medium access control protocols for ad hoc wireless networks

8/2/2019 Medium Access Control Protocols for Ad Hoc Wireless Networks

1/33

Medium Access Control protocols for ad hocwireless networks: A survey

Sunil Kumar a,*, Vineet S. Raghavan b, Jing Deng c

a Department of Electrical and Computer Engineering, Clarkson University, Potsdam, NY 13699, United Statesb Digital Television Group, ATI Technologies Inc., Marlborough, MA 01752, United States

c Department of Computer Science, University of New Orleans, New Orleans, LA 70148, United States

Received 17 October 2003; received in revised form 13 September 2004; accepted 8 October 2004Available online 2 November 2004

Abstract

Studies of ad hoc wireless networks are a relatively new field gaining more popularity for various new applications.In these networks, the Medium Access Control (MAC) protocols are responsible for coordinating the access from activenodes. These protocols are of significant importance since the wireless communication channel is inherently prone toerrors and unique problems such as the hidden-terminal problem, the exposed-terminal problem, and signal fading

effects. Although a lot of research has been conducted on MAC protocols, the various issues involved have mostly beenpresented in isolation of each other. We therefore make an attempt to present a comprehensive survey of majorschemes, integrating various related issues and challenges with a view to providing a big-picture outlook to this vastarea. We present a classification of MAC protocols and their brief description, based on their operating principlesand underlying features. In conclusion, we present a brief summary of key ideas and a general direction for future work. 2004 Elsevier B.V. All rights reserved.

Keywords: Ad hoc networks; Wireless networks; MAC; Medium Access Control; Quality of Service (QoS); MANET

1. Introduction

Back in the 1970s, the Defense Advanced Re-search Projects Agency (DARPA) was involvedin the development of packet radio networks foruse in the battlefields. Around the same time, the

ALOHA [1] project used wireless data broadcast-ing to create single hop radio networks. This sub-

sequently led to development of the multi-hopmultiple-access Packet Radio Network (PRNET),which allowed communication coverage over awide area. The term multi-hop refers to the factthat data from the source needs to travel throughseveral other intermediate nodes before it reachesthe destination. One of the most attractive featuresof PRNET was rapid deployment. Also, after

1570-8705/$ - see front matter 2004 Elsevier B.V. All rights reserved.doi:10.1016/j.adhoc.2004.10.001

* Corresponding author. Tel.: +1 315 268 6602; fax: +1 315268 7600.

E-mail address: [email protected] (S. Kumar).

Ad Hoc Networks 4 (2006) 326358

www.elsevier.com/locate/adhoc
mailto:[email protected]:[email protected]:[email protected]


2/33

installation, the whole system was self-initializingand self-organizing. The network consisted of mo-bile radio repeaters, wireless terminals and dedi-

cated mobile stations. Packets were relayed fromone repeater to the other until data reached itsdestination.

With the development of technology, deviceshave shrunk in size and they now incorporatemore advanced functions. This allows a node toact as a wireless terminal as well as a repeaterand still be compact enough to be mobile. A self-organizing and adaptive collection of such devicesconnected with wireless links is now referred to asan Ad Hoc Network. An ad hoc network does notneed any centralized control. The network shoulddetect any new nodes automatically and inductthem seamlessly. Conversely, if any node movesout of the network, the remaining nodes shouldautomatically reconfigure themselves to adjust tothe new scenario. If nodes are mobile, the networkis termed as a MANET (Mobile Ad hoc NET-work). The Internet Engineering Task Force(IETF) has set up a working group named MAN-ET for encouraging research in this area [2].

Typically, there are two types of architectures inad hoc networks: flat and hierarchical [3,6]. Each

node in an ad hoc network is equipped with atransceiver, an antenna and a power source. Thecharacteristics of these nodes can vary widely interms of size, processing ability, transmissionrange and battery power. Some nodes lend them-selves for use as servers, others as clients and yetothers may be flexible enough to act as both,depending on the situation. In certain cases, eachnode may need to act as a router in order to con-vey information from one node to another [4,5].

1.1. Applications

Coupled with global roaming capabilities andseamless integration with existing infrastructure,if any, ad hoc wireless networks can be used inmany new applications [6,8]. In case of naturalor other disasters, it is possible that existing com-munication infrastructure is rendered unusable.In such situations, an ad hoc wireless network fea-turing wideband capabilities can be set up almostimmediately to provide emergency communication

in the affected region. In mobile computing envi-ronments, mobile wireless devices that have thecapability to detect the presence of existing net-

works can be used to synchronize data with theusers conventional desktop computers automati-cally, and download appointment/schedule data.A user carrying a handheld Personal Digital Assis-tant (PDA) device can download Context sensitivedata in a shopping mall or museum featuring suchwireless networks and services. The PDA would beable to detect the presence of the network and con-nect itself in an ad hoc fashion. Depending on theusers movement, the PDA can poll the networkfor relevant information based on its current loca-tion. For instance, if the user is moving throughthe clothing section of the shopping mall, informa-tion on special deals or pricing can be made avail-able. Similarly, ad hoc networks can be used intravel-related and customized household applica-tions, telemedicine, virtual navigation, etc.

1.2. Important issues

There are several important issues in ad hocwireless networks [3,68,70]. Most ad hoc wirelessnetwork applications use the Industrial, Scientific

and Medical (ISM) band that is free from licensingformalities. Since wireless is a tightly controlledmedium, it has limited channel bandwidth that istypically much less than that of wired networks.Besides, the wireless medium is inherently errorprone. Even though a radio may have sufficientchannel bandwidth, factors such as multiple ac-cess, signal fading, and noise and interferencecan cause the effective throughput in wireless net-works to be significantly lower. Since wirelessnodes may be mobile, the network topology can

change frequently without any predictable pattern.Usually the links between nodes would be bi-direc-tional, but there may be cases when differences intransmission power give rise to unidirectional links,which necessitate special treatment by the MediumAccess Control (MAC) protocols. Ad hoc networknodes must conserve energy as they mostly rely onbatteries as their power source. The security issuesshould be considered in the overall network design,as it is relatively easy to eavesdrop on wirelesstransmission. Routingprotocols require information

S. Kumar et al. / Ad Hoc Networks 4 (2006) 326358 327


3/33

about the current topology, so that a route from asource to a destination may be found. However,the existing routing schemes, such as distance-vec-

tor and link-state based protocols, lead to poorroute convergence and low throughput for dy-namic topology. Therefore, a new set of routingschemes is needed in the ad hoc wireless context[5,8].

MAC layer, sometimes also referred to as a sub-layer of the Data Link layer, involves the func-tions and procedures necessary to transfer databetween two or more nodes of the network. It isthe responsibility of the MAC layer to performerror correction for anomalies occurring in thephysical layer. The layer performs specific activi-ties for framing, physical addressing, and flowand error controls. It is responsible for resolvingconflicts among different nodes for channel access.Since the MAC layer has a direct bearing on howreliably and efficiently data can be transmittedbetween two nodes along the routing path in thenetwork, it affects the Quality of Service (QoS) ofthe network. The design of a MAC protocol shouldalso address issues caused by mobility of nodes andan unreliable time varying channel [68].

1.3. Need for special MAC protocols

The popular Carrier Sense Multiple Access(CSMA) [9] MAC scheme and its variations suchas CSMA with Collision Detection (CSMA/CD)developed for wired networks, cannot be used di-rectly in the wireless networks, as explained below.

In CSMA-based schemes, the transmitting nodefirst senses the medium to check whether it is idleor busy. The node defers its own transmission toprevent a collision with the existing signal, if the

medium is busy. Otherwise, the node begins totransmit its data while continuing to sense themedium. However, collisions occur at receivingnodes. Since, signal strength in the wireless med-ium fades in proportion to the square of distancefrom the transmitter, the presence of a signal atthe receiver node may not be clearly detected atother sending terminals, if they are out of range.As illustrated in Fig. 1, node B is within the rangeof nodes A and C, but A and C are not in eachothers range. Let us consider the case where A is

transmitting to B. Node C, being out ofA s range,cannot detect carrier and may therefore send datato B, thus causing a collision at B. This is referredto as the hidden-terminal problem, as nodes A andC are hidden from each other [10,11].

