802.11have the largest delay. DMAC-OM-BO experiences largedelay due to its omni-directional backoff that limits thespatial reuse. Our OPDMAC protocol has the smallestend-to-end delay due to its novel backoff mechanismthat minimizes the idle waiting time and eliminates theHoL blocking. We also notice that DMAC experiences anaverage delay that is smaller than DMAC-OM-BO. The

20.pdf 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1


F a i r n e s s I n d e x

OPDMAC

DMAC−OM−BO

DMAC

802.11

CDR

Fig. 20. The Fairness Index for random multi-hop topolo-gies with five multi-hop flows.

21.pdf 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

1

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8


C o n t r o l O v

e r h e a d

OPDMAC

DMAC−OM−BO

DMAC

802.11

CDR

Fig. 21. Control overhead for random multi-hop topolo-gies with five multi-hop flows.

reason is that some flows are completely blocked due topersistent deafness. In DMAC, if an intermediate node

on the route of a certain flow is also the originator of anew flow, the first flow is blocked as the intermediatenode remains deaf as long as its own flow has packetsto send. This results in fewer active flows in the networkexperiencing a relatively lower delay.

In Fig. 20, we plot the fairness index versus the per-flow offered load. As we can see, the OPDMAC is thefairest among the protocols we compared it with. Thisis because OPDMAC protocol reduces the impact of deafness and does not rely on the binary exponential

backoff mechanism, rather it employs a constant windowfor the listening period.

Fig. 21 shows the control overhead. The overhead isdefined as the average number of bits transmitted todeliver one bit of payload to the receiver at the MAClayer. We can see that CDR-MAC has large overhead dueto the circular transmission of RTS packets. DMAC hasslightly more overhead than the rest of protocols since itsuffers from more transmission failures due to deafness.OPDMAC has small control overhead similar to DMAC-OM-BO and IEEE 802.11. This proves that OPDMACis a lightweight protocol that is able to enhance thespatial reuse and reduce the impact of deafness withoutadditional control overhead.






14

8 DISCUSSION

8.1 Deafness Mitigation

The deafness problem is a critical challenge to the mech-anism of exploiting the spatial reusability using beam-forming antennas. It is initiated when a transmitter failsto communicate with a receiver because the receiver is

beamformed in another direction. It is aggravated when

the transmitter reacts inappropriately to such failures[3]. While the majority of existing approaches focus onresolving the occurrence of deafness at the expense of compromising the spatial reuse [4]–[7], we primarilyaddressed the more important cause which is the binaryexponential backoff mechanism commonly used by theseMAC protocols. With our active backoff algorithm, theconsequences of the deafness problem are significantlyreduced. Hence, our approach can be used along withthe approaches that rely on the use of additional controloverhead to inform neighbors about possible deafness,however, the negative impact of overhead in this case isstill questionable.

8.2 Mobility

In this work, we have not considered mobility in ourevaluations. Our results are more applicable to multi-hop wireless networks with static topologies such asmesh networks and wireless backbones. Node mobilitycould result in stale information in neighbor look-uptables causing inaccurate beamforming. In this paper,the network layer is responsible to provide the MAClayer with the beamforming information needed to com-municate with the neighbors. The information could becollected during route discovery phase or through peri-

odic HELLO packets. Alternatively, a separate neighbordiscovery mechanism similar to that proposed in [21]can be employed in conjunction with our protocol. Theimpact of mobility is closely related to the efficiencyof the neighbor discovery module. We plan to evaluatethe impact of mobility in our future work. However,we expect our proposed OPDMAC protocol to performequally well in a mobile scenario since OPDMAC tendsto minimize the idle waiting time following a transmis-sion failure so the mobility of one neighbor may nothave a significant impact on the transmissions destinedto other neighbors. However, further investigation isneeded to evaluate the OPDMAC performance in amobile network.

8.3 Compatibility with IEEE 802.11 MAC

Due to the vast spread of IEEE 802.11 wireless cards,the incremental deployment of beamforming antenna-

based wireless devices is inevitable. Unlike most ex-isting directional MAC protocols that employs specificcontrol packets, additional fields in the RTS/CTS pack-ets, antenna-dependent inter-frame spacing and/or busytones, OPDMAC uses the same RTS/CTS packet formatand the same SIFS as the IEEE 802.11 standard. This

ensures the compatibility of OPDMAC, thus facilitatingthe partial deployment of beamforming antennas.