Let us now consider another case where B istransmitting to A. Since C is within Bs range, itsenses carrier and decides to defer its own trans-mission. However, this is unnecessary becausethere is no way Cs transmission can cause any col-lision at receiver A. This is referred to as theexposed-terminal problem, since B being exposed

to C caused the latter to needlessly defer its trans-mission [11]. MAC schemes are designed to over-come these problems.

The rest of the paper is organized as follows. Aclassification of ad hoc network MAC schemes isgiven in Section 2. Details of various MACschemes in each class are discussed in Sections 3and 4. The summary and future research directionsare described in Section 5, followed by conclusionin Section 6.

2. Classification

Various MAC schemes developed for wirelessad hoc networks can be classified as shown inFig. 2. In contention-free schemes (e.g., TDMA,FDMA, CDMA), certain assignments are usedto avoid contentions [6]. Contention basedschemes, on the other hand, are aware of the riskof collisions of transmitted data. Since conten-tion-free MAC schemes are more applicable to

Fig. 1. Illustration of the hidden and exposed terminalproblems.

328 S. Kumar et al. / Ad Hoc Networks 4 (2006) 326358


4/33

static networks and/or networks with centralizedcontrol, we shall focus on contention-based MACschemes in this survey.

We can view this category as a collection ofrandom access and dynamic reservation/collisionresolution protocols as shown in Fig. 2(a) [12]. Inrandom access based schemes, such as ALOHA, a

node may access the channel as soon as it isready. Naturally, more than one node may trans-mit at the same time, causing collisions. ALOHAis more suitable under low system loads withlarge number of potential senders and it offers rel-atively low throughput. A variation of ALOHA,termed Slotted ALOHA, introduces synchronized

(a)

(b)

Fig. 2. Classification of MAC schemes.



5/33

transmission time-slots similar to TDMA. In thiscase, nodes can transmit only at the beginning ofa time-slot. The introduction of time slot doubles

the throughput as compared to the pure ALOHAscheme, with the cost of necessary time synchroni-zation. The CSMA-based schemes further reducethe possibility of packet collisions and improvethe throughput.

In order to solve the hidden and exposed termi-nal problems in CSMA, researchers have come upwith many protocols, which are contention basedbut involve some forms of dynamic reservation/collision resolution. Some schemes use the Re-quest-To-Send/Clear-To-Send (RTS/CTS) controlpackets to prevent collisions, e.g. Multiple AccessCollision Avoidance (MACA) [13] and MACAfor Wireless LANs (MACAW) [14]. Yet othersuse a combination of carrier sensing and controlpackets [15,16,23], etc.

As shown in Fig. 2(b), the contention-basedMAC schemes can also be classified as sender-initiated vs. receiver-initiated, single-channel vs.multiple-channel, power-aware, directional anten-na based, unidirectional link based and QoS awareschemes. We briefly discuss these categories in thefollowing:

One distinguishing factor for MAC protocols iswhether they rely on the sender initiating the datatransfer, or the receiver requesting the same [6]. Asmentioned above, the dynamic reservation ap-proach involves the setting up of some sort of areservation prior to data transmission. If a nodethat wants to send data takes the initiative of set-ting up this reservation, the protocol is consideredto be a sender-initiated protocol. Most schemesare sender-initiated. In a receiver-initiated protocol,the receiving node polls a potential transmitting

node for data. If the sending node indeed hassome data for the receiver, it is allowed to trans-mit after being polled. The MACABy Invitation(MACA-BI) [17] and Receiver Initiated BusyTone Multiple Access (RI-BTMA) [18] are exam-ples of such schemes. As we shall see later,MACA-BI is slightly more efficient in terms oftransmit and receive turn around times comparedto MACA.

Another classification is based on the number ofchannels used for data transmission. Single chan-

nel protocols set up reservations for transmissions,and subsequently transmit their data using thesame channel or frequency. Many MAC schemes

use a single channel [1,9,1315, etc.]. Multiplechannel protocols use more than one channel inorder to coordinate connection sessions amongthe transmitter and receiver nodes. The FCC man-dates that all radios using the ISM band must em-ploy either DSSS or FHSS schemes. Several MACprotocols have been developed for using multiplechannels through frequency-hopping techniques,e.g., Hop-Reservation Multiple Access (HRMA)scheme [19]. Some others use a special control-signal on a separate channel for protecting the ac-tual data that is transmitted on the data channel(s)[20,4753].

As mentioned earlier, it becomes important inthe context of low power devices, to have energyefficient protocols at all layers of the networkmodel. Much work has already been done forstudying and developing appropriate MAC proto-cols that are also power aware ([2736], etc).

Yet another class of MAC protocols uses direc-tional antennas [5664]. The advantage of thismethod is that the signals are transmitted only inone direction. The nodes in other directions are

therefore no longer prone to interference or colli-sion effects, and spatial reuse is facilitated.

Usually the links between nodes are bi-direc-tional, but there may be cases when differences intransmission power give rise to unidirectionallinks, which necessitate special treatment by theMAC protocols. Prakash [66] pointed out someof the issues to be taken care of in unidirectionallink networks. Several MAC schemes have beenproposed for unidirectional links [10,6769].

With the growing popularity of ad hoc net-

works, it is reasonable to expect that users willdemand some level of QoS from it, such as end-to-end delay, available bandwidth, probability ofpacket loss, etc. However, the lack of centralizedcontrol, limited bandwidth channels, node mobil-ity, power or computational constraints and theerror-prone nature of the wireless medium makeit very difficult to provide effective QoS in ad hocnetworks [3,7274]. Since the MAC layer has a di-rect bearing on how reliably and efficiently datacan be transmitted from one node to the next



6/33

along the routing path in the network, it affects theQuality of Service (QoS) of the network. SeveralQoS-aware MAC schemes have been reported in

the literature [8699].Note that the above categories are not totallyindependent of each other. In fact, a given MACprotocol may belong to more than one category.For example, Power Aware Medium AccessControl with Signaling (PAMAS) [27] is apower-aware protocol that also uses two channels.Similarly; RI-BTMA is a receiver-initiated MACscheme that uses multiple channels.

Several representative MAC schemes for ad hocwireless networks are briefly discussed and sum-marized in the following two sections. For the sakeof convenience in discussion, we have broadly ar-ranged the schemes in non-QoS and QoS-awareclasses. The non-QoS MAC schemes in Section 3have been further divided in the following catego-ries: general, power-aware, multiple channel,directional antenna-based, and unidirectionalMAC protocols. Similarly, QoS-aware schemes(in Section 4) have been arranged in a few catego-ries according to their properties. In the process ofchoosing these MAC schemes, we tended to selectthose that are more representative in their

category.

3. Review of non-QoS MAC protocols

In particular, we shall discuss several importantcontention based MAC schemes in the single chan-nel, receiver initiated, power-aware, and multiplechannel categories. Due to space limitation, wewill only briefly discuss other categories. However,it should not mean that these other categories are

less important.

3.1. General MAC protocols

We have mostly included the single channelprotocols in this sub-section. A receiver initiatedMACA-BI scheme is also discussed.

3.1.1. Multiple access collision avoidance (MACA)

The MACA protocol was proposed by Karn toovercome the hidden and exposed terminal prob-

lems in CSMA family of protocols [13]. MACAuses two short signaling packets, similar to theAppleTalk protocol [21]. In Fig. 1, if node A

wishes to transmit to node B, it first sends anRTS packet to B, indicating the length of the datatransmission that would later follow. If B receivesthis RTS packet, it returns a CTS packet to A thatalso contains the expected length of the data to betransmitted. When A receives the CTS, it immedi-ately commences transmission of the actual data toB. The key idea of the MACA scheme is that anyneighboring node that overhears an RTS packethas to defer its own transmissions until some timeafter the associated CTS packet would have fin-ished, and that any node overhearing a CTS pack-et would defer for the length of the expected datatransmission.

In a hidden terminal scenario (see Fig. 1) as ex-plained in Section 1, C will not hear the RTS sentby A, but it would hear the CTS sent by B.Accordingly, C will defer its transmission duringA s data transmission. Similarly, in the exposedterminal situation, C would hear the RTS sentby B, but not the CTS sent by A. Therefore C willconsider itself free to transmit during Bs transmis-sion. It is apparent that this RTSCTS exchange

enables nearby nodes to reduce the collisions atthe receiver, not the sender. Collisions can still oc-cur between different RTS packets, though. If twoRTS packets collide for any reason, each sendingnode waits for a randomly chosen interval beforetrying again. This process continues until one ofthe RTS transmissions elicits the desired CTS fromthe receiver.

MACA is effective because RTS and CTS pack-ets are significantly shorter than the actual datapackets, and therefore collisions among them are

less expensive compared to collisions among thelonger data packets. However, the RTSCTS ap-proach does not always solve the hidden terminalproblem completely, and collisions can occur whendifferent nodes send the RTS and the CTS packets.Let us consider an example with four nodes A, B,Cand D in Fig. 3. Node A sends an RTS packet toB, and B sends a CTS packet back to A. At C,however, this CTS packet collides with an RTSpacket sent by D. Therefore C has no knowledgeof the subsequent data transmission from A to B.



7/33

While the data packet is being transmitted, Dsends out another RTS because it did not receivea CTS packet in its first attempt. This time, C re-plies to D with a CTS packet that collides withthe data packet at B. In fact, when hidden termi-nals are present and the network traffic is high,the performance of MACA degenerates to thatof ALOHA [20].

Another weakness of MACA is that it does notprovide any acknowledgment of data transmissionsat the data link layer. If a transmission fails for

any reason, retransmission has to be initiated bythe transport layer. This can cause significant de-lays in the transmission of data.

In order to overcome some of the weaknesses ofMACA, Bharghavan et al. [14] proposed MACAfor Wireless (MACAW) scheme that uses a fivestep RTSCTSDSDATAACK exchange. MA-CAW allows much faster error recovery at thedata link layer by using the acknowledgment pack-et (ACK) that is returned from the receiving nodeto the sending node as soon as data reception is

completed. The backoff and fairness issues amongactive nodes were also investigated. MACAWachieves significantly higher throughput comparedto MACA. It however does not fully solve the hid-den and exposed terminal problems [15,20].

The Floor Acquisition Multiple Access (FAMA)is another MACA based scheme that requiresevery transmitting station to acquire control ofthe floor (i.e., the wireless channel) before it actu-ally sends any data packet [15]. Unlike MACA orMACAW, FAMA requires that collision avoid-

ance should be performed both at the sender aswell as the receiver. In order to acquire thefloor, the sending node sends out an RTS using

either non-persistent packet sensing (NPS) ornon-persistent carrier sensing (NCS). The receiverresponds with a CTS packet, which contains theaddress of the sending node. Any station overhear-ing this CTS packet knows about the station thathas acquired the floor. The CTS packets are re-peated long enough for the benefit of any hiddensender that did not register another sending nodesRTS. The authors recommend the NCS variant forad hoc networks since it addresses the hidden ter-minal problem effectively.

3.1.2. IEEE 802.11 MAC scheme

The IEEE 802.11 specifies two modes ofMAC protocol: distributed coordination function(DCF) mode (for ad hoc networks) and pointcoordination function (PCF) mode (for centrallycoordinated infrastructure-based networks) [2225]. The DCF in IEEE 802.11 is based on CSMAwith Collision Avoidance (CSMA/CA), which canbe seen as a combination of the CSMA andMACA schemes. The protocol uses the RTS

CTSDATAACK sequence for data transmis-sion. Not only does the protocol use physicalcarrier sensing, it also introduces the novel conceptof virtual carrier sensing. This is implemented inthe form of a Network Allocation Vector (NAV),which is maintained by every node. The NAV con-tains a time value that represents the duration upto which the wireless medium is expected to bebusy because of transmissions by other nodes.Since every packet contains the duration informa-tion for the remainder of the message, every node

overhearing a packet continuously updates its ownNAV.Time slots are divided into multiple frames and

there are several types of inter frame spacing (IFS)slots. In increasing order of length, they are theShort IFS (SIFS), Point Coordination FunctionIFS (PIFS), DCF IFS (DIFS) and Extended IFS(EIFS). The node waits for the medium to be freefor a combination of these different times before itactually transmits. Different types of packets canrequire the medium to be free for a different num-

Fig. 3. Illustration of failure of RTSCTS mechanism insolving Hidden and Exposed terminal problems.



8/33

ber or type of IFS. For instance, in ad hoc mode, ifthe medium is free after a node has waited forDIFS, it can transmit a queued packet. Otherwise,

if the medium is still busy, a backoff timer is initi-ated. The initial backoff value of the timer is cho-sen randomly from between 0 and CW-1 whereCW is the width of the contention window, interms of time-slots. After an unsuccessful trans-mission attempt, another backoff is performedwith a doubled size of CW as decided by binaryexponential backoff (BEB) algorithm. Each timethe medium is idle after DIFS, the timer is decre-mented. When the timer expires, the packet istransmitted. After each successful transmission,another random backoff (known as post-backoff)is performed by the transmission-completing node.A control packet such as RTS, CTS or ACK istransmitted after the medium has been free forSIFS. Fig. 4 shows the channel access in IEEE802.11.

IEEE 802.11 DCF is a widely used protocol forwireless LANs. Many of the MAC schemes dis-cussed in this paper are based on it. Some otherfeatures of this protocol will be discussed alongwith such schemes.

3.1.3. Multiple access collision avoidance-byinvitation (MACA-BI)

In typical sender-initiated protocols, the send-ing node needs to switch to receive mode (to getCTS) immediately after transmitting the RTS.Each such exchange of control packets adds toturnaround time, reducing the overall throughput.MACA-BI [17] is a receiver-initiated protocol and

it reduces the number of such control packet ex-changes. Instead of a sender waiting to gain accessto the channel, MACA-BI requires a receiver to re-

quest the sender to send the data, by using aReady-To-Receive (RTR) packet instead of theRTS and the CTS packets. Therefore, it is a two-way exchange (RTRDATA) as against thethree-way exchange (RTSCTSDATA) ofMACA [13].

Since the transmitter cannot send any data be-fore being asked by the receiver, there has to bea traffic prediction algorithm built into the receiverso it can know when to request data from the sen-der. The efficiency of this algorithm determines thecommunication throughput of the system. Thealgorithm proposed by the authors piggybacksthe information regarding packet queue lengthand data arrival rate at the sender in the datapacket. When the receiver receives this data, it isable to predict the backlog in the transmitter andsend further RTR packets accordingly. There is aprovision for a transmitter to send an RTS packetif its input buffer overflows. In such a case, the sys-tem reverts to MACA.

The MACA-BI scheme works efficiently in net-works with predictable traffic pattern. However, if

the traffic is bursty, the performance degrades tothat of MACA.

3.1.4. Group allocation multiple access with packet

sensing (GAMA-PS)

GAMA-PS incorporates features of contentionbased as well as contention free methods [26]. It di-vides the wireless channel into a series of cycles.

Immediate access whenmedium is idle >= DIFS

Busy Medium

Contention Window

Slot TimeDefer Access

Select Slot and decrement backoff as long

as medium stays idle

DIFS

DIFS

PIFS

SIFSBackoff Window Next Frame

Fig. 4. IEEE 802.11 DCF channel access.



9/33

Every cycle is divided in two parts for contentionand group transmission. Although the grouptransmission period is further divided into individ-

ual transmission periods, GAMA-PS does not re-quire clock or time synchronization amongdifferent member nodes. Nodes wishing to makea reservation for access to the channel employthe RTSCTS exchange. However, a node willbackoff only if it understands an entire packet.Carrier sensing alone is not sufficient reason forbacking off.