8.4 Multi-path Environment

In this paper, we evaluated the performance of theproposed OPDMAC under a single-path propagationmodel. Although the use of beamforming antennas re-

duces the effect of multi-path fading, signals transmitted by neighboring nodes can be received from severaldirections and may interfere with ongoing directionalcommunications. On the other hand, the presence of multiple paths between a transmitter-receiver pair couldallow nodes outside their ideal communication regionto learn about the ongoing communication and hencetransmission failures due to deafness can be reduced. Inour future work, we plan to evaluate the performanceof the OPDMAC protocol in a multi-path environment.

8.5 Implication of Constant Contention Window

Following a transmission failure, if the node cannot finda packet to transmit in another direction, it is forcedto enter an idle backoff phase as mentioned in Section5.4. The OPDMAC employs random backoff derivedfrom a constant contention window to reduce the idlewaiting time since most transmission failures are dueto deafness [4]. Deadlock may occur between two ormore contending nodes when they try to send packetsto a common receiver as a result of using a constantcontention window. This may happen under strict con-ditions such that the contending senders are within thesame antenna beam of the receiver as well as theirtransmit queues have no packets destined to any other

receiver. The deadlock will most likely occur in a WLANsituation where fixed nodes are attached to a singleaccess point. In case of a multi-hop wireless network,which is the target network for OPDMAC deployment,a node usually acts as forwarder and lies on a number of flow paths. Hence, its queue contains packets for mul-tiple receivers preventing deadlock from occurring. Inaddition, there are two factors that can break an alreadyestablished deadlock: (i) A deadlocked node will get outof the deadlock situation as soon as it receives a packet(during its idle omni-directional backoff) to be deliveredto another receiver. (ii) A deadlocked mobile node islikely to move away from the receiver’s beam, in whichit is in deadlock with other senders, to another receiving

beam. Although theoretically possible, deadlock would be rare and short-lived in a multi-hop wireless networkwith dynamic traffic, retransmission count limits andnode mobility.

9 CONCLUSION

This paper addresses the problem of designing op-portunistic medium access control protocols for multi-hop wireless networks with beamforming antennas. Wediscussed various design issues and showed that the






15

binary exponential backoff algorithm, commonly usedamong directional MAC protocols, is over-conservativeand should not be used in the presence of beamformingantennas. Based on our discussion, we presented a set of observations that forms the foundation of the proposedOPDMAC protocol. OPDMAC employs a new active

backoff mechanism in which the node is not forced to un-dergo idle backoff following a transmission failure ratherit is allowed to explore transmission opportunities foroutstanding packets in other directions. We also intro-duced a listening period in which the node remains idlein an omni-directional mode after each successful trans-mission to avoid possible starvation due to prolongedperiods of deafness. Decoupling the listening phase andthe backoff phase is a unique feature of our proposedprotocol. This enables OPDMAC to enhance the spatialreusability of the wireless channel and simultaneouslyreduce the impact of the deafness problem withoutadditional overhead. Through extensive simulations, wedemonstrated that the OPDMAC protocol enhances the

performance in terms of throughput, delay, packet de-livery ratio and fairness. Moreover, we discussed andevaluated the benefits of carefully choosing some of theprotocol parameters instead of using the default valuesgenerally used by existing directional MAC protocols.

ACKNOWLEDGMENT

The authors would like to thank the anonymous review-ers for their valuable comments that have significantlyimproved the quality of the paper.

REFERENCES[1] R. Ramanathan, “On the Performance of Ad Hoc Networks with

Beamforming Antennas,” in ACM MobiHoc, October 2001, pp. 95–105.

[2] R. Choudhury, X. Yang, R. Ramanathan, and N. Vaidya, “UsingDirectional Antennas for Medium Access Control in Ad HocNetworks,” in ACM Mobicom, September 2002, pp. 59–70.

[3] R. Choudhury and N. Vaidya, “Deafness: A MAC Problem in AdHoc Networks when Using Directional Antennas,” in IEEE ICNP,October 2004, pp. 283–292.

[4] M. Takata, M. Bandai, and T. Watanabe, “A MAC Protocol withDirectional Antennas for Deafness Avoidance in Ad Hoc Net-works,” in IEEE GLOBECOM, November 2007, pp. 620–625.