GAMA-PS organizes nodes into transmissiongroups, which consist of nodes that have been allo-cated a transmission period. Every node in thegroup is expected to listen in on the channel.Therefore, there is no need of any centralized con-trol. Every node in the group is aware of all thesuccessful RTSCTS exchanges and by extension,of any idle transmission periods.

Members of the transmission group take turnstransmitting data, and every node is expected tosend a Begin Transmission Period (BTP) packetbefore actual data. The BTP contains the stateof the transmission group, position of the nodewithin that group and the number of groupmembers. A member station can transmit up to

a fixed length of data, thereby increasing effi-ciency. The last member of the transmissiongroup broadcasts a Transmit Request (TR) pack-et after it sends its data. Use of the TR shortensthe maximum length of the contention period byforcing any station that might contend for groupmembership to do so at the start of the conten-tion period.

GAMA-PS assumes that there are no hiddenterminals. As a result, this scheme may notwork well for mobile ad hoc networks. When

there is not enough traffic in the network,GAMA-PS behaves almost like CSMA. How-ever, as the load grows, it starts to mimicTDMA and allows every node to transmit oncein every cycle.

3.2. Power aware MAC protocols

Since mobile devices are battery powered, it iscrucial to conserve energy and utilize power as effi-ciently as possible. In fact, the issue of power con-

servation should be considered across all the layersof the protocol stack. The following principlesmay serve as general guidelines for power conser-

vation in MAC protocols [2730]. First, collisionsare a major cause of expensive retransmissionsand should be avoided as far as possible. Second,the transceivers should be kept in standby mode(or switched off) whenever possible as they con-sume the most energy in active mode. Third, in-stead of using the maximum power, thetransmitter should switch to a lower power modethat is sufficient for the destination node to receivethe transmission. Several researchers, includingGoldsmith and Wicker [31], have conducted stud-ies in this area.

As we mentioned in the context of classifyingMAC protocols, some approaches implementpower management by alternating sleep and wakecycles [27,3234]. Other approaches, classified aspower control, use a variation in the transmissionpower [35,36]. We now present the details of someselected schemes in both categories.

3.2.1. Power aware medium access control with

signaling (PAMAS)

The basic idea of PAMAS developed by Ragha-

vendra and Singh [27] is that all the RTSCTS ex-changes are performed over the signaling channeland the data transmissions are kept separate overa data channel. While receiving a data packet,the destination node starts sending out a busy toneover the signaling channel. Nodes listen in on thesignaling channel to deduce when it is optimalfor them to power down their transceivers. Everynode makes its own decision whether to poweroff or not such that there is no drop in the through-put. A node powers itself off if it has nothing to

transmit and it realizes that its neighbor is trans-mitting. A node also powers off if at least oneneighbor is transmitting and another is receivingat the same time. The authors have developed sev-eral rules to determine the length of a power-downstate.

The authors also mention briefly some strate-gies, to use this scheme with other protocols likeFAMA [15]. They have also noted that the useof ACK and transmission of multiple packetstogether will also enhance the performance of



10/33

PAMAS. However, the radio transceiver turn-around time, which might not be negligible, wasnot considered in the PAMAS scheme.

3.2.2. Dynamic power saving mechanism (DPSM)

Jung and Vaidya [32] proposed DPSM based onthe idea of using sleep and wake states for nodes inorder to conserve power. It is a variation of theIEEE 802.11 scheme, in that it uses dynamicallysized Ad-hoc Traffic Indication Message (ATIM)windows to achieve longer dozing times for nodes.

The IEEE 802.11 DCF mode has a power sav-ing mechanism, in which time is divided into bea-con intervals that are used to synchronize thenodes [23]. At the beginning of each beacon inter-val, every node must stay awake for a fixed timecalled ATIM window. This window is used to an-nounce the status of packets ready for transmis-sion to any receiver nodes. Such announcementsare made through ATIM frames, and they areacknowledged with ATIM-ACK packets duringthe same beacon interval. Fig. 5 illustrates themechanism. Earlier work [33] shows that if the sizeof the ATIM window is kept fixed, performancesuffers in terms of throughput and energyconsumption.

In DPSM, each node dynamically and indepen-dently chooses the length of the ATIM window.As a result, every node can potentially end up hav-ing a different sized window. It allows the senderand receiver nodes to go into sleep state immedi-ately after they have participated in the transmis-sion of packets announced in the prior ATIMframe. Unlike the DCF mechanism, they do not

even have to stay awake for the entire beaconinterval. The length of the ATIM window is in-creased if some packets queued in the outgoing

buffer are still unsent after the current window ex-pires. Also, each data packet carries the currentlength of the ATIM window and any nodes thatoverhear such information may decide to modifytheir own window lengths based on the receivedinformation.

DPSM is found to be more effective than IEEE802.11 DCF in terms of power saving andthroughput. However, IEEE 802.11 and DPSMare not suitable for multi-hop ad hoc networksas they assume that the clocks of the nodes aresynchronized and the network is connected. Tsenget al. [34] have proposed three variations of DPSMfor multi-hop MANETs that use asynchronousclocks.

3.2.3. Power control medium access control (PCM)

Previous approaches of power control usedalternating sleep and wake states for nodes[27,32,34]. In PCM [35], the RTS and CTS packetsare sent using the maximum available power,whereas the data and ACK packets are sent withthe minimum power required to communicate

between the sender and receiver.The method for determining these lower power

levels, described below, has also been used by ear-lier researchers in [13,43]. An example scenario isdepicted in Fig. 6. Node D sends the RTS to nodeEat a transmit power level Pmax, and also includesthis value in the packet. Emeasures the actual sig-nal strength, say Pr, of the received RTS packet.

A

B

C

ATIM DATA

ATIM window

ATIM windowATIM-ACK ACK

ATIM window

Dozing

ATIM window

Next beacon intervalBeacon interval

Fig. 5. Power saving mechanism for DCF: Node A announces a buffered packet for B using an ATIM frame. Node B replies bysending an ATIM-ACK, and both A and B stay awake during the entire beacon interval. The actual data transmission from A to B iscompleted during the beacon interval. Since C does not have any packet to send or receive, it dozes after the ATIM window [32].



11/33

Based on Pmax, Pr and the noise level at its loca-tion, E then computes the minimum necessarypower level (say, Psuff) that would actually be suf-ficient for use by D. Now, when E responds withthe CTS packet using the maximum power it hasavailable, it includes the value of Psuffthat D sub-sequently uses for data transmission. G is able to

hear this CTS packet and defers its own transmis-sions. E also includes the power level that it usedfor the transmission in the CTS packet. D thenfollows a similar process and calculates the mini-mum required power level that would get a pack-et from E to itself. It includes this value in thedata packet so that E can use it for sending theACK.

PCM also stipulates that the source node peri-odically transmits the DATA packet at the maxi-mum power level, for just enough time so that

nodes in the carrier sensing range, such as A maysense it. PCM thus achieves energy savings with-out causing throughput degradation.

The operation of the PCM scheme requires arather accurate estimation of received packet sig-nal strength. Therefore, the dynamics of wirelesssignal propagation due to fading and shadowingeffect may degrade its performance. Anotherdrawback of this scheme is the difficulty in imple-menting frequent changes in the transmit powerlevels.

3.2.4. Power controlled multiple access (PCMA)

PCMA, proposed by Monks et al. [36], relies oncontrolling transmission power of the sender so

that the intended receiver is just able to decipherthe packet. This helps in avoiding interference withother neighboring nodes that are not involved inthe packet exchange. PCMA uses two channels,one for sending out busy tones and the other fordata and other control packets. Power controlmechanism in PCMA has been used for increasingchannel efficiency through spatial frequency reuserather than only increasing battery life. Therefore,an important issue is for the transmitter and recei-ver pair to determine the minimum power levelnecessary for the receiver to decode the packet,while distinguishing it from noise/interference.Also, the receiver has to advertise its noise toler-ances so that no other potential transmitter willdisrupt its ongoing reception.

In the conventional methods of collision avoid-ance, a node is either allowed to transmit or not,depending on the result of carrier sensing. InPCMA, this method is generalized to a boundedpower model. Before data transmission, the sendersends a Request Power To Send (RPTS) packet onthe data channel to the receiver. The receiver re-

sponds with an Accept Power To Send (APTS)packet, also on the data channel. This RPTS-APTS exchange is used to determine the minimumtransmission power level that will cause a success-ful packet reception at the receiver. After this ex-change, the actual data is transmitted andacknowledged with an ACK packet.

In a separate channel, every receiver sets up aspecial busy tone as a periodic pulse. The signalstrength of this busy tone advertises to the othernodes the additional noise power the receiver node

can tolerate. When a sender monitors the busytone channel, it is essentially doing something sim-ilar to carrier sensing, as in CSMA/CA model.When a receiver sends out a busy tone pulse, it isdoing something similar to sending out a CTSpacket. The RPTS-APTS exchange is analogousto the RTSCTS exchange. The major differencehowever is that the RPTS-APTS exchange doesnot force other hidden transmitters to backoff.Collisions are resolved by the use of some appro-priate backoff strategy.

A D E H

Range of

DataRange of

ACK

TR for CTSTR for RTS

CS Zone for

RTS

CS Zone for

CTS

G

Fig. 6. Illustration of power control scheme: (CS) carrier senseand (TR) transmission range [35].



12/33

The authors claim improvements in aggregatechannel utilization by more than a factor of 2 com-pared to IEEE 802.11 protocol. Since carrier

sensing while simultaneously transmitting is acomplicated operation, there could be a problemof the ACK packet being subjected to collision.This is an issue because the noise level at thesource cannot be updated during data transmis-sion. This seems to be an open problem with allschemes that use such power control measures.

Woesner et al. [33] also presented the powersaving techniques for IEEE 802.11 and the HighPerformance LAN (HIPERLAN) [46] standards.Chen et al. [37] developed a distributed algorithmcalled Span, wherein every node takes into accountits own power reserve and the advantage to itsneighbors before deciding on staying awake (orgoing to sleep) and acting as a coordinator node.The nodes that are awake take care of routing du-ties. Sivalingam et al. [29] have identified some ofthe ideas that can be used to conserve power atthe MAC layer. They have also performed studieson some protocols in order to compare their per-formance vis-a-vis power efficiency. In fact, powercontrol has also been used for network topologycontrol in [3840] and to generate energy efficient

spanning trees for multicasting and broadcastingin [41,42].

3.3. Multiple channel protocols

A major problem of single shared channelschemes is that the probability of collision in-creases with the number of nodes. It is possibleto solve this problem with multi-channel ap-proaches. As seen in the classification, some mul-ti-channel schemes use a dedicated channel for

control packets (or signaling) and one separatechannel for data transmissions [9,18,20,27,47].They set up busy tones on the control channel, al-beit one with small bandwidth consumption, sothat nodes are aware of ongoing transmissions.

Another approach is to use multiple channelsfor data packet transmissions. This approach hasthe following advantages [52]. First, since the max-imum throughput of a single channel scheme islimited by the bandwidth of that channel, usingmore channels appropriately can potentially in-

crease the throughput. Second, data transmittedon different channels does not interfere with eachother, and multiple transmissions can take place

in the same region simultaneously. This leads tosignificantly fewer collisions. Third, it is easier tosupport QoS by using multiple channels. Schemesproposed in [19,4853] employ such an approach.In general, a multiple data-channel MAC protocolhas to assign different channels to different nodesin real time. The issue of medium access still needsto be resolved. This involves deciding, for instance,the time slots at which a node would get access to aparticular channel. In certain cases, it may be nec-essary for all the nodes to be synchronized witheach other, whereas in other instances, it may bepossible for the nodes to negotiate schedulesamong themselves.

We discuss below the details of some of themultiple channel MAC schemes.

3.3.1. Dual busy tone multiple access (DBTMA)

In the schemes based on the exchange of RTS/CTS dialogue, these control packets themselves areprone to collisions. Thus, in the presence of hiddenterminals, there remains a risk of subsequent datapackets being destroyed because of collisions. The

DBTMA scheme [20] uses out-of-band signaling toeffectively solve the hidden and the exposed termi-nal problems. Data transmission is however on thesingle shared wireless channel. It builds upon ear-lier work on the Busy Tone Multiple Access(BTMA) [9] and the Receiver Initiated-Busy ToneMultiple Access (RI-BTMA) [18] schemes.

DBTMA decentralizes the responsibility ofmanaging access to the common medium and doesnot require time synchronization among the nodes.As in several schemes discussed earlier, DBMTA

sends RTS packets on data channel to set up trans-mission requests. Subsequently, two different busytones on a separate narrow channel are used toprotect the transfer of the RTS and data packets.The sender of the RTS sets up a transmit-busytone (BTt). Correspondingly, the receiver sets upa receive-busy tone (BTr) in order to acknowledgethe RTS, without using any CTS packet.

Any node that senses an existing BTr or BTtdefers from sending its own RTS over the chan-nel. Therefore, both of these busy tones together



13/33

guarantee protection from collision from othernodes in the vicinity. Through the use of the BTtand BTr in conjunction, exposed terminals are

able to initiate data packet transmissions. Also,hidden terminals can reply to RTS requests assimultaneous data transmission occurs betweenthe receiver and sender. The authors claimed a sig-nificant improvement of 140% over the MACAprotocol under certain scenarios. However, theDBTMA scheme does not use ACK to acknowl-edge the received data packets. It also requiresadditional hardware complexity.

Yeh and Zhou [47] have recently proposed anRTS/OTS/CTS (ROC) scheme for efficiently sup-porting networks with devices having heteroge-neous power levels and transmission ranges. Thisscheme uses an additional Object To Send (OTS)control packet. By the use of a separate controlchannel and single data channel, the proposedschemes solved problems due to hidden, exposed,moving, temporarily deaf and heterogeneousnodes. However, the authors did not present thesimulation results to support their claim.

3.3.2. Multi channel CSMA MAC protocol

The multi-channel CSMA protocol proposed

by Nasipuri et al. [48] divides the total availablebandwidth (W) into N distinct channels of W/Nbandwidth each. Here N may be lower than thenumber of nodes in the network. Also, the chan-nels may be divided based on either an FDMAor CDMA. A transmitter would use carrier sensingto see if the channel it last used is free or not. Ituses the last used channel if found free. Otherwise,another free channel is chosen at random. If nofree channel is found, the node should backoffand retry later. Even when traffic load is high

and sufficient channels are not available, chancesof collisions are somewhat reduced since each nodetends to prefer its last used channel instead of sim-ply choosing a new channel at random.

This protocol has been shown to be more effi-cient than single channel CSMA schemes. Interest-ingly, the performance of this scheme is lower thanthat of the single channel CSMA scheme at lowertraffic load or when there are only a small numberof active nodes for a long period of time. This isdue to the waste of idling channels. In [50] the pro-

tocol is extended to select the best channel basedon the signal power observed at the sender side.

3.3.3. Hop-reservation multiple access (HRMA)HRMA [19] is an efficient MAC protocol basedon FHSS radios in the ISM band. Earlier proto-cols such as [54,55] used frequency-hopping radiosto achieve effective CDMA by requiring the radioto hop frequencies in the middle of data packets.HRMA uses time-slotting properties of very-slowFHSS such that an entire packet is sent in the samehop. HRMA requires no carrier sensing, employsa common frequency hopping sequence, and al-lows a pair of nodes to reserve a frequency hop(through the use of an RTSCTS exchange) forcommunication without interference from othernodes.

One of the N available frequencies in the net-work is reserved specifically for synchronization.The remaining (N 1) frequencies are divided intoM= floor ((N 1)/2) pairs of frequencies. Foreach pair, the first frequency is used for Hop Res-ervation (HR), RTS, CTS and data packets, whilethe second frequency is used for ACK packets.HRMA can be treated as a TDMA scheme, whereeach time slot is assigned a specific frequency and

subdivided into four partssynchronizing, HR,RTS and CTS periods. Fig. 7 shows an exampleof the HRMA frame. During the synchronizationperiod of every time slot, all idle nodes synchronizeto each other. On the other three periods, they hoptogether on the common frequency hops that havebeen assigned to the time slots.