[5] T. Korakis, G. Jakllari, and L. Tassiulas, “CDR-MAC: A Protocolfor Full Exploitation of Directional Antennas in Ad Hoc WirelessNetworks,” IEEE Transactions on Mobile Computing, vol. 7, no. 2,

pp. 145–155, Febraury 2008.[6] G. Jakllari, I. Broustis, T. Korakis, S. V. Krishnamurthy, , andL. Tassiulas, “ Handling Asymmetry in Gain in Directional An-tenna Equipped Ad Hoc Networks,” in IEEE PIMRC, September2005, pp. 1284–1288.

[7] H. Gossain, C. Cordeiro, , and D. P. Agrawal, “MDA: An EfficientDirectional MAC scheme for Wireless Ad Hoc Networks,” in IEEEGLOBECOM, vol. 6, November 2005, pp. 3633–3637.

[8] M. Takata and M. Bandai and T. Watanabe, “A Receiver-InitiatedDirectional MAC Protocol for Handling Deafness in Ad HocNetworks,” in IEEE ICC, vol. 9, June 2006, pp. 4089–4095.

[9] V. Kolar, S. Tilak, and N. B. Abu-Ghazaleh, “Avoiding Head of Line Blocking in Directional Antenna,” in IEEE LCN , November2004, pp. 385–392.

[10] IEEE, “IEEE 802.11 Standard: Wireless LAN Medium AccessControl (MAC) and Physical Layer (PHY) Specification,” 1999.

[11] S. S. V. Bharghavan, A. Demers and L. Zhang, “MACAW: A MediaAccess Protocol for Wireless LANs,” in ACM SIGCOMM, August1994, pp. 212–225.

[12] Y. Ko, V. Shankarkumar, and N. Vaidya, “Medium Access ControlProtocols Using Directional Antennas in Ad Hoc Networks,” inIEEE INFOCOM, March 2000, pp. 13–21.

[13] A. Nasipuri, S.Ye, J. You, and R. E. Hiromoto, “A MAC Protocolfor Mobile Ad Hoc Networks Using Directional Antennas,” inIEEE WCNC, September 2000, pp. 1214–1219.

[14] T. ElBatt, T. Anderson, and B. Ryu, “Performance Evaluation of

Multiple Access Protocols for Ad Hoc Networks Using DirectionalAntennas,” in IEEE WCNC, vol. 2, March 2003, pp. 982–987.

[15] Y. Wang and J. J. Garcia-Luna-Aceves, “Spatial Reuse and Colli-sion Avoidance in AdHoc Networks with Directional Antennas,”in IEEE GLOBECOM, vol. 1, November 2002, pp. 112–116.

[16] S. Bandyopadhyay, K. Hasuike, S. Horisawa, and S. Tawara,“An Adaptive MAC and Directional Routing Protocol for AdHoc Wireless Network using ESPAR Antenna,” in ACM MobiHoc,October 2001, pp. 243–246.

[17] M. Takai, J. Martin, A. Ren, and R. Bagrodia, “Directional VirtualCarrier Sensing for Directional Antennas in Mobile Ad HocNetworks,” in ACM MobiHoc, June 2002, pp. 183–193.

[18] R. R. Choudhury, X. Yang, R. Ramanathan, and N. H. Vaidya,“On Designing MAC Protocols for Wireless Networks UsingDirectional Antennas,” IEEE Transactions on Mobile Computing,vol. 5, no. 5, pp. 477–491, May 2006.

[19] V. A. RamMohan, H. Sethu, M. R. Hosaagrahara, and K. R.

Dandekar, “A New Protocol to Mitigate the Unheard RTS/CTSProblem in Networks with Switched Beam Antennas,” in IEEEISWPC, February 2007, pp. 129–134.

[20] A. P. Subramanian and S. R. Das, “Addressing Deafness andHidden Terminal Problem in Directional Antenna based WirelessMulti-Hop Networks,” in IEEE COMSWARE, Bangalore, India,

January 2007, pp. 1–6.[21] R. Ramanathan, J. Redi, C. Santivanez, D. Wiggins, and S. Polit,

“Ad Hoc Networking with Directional Antennas: A CompleteSystem Solution,” IEEE Journal on Selected Areas in Communications,vol. 23, no. 3, pp. 496–506, March 2005.

[22] O. Bazan and M. Jaseemuddin, “An Opportunistic DirectionalMAC Protocol for Multihop Wireless Networks with SwitchedBeam Directional Antennas,” in IEEE ICC, May 2008, pp. 2775–2779.