A sender-node first sends an RTS packet to thereceiver in the RTS period of the time slot. The re-ceiver sends a CTS packet to the sender in the CTS

s. slot slot 1 slot 2 slot 3 slot 4

f0

SYN HR RTS CTS

f2f0

Fig. 7. Structure of HRMA slot and frame [19].



14/33

period of that same time slot. Now, the sendersends the data on the same frequency (at this time,the other idle nodes are synchronizing), and then

hops to the acknowledgement frequency on whichthe receiver sends an ACK. If the data is large andrequires multiple time slots, the sender indicatesthis in the header of the data packet. The receiverthen sends an HR packet in the HR period of thenext time slot, to extend the reservation of the cur-rent frequency for the sender and receiver. Thistells the other nodes to skip this frequency in thehopping sequence.

The authors claim that HRMA achieves signif-icantly higher throughput than Slotted ALOHA inFHSS channels. It uses simple half-duplex slowfrequency hopping radios that are commerciallyavailable. It however requires synchronizationamong nodes, which is not suitable for multi-hopnetworks.

3.3.4. Multi-channel medium access control

(MMAC)

So and Vaidya proposed MMAC [49], whichutilizes multiple channels by switching amongthem dynamically. Although the IEEE 802.11 pro-tocol has inherent support for multiple channels in

DCF mode, it only utilizes one channel at present[23]. The primary reason is that hosts with a singlehalf duplex transceiver can only transmit or listento one channel at a time.

MMAC is an adaptation to the DCF in order touse multiple channels. Similar to the DPSMscheme [32], time is divided into multiple fixed-time beacon intervals. The beginning of everyinterval has a small ATIM window. During thiswindow ATIM packets are exchanged amongnodes so that they can coordinate the assignment

of appropriate channels for use in the subsequenttime slots of that interval. Unlike other multi-channel protocols (e.g., [5153]), MMAC needsonly one transceiver. At the beginning of everybeacon interval, every node synchronizes itself toall other nodes by tuning in to a common synchro-nization channel on which ATIM packets are ex-changed. No data packet transmission is allowedduring this period of time. Further, every nodemaintains a preferred channel list (PCL) thatstores the usage of channels within its transmission

range, and also allows for marking priorities forthose channels.

If a node has a data packet to send, it sends out

an ATIM packet to the recipient that includes sen-ders PCL. The receiver in turn compares the sen-ders PCL with that of its own and selects anappropriate channel for use. It then responds withan ATIM-ACK packet and includes the chosenchannel in it. If the chosen channel is acceptableto the sender, it responds with an ATIM-RES(Reservation) packet. Any node overhearing anATIM-ACK or ATIM-RES packet updates itsown PCL. Subsequently, the sender and receiverexchange RTS/CTS messages on the selected chan-nel prior to data exchange. Otherwise, if the cho-sen channel is not suitable for the sender, it hasto wait till the next beacon interval to try anotherchannel.

The authors have shown using simulations thatthe performance of MMAC is better than IEEE802.11 and DCA [51] in terms of throughput. Alsoit can be easily integrated with IEEE 802.11 PSMmode while using a simple hardware. However, ithas longer packet delay than DCA. Moreover, itis not suitable for multi-hop ad hoc networks asit assumes that the nodes are synchronized. It

should be interesting to study its extension tomulti-hop networks by using the approach pro-posed by Tseng et al. [34].

3.3.5. Dynamic channel assignment with power

control (DCA-PC)

DCA-PC proposed by Tseng et al. [52] is anextension of their DCA protocol [51] that didnot consider the issue of power control. It com-bines concepts of power control and multiplechannel medium access in the context of MAN-

ETs. The hosts are assigned channels dynamically,as and when they need them. Every node isequipped with two half-duplex transceivers andthe bandwidth is divided into a control channeland multiple data channels. One transceiver oper-ates on the control channel in order to exchangecontrol packets (using maximum power) forreserving the data channel, and the other switchesbetween the data channels for exchanging data andacknowledgments (with power control). When ahost needs a channel to talk to another, it engages



15/33

in an RTS/CTS/RES exchange, where RES is aspecial reservation packet, indicating the appropri-ate data channel to be used.

Every node keeps a table of power levels to beused when communicating with any other node.These power levels are calculated based on theRTS/CTS exchanges on the control channel. Sinceevery node is always listening to the control chan-nel, it can even dynamically update the power val-ues based on the other control exchangeshappening around it. Every node maintains a listwith channel usage information. In essence this listtells the node which channel its neighbor is usingand the times of such usage.

DCA-PC has been shown to achieve higherthroughput than DCA. However, it is observedthat when the number of channels is increased be-yond a point, the effect of power control is less sig-nificant due to overloading of the control channel[52]. In summary, DCA-PC is a novel attempt atsolving dynamic channel assignment and powercontrol issues in an integrated fashion.

3.4. Protocols using directional antennas

MAC protocols for ad hoc networks typically

assume the use of omni-directional antennas,which transmit radio signals to and receives themfrom all directions. These MAC protocols requireall other nodes in the vicinity to remain silent. Withdirectional antennas, it is possible to achieve highergain and restrict the transmission to a particulardirection. Similarly, packet reception at a nodewith directional antenna is not affected by interfer-ence from other directions. As a result, it is possiblethat two pairs of nodes located in each othersvicinity communicate simultaneously, depending

on the direction of transmission. This would leadto better spatial reuse in the other unaffected direc-tions [56]. Using these antennas, however, is not atrivial task as the correct direction should be pro-vided and turned to in real time. Besides, new pro-tocols would need to be designed for takingadvantage of the new features enabled by direc-tional antennas because the current protocols(e.g., IEEE 802.11) cannot benefit from these fea-tures. Currently, directional antenna hardware isconsiderably bulkier and more expensive than

omni-directional antennas of comparable capabili-ties. Applications involving large military vehiclesare however suitable candidates for wireless devices

using such antenna systems. The use of higher fre-quency bands (e.g., ultra wide band transmission)will reduce the size of directional antennas.

Studies have been undertaken for adapting theslotted ALOHA scheme for use with packet radionetworks and directional antennas [57]. Similar re-search on packet radio networks involving multi-ple and directional antennas has also beenpresented in [5860]. Recently, Ramanathan [61]has discussed channel-access models, link powercontrol and directional neighbor discovery, in thecontext of beam forming directional antennas. Ef-fects such as improved connectivity and reducedlatency are also discussed. Bandyopadhyay et al.[62] suggested a scheme in which every nodedynamically stores some information about itsneighbors and their transmission schedulesthrough the use of special control packets. This al-lows a node to steer its antenna appropriatelybased on the on-going transmissions in the neigh-borhood. A method for using the directionalantennas to implement a new form of link-statebased routing is also proposed.

Ko et al. [63] suggested two variations of theirDirectional MAC (D-MAC) scheme using direc-tional antennas. This scheme uses the familiarRTS/CTS/Data/ACK sequence where only theRTS packet is sent using a directional antenna.Every node is assumed to be equipped with severaldirectional antennas, but only one of them is al-lowed to transmit at any given time, depending onthe location of the intended receiver. In this scheme,every node is assumed to be aware of its own loca-tion as well as the locations of its immediate neigh-

bors. This scheme gives better throughput thanIEEE 802.11 by allowing simultaneous transmis-sions that are not possible in current MAC schemes.

Based on the IEEE 802.11 protocol, Nasipuriet al. [64] proposed a relatively simple scheme, inwhich every node has multiple antennas. Any nodethat has data to send first sends out an RTS in alldirections using every antenna. The intended recei-ver also sends out the CTS packet in all directionsusing all the antennas. The original sender isnow able to discern which antenna picked up the



16/33

strongest CTS signal and can learn the relativedirection of the receiver. The data packet is sentusing the corresponding directional antenna in

the direction of the intended receiver. Thus, theparticipating nodes need not know their locationinformation in advance. Please note that only oneradio transceiver in a node can transmit and receiveat a time. Using simulation the authors have shownthat this scheme can achieve up to 23 times betteraverage throughput than CSMA/CA with RTS/CTS scheme (using omni-directional antennas).