[23] G. Jakllari, W. Luo, and S. V. Krishnamurthy, “An IntegratedNeighbor Discovery and MAC Protocol for Ad Hoc Networks

Using Directional Antennas,” IEEE Transactions On Wireless Com-munications, vol. 6, no. 3, pp. 11–21, March 2007.

[24] E. Shihab, L. Cai, and J. Pan, “A Distributed Directional-to-Directional MAC Protocol for Asynchronous Ad Hoc Networks,”in IEEE GLOBECOM, November 2008, pp. 1–5.

[25] Z. Huang, C. chung Shen, C. Srisathapornphat, and C. Jaikaeo,“A Busy-Tone Based Directional MAC Protocol for Ad Hoc Net-works,” in IEEE Milcom, vol. 2, October 2002, pp. 1233–1238.

[26] S. S. Kulkarni and C. Rosenberg, “DBSMA: A MAC Protocol forMulti-hop Ad-hoc Networks with Directional Antennas,” in IEEEPIMRC, vol. 12, September 2005, pp. 1371–1377.

[27] E. Ulukan and O. Gurbuz, “Angular MAC Protocol with LocationBased Scheduling for Wireless Ad Hoc Networks,” in IEEE VTC,vol. 3, May 2005, pp. 1473–1478.

[28] R. Choudhury and N. Vaidya, “MAC-Layer Capture: A Problemin Wireless Mesh Networks using Beamforming Antennas,” inIEEE SECON , June 2007, pp. 401–410.

[29] L. Bao and J. Garcia-Luna-Aceves, “Transmission Scheduling inAd Hoc Networks with Directional Antennas,” in ACM Mobicom,September 2002, pp. 48–58.

[30] J. Wang, Y. Fang, and D. Wu, “SYN-DMAC: A Directional MACProtocol for Ad Hoc Networks with Synchronization,” in IEEE

Milcom, vol. 4, October 2005, pp. 2258–2263.[31] G. Bianchi, “Performance Analysis of the IEEE 802.11 Distributed

Coordination Function,” IEEE Journal on Selected Areas in Commu-nications, vol. 18, no. 3, pp. 535–547, March 2000.

[32] OPNET Technologies, http://www.opnet.com/.[33] R. Jain, D. Chiu, and W. Hawe, A Quantitative Measure of Fair-

ness and Discrimination for Resource Allocation in Shared ComputerSystem, 1984, DEC Technical Report 301.






16

Osama Bazan (S’06, M’10) received his Bache-lor (BSc.) and Master (MSc.) Degrees in Elec-tronics and Electrical Communications Engi-neering from Cairo University, Egypt in 2000 and2003 respectively and his PhD degree in Elec-trical and Computer Engineering from RyersonUniversity, Canada in 2009. Osama is currentlya Post Doctoral Fellow in the Department ofElectrical and Computer Engineering at RyersonUniversity. He served as technical program com-

mittee member for several international confer-ences and as a peer reviewer for numerous journals and conferences.His current research interests include medium access control in ad hocnetworks, multi-hop wireless networks with heterogeneous antennasand interference-aware routing in wireless mesh networks.

Muhammad Jaseemuddin (M ’98) receivedB.E. from N.E.D. University of Engg. & Tech.,Karachi, Pakistan, in 1989, M. S. from The Uni-versity of Texas at Arlington in 1991, and Ph.D.from University of Toronto in 1997. He workedin Advanced IP group and Wireless TechnologyLab (WTL) at Nortel Networks. He worked on theprototypes of Wireless Service Delivery Platformand UMTS VHE framework. In WTL, he workedon QoS, Routing and Handover issues in mobile

wireless IP access network. He has been Asso-ciate Professor at Ryerson University since 2002. His research interestsinclude investigating MAC and Routing for Smart Antennas and Co-operative Communication; the impact of mobility on Routing and Trans-port layers; Mobile Application development platforms using overlay ad-hoc network, Heterogeneous wireless networks, and IP Routing andtraffic engineering. He has been working in collaboration with SolanaNetworks on designing Routing Failure Detection and Recovery tool.URL: http://www.ee.ryerson.ca/ jaseem/



On The Design of Opportunistic MAC Protocols - Copy

Documents

Transcript of On The Design of Opportunistic MAC Protocols - Copy