Choudhury et al. [56] presented a Multi-HopRTS MAC (M-MAC) scheme for transmissionon multi-hop paths. Since directional antennashave a higher gain and transmission range thanomni-directional antennas, it is possible for a nodeto communicate directly with another node that isfar away. M-MAC therefore uses multiple hops tosend RTS packets to establish links between dis-tant nodes, but the subsequent CTS, data andACK packets are sent in a single hop. Simulationresults indicate that this protocol can achieve bet-ter throughput and end-to-end delay than the basicIEEE 802.11 [23] and the D-MAC [63] schemespresented earlier. The authors however note thatthe performance also depends on the topology

configuration and flow patterns in the system.The use of directional antennas can introduce

three new problems: new kinds of hidden termi-nals, higher directional interference and deafness[56]. These problems depend on the topology andflow patterns. For example, the deafness is a prob-lem if routes of two flows share a common link.Similarly, nodes that are in a straight line witnesshigher directional interference. The performanceof these schemes will degrade with node mobility.Some of the current protocols (e.g., [63,64]) inac-

curately assume that the gain of directional anten-na is the same as that of omni-directional antenna.Similarly, none of them considers the effect oftransmit power control, use of multiple channelsand support for real-time traffic.

3.5. Unidirectional MAC protocols

When low-power and battery-operated nodescoexist with more powerful nodes tethered topower sources in ad hoc networks, disparities in

the transmission powers and asymmetric links be-tween nodes are introduced. Such a network istherefore heterogeneous in terms of power levels.

This gives rise to situations where a node A is ableto transmit to another node B, but B s transmis-sion may not reach A. Recently, some schemeshave been proposed that control the transmissionrange of individual node(s) to maintain optimumnetwork topology [38,40,65]. As a result, theremight be unidirectional links in these networks.

Several studies have been presented on unidirec-tional MAC. Prakash [66] pointed out some of theissues to be taken care of in unidirectional link net-works. In a network of devices having heteroge-neous power levels, when a low power node triesto reserve the channel for data transmission, itmay not be heard due to higher power nodes thatare close enough to disrupt its data exchange. As aresult, a successful RTSCTS exchange does notguarantee successful transmission of data. Fur-thermore, it is important to ensure that the MACprotocol does not favor certain higher powernodes. In order to overcome this problem, Poojaryet al. [10] proposed a scheme to extend the reach ofRTS/CTS exchange information in the IEEE802.11 protocol. This ensures that all hidden high-

er power nodes that could otherwise interfere withthe subsequent DATA transmission are madeaware of the reservation of the channel. Bao etal. [67] proposed a set of collision-free channel ac-cess schemes, known as PANAMA, for ad hocnetworks with unidirectional links. In each conten-tion slot, one or multiple winners are elected deter-ministically to access the channel.

Agarwal et al. [68] summarized the problemscaused by unidirectional links in ad hoc wirelessnetworks and presented some modifications of

MAC and routing protocols. Ramasubramanian[69] presented a Sub Routing Layer (SRL) as abidirectional abstraction over unidirectional linksin lower layers. SRL uses different reverse linksas the abstract reverse link to routing layer.

4. QoS-aware MAC protocols

With the growing popularity of ad hocnetworks, it is reasonable to expect that users will



17/33

demand some level of QoS from them. Some of theQoS related parameters that may be quantified areend-to-end delay, available bandwidth, probability

of packet loss, etc. However, the lack of central-ized control, limited bandwidth, error-prone wire-less channels, node mobility, and power orcomputational constraints makes it very difficultto provide effective QoS in such networks [3,7274].

When the nodes join or leave an ad hoc wirelessnetwork at random, periodic topology updates arerequired so that every node is aware of the currentnetwork configuration. In case the topology of thenetwork changes so rapidly that the routine up-dates are unable to cope up with the same, the net-work is not combinatorially stable and it may notbe possible to guarantee certain levels of QoS.However, if such guarantees are maintainedregardless of the changes in topology, the networkis said to be QoS-robust. Otherwise, if the guaran-tees are maintained between any two consecutiveupdates to the topology, the network is said tobe QoS-preserving [3]. The use of priority to real-ize QoS is known as prioritized QoS. PrioritizedQoS lets the applications specify a higher priorityfor accessing network resources than other appli-

cations. A parameterized QoS involves reservingresources for the end-to-end path of the applica-tion data stream. A new stream is not admittedif enough bandwidth is not available to supportit. This ensures that the already admitted flows re-main unaffected. In certain situations, the conceptof soft or dynamic QoS may be rather useful. Insoft-QoS [75], after the initial connection is setup, there can be brief periods of time when thereis a disruption in providing the pre-decided QoSguarantees. In dynamic-QoS [76], a resource reser-

vation request specifies a range of values (i.e., theminimum level of service that the applicationsare willing to accept and the maximum level of ser-vice they are able to utilize), and the networkmakes a commitment to provide service at a spec-ified point within this range. In such a case, alloca-tion of resources needs to be dynamically adjustedacross all layers of the network. Treating the reser-vations as ranges provides the flexibility needed foroperation in a dynamic ad hoc network environ-ment. Real-time consumer applications such as

streaming audio/video require a reserved share ofthe channel capacity over relatively long durationsso that QoS requirements are met. However, strin-

gent delivery guarantees, particularly on shorttime scales, need not always be fulfilled for suchapplications. Therefore, these applications can besatisfied by soft or dynamic QoS. Other applica-tions such as inter-vehicle communication forsafety require guaranteed delivery of short burstsof data with a bounded delay. These applicationswill require parameterized QoS. In fact, maintain-ing QoS guarantees for delay sensitive traffic isquite difficult in MANETs because obtaining aconsistent network-wide distributed snapshot ofthe state of the queues and the channel at individ-ual nodes at any given instant is an intractableproblem.

The issues affecting support for QoS in ad hocnetworks are briefly discussed below, followed bybrief explanation of selected protocols.

4.1. Issues affecting QoS

Several issues, such as the service model, rout-ing strategies, admission control, resource reserva-tion, signaling techniques, and MAC protocols

need to be considered in the context of providingQoS in ad hoc wireless networks. In fact, everylayer of the network has to be made QoS aware be-cause only when all the factors are considered to-gether in the overall scenario can effective QoSbe provided for the end-user application. Webriefly discuss below the important issues acrossdifferent layers.

A QoS service model specifies an overall archi-tectural framework, within which certain types ofservices can be provided in the network. Some of

the prominent service models suggested for adhoc networks are: flexible quality of service modelfor MANETs (FQMM) [77], cross layer servicemodel [78], and stateless model for wireless adhoc networks (SWAN) [79].

Signaling is used in order to negotiate, reserve,maintain and free up resources, and is one of themost complicated aspects of the network. It shouldbe performed reliably (including topologychanges) with minimum overhead. Out-of-bandand in-band signaling are two commonly used



18/33

approaches. INSIGNIA [80] and Integrated Mo-bile Ad Hoc QoS (iMAQ) framework [81] areexamples of signaling schemes proposed in the

literature.If the application needs to be guaranteed a cer-tain minimum bandwidth or end-to-end delay, therouting scheme should also be QoS aware. Notonly does the route have to be valid at the timethe data is to be transported, but also all the nodesalong that route need to have sufficient resourcesin order to support the QoS requirement of thedata flow and the application. For extensive surveyof routing techniques, the reader is referred to[3,5,72,82]. Once a potential route is established,it is necessary to reserve and allocate the requiredresources in all the nodes of that route so that thedemands of the application can be met. The admis-sion control becomes important in this context.

QoS supporting components at upper layers as-sume the existence of a QoS-aware MAC protocol,which takes care of medium contention, supportsreliable unicast communication, and provides re-source reservation for rt traffic in a distributedenvironment. The MAC protocols are thereforevery important for QoS support, since they havea direct bearing on how reliably and efficiently

data can be transmitted from one node to the nextalong any path in the network. The MAC protocolshould address issues caused by node mobility andunreliable time-varying channel.

4.2. Review of selected QoS-aware MAC protocols

In ad hoc networks, MAC protocols aim tosolve the problem of contention by addressingthe issues of hidden or exposed terminals. Forreal-time applications requiring certain level of

QoS, the MAC layer protocol should also supportresource reservation and real time (rt) traffic.MAC is a lower level function and needs to be clo-sely integrated with upper layers such as the net-work layer for routing. Since centralized controlis not available, it is difficult to maintain informa-tion about connections and reservations.

There are two ways to avoid the use of a central-ized coordinator node in QoS-aware MACschemes. The first approach involves synchronousschemes like Cluster TDMA [83,84], Cluster Token

[85], and Soft Reservation Multiple Access withPriority Assignment (SRMA/PA) [86]. In ClusterTDMA, the nodes are organized into clusters, and

each cluster has a cluster head that is responsiblefor coordinating the activities of the nodes underits purview. Each cluster uses a different DS-SpreadSpectrum code. A common, globally synchronousslotted TDM frame is defined among clusters. Slotscan be reserved by rt traffic and free slots are usedby non-real-time (nrt) data. The rt traffic handlingperformance of the scheme is very good. However,time synchronization is a resource intensive processand should ideally be avoided in ad hoc networks.Similarly, the implementation of multiple codesand associated power control is non-trivial. Themerits of such TDMA schemes have been discussedin [7]. In Cluster Token scheme, the TDM accessscheme is replaced by an implicit token schemewithin each cluster. Also, no synchronization is re-quired across different clusters.

The other option is to use asynchronous ap-proaches that do not require global time synchro-nization and therefore are more suitable for ad hocnetworks. IEEE 802.11 DCF is a widely used asyn-chronous protocol that uses a best effort deliverymodel [2225]. It does not support rt traffic as its

random backoff mechanism cannot provide deter-ministic upper bounds on channel access delays. Anumber of QoS-aware MAC schemes have beenproposed in the past few years and most of themare more or less based on the IEEE 802.11 DCF.No formal classification exists in the literature togroup these schemes. We have attempted to clas-sify below these schemes according to their majorfeatures: i. some schemes, such as real-time MAC[87], DCF with priority classes [88] and enhancedDCF (EDCF) [8991], use shorter inter-frame

spacing and backoff contention values to meetthe delay and bandwidth requirements of rt traffic.These schemes are relatively straightforwardextensions of IEEE 802.11 DCF and can be over-laid on this protocol. ii. The black burst (BB) con-tention [92,93], elimination by sieving (ES-DCF)and deadline bursting (DB-DCF) [94,95] schemesuse a shorter inter-frame spacing and a differentapproach than the backoff window for chan-nel contention to support bounded time delay ofrt traffic. iii. Instead of directly manipulating



19/33

inter-frame spacing and contention window, an-other group of schemes uses reserved time slotsat nodes to provide bounded delay and required

bandwidth for the rt traffic. The nrt data traffic istreated exactly as in IEEE 802.11. Examples of thisclass of schemes are: MACA/PR [96], asynchro-nous QoS enabled multi-hop MAC [97] and dy-namic bandwidth allocation/sharing/extension(DBASE) protocol [98]. iv. The above classes ofschemes may not guarantee a fair proportion ofchannel to different flows. Therefore, someresearchers have proposed MAC schemes (e.g.,distributed fair scheduling [99]) to provide a rea-sonably fair channel allocation to different flows(often according to their priority).

It should be pointed out that the schemes of dif-ferent classes often have some common features.We discuss below salient features of major schemesin each category.

4.2.1. Real-time MAC (RT-MAC)

In IEEE 802.11 protocol, packets that havemissed their deadlines are still retransmitted, eventhough they are not useful any more. This causesbandwidth and resources to be wasted. Baldwinet al. [87] proposed a variation of the IEEE

802.11 protocol called RT-MAC that supports rttraffic by avoiding packet collisions and the trans-mission of already expired packets. To achieve this,RT-MAC scheme uses a packet transmission dead-line and an enhanced collision avoidance schemeto determine the transmission stations next backoffvalue. When an rt packet is queued for transmis-sion, a timestamp is recorded locally in the nodeindicating the time by when the packet should betransmitted. The sending node checks whether apacket has expired at three points: before sending

the packet, when its backoff timer expires and whena transmission goes unacknowledged. An expiredpacket is immediately dropped from the transmis-sion queue. When the packet is actually about tobe sent out, the sending node chooses the nextbackoff value and records it in the packet header.Any node that overhears this packet then ensuresthat it chooses a different backoff value. This elim-inates the possibility of collision. The range of val-ues (i.e., contention window, CW) from which thebackoff value is chosen, is made a function of the

number of nodes in the system. Therefore, thenumber of nodes should be known or at least esti-mated in this scheme.

RT-MAC scheme has been shown to achievedrastic reductions in mean packet delay, misseddeadlines, and packet collisions as compared toIEEE 802.11. However, the contention windowmay typically become quite large in a network withlarge number of nodes. This will result in wastedbandwidth when the network load is light.

4.2.2. DCF with priority classes

Deng et al. [88] proposed another variation ofthe IEEE 802.11 protocol (henceforth calledDCF-PC) that supports priority based access fordifferent classes of data. The basic idea is to usea combination of shorter IFS or waiting timesand shorter backoff time values (i.e., maximumallowable size of contention window) for higherpriority data (i.e., rt traffic). As already mentioned,some different IFS intervals specified in the IEEE802.11 protocol are SIFS, PIFS and DIFS [2325]. While normal nodes wait for the channel to re-main idle after DIFS interval before they transmitdata, a higher priority node waits for only PIFS.However, if the chosen backoff value happens to

be longer, the higher priority node can still loseout to another node that has a larger IFS but ashorter random backoff value. In order to solvethis problem, the authors have proposed two dif-ferent formulae for generating the random backoffvalues so that the higher priority nodes are as-signed shorter backoff time.

Using simulations, the authors have demon-strated that this scheme has better performancethan 802.11 DCF, in terms of throughput, accessdelay and frame loss probability for higher priority

(rt) traffic. It can support more than two traffic pri-orities. However, this scheme lacks the ability toprovide deterministic delay bounds for rt traffic.Moreover, normal data traffic suffers higher delaydue to a longer backoff time even when no higherpriority node is transmitting. Channel bandwidthis also wasted in such cases.

4.2.3. Enhanced DCF

IEEE 802.11 DCF is designed to provide achannel access with equal probabilities to all the



20/33

contending nodes in a distributed manner. EDCFenhances the DCF protocol to provide differenti-ated channel access according to the frame

priorities. It has been developed as a part of thehybrid coordination function (HCF) of IEEE802.11e [8991]. We discuss below its workingprinciple, independent of the details of IEEE802.11e HCF.

Each data frame is assigned a traffic class (TC)in the MAC header, based on its priority asdetermined in the higher layers. During the con-tention process, EDCF uses AIFS[TC], CWmin[TC] and CWmax [TC] instead of DIFS, CWminand CWmax of the DCF, respectively, for a framebelonging to a particular TC. Here AIFS (Arbi-tration Inter Frame Space) duration is at leastDIFS, and can be enlarged individually for eachTC. The CWmin of the backoff mechanism is setdifferently for different priority classes. EDCFthus combines two measures to provide servicedifferentiation. Fig. 8 illustrates the EDCF chan-nel access.

Based on the analysis of delay incurred by IEEE802.11 DCF, Veres et al. [75] proposed a fully dis-tributed Virtual MAC (VMAC) scheme that sup-ports service differentiation, radio monitoring,

and admission control for delay-sensitive andbest-effort traffic. VMAC passively monitors theradio channel and estimates locally achievable ser-vice levels. It also estimates key MAC-level QoSstatistics, such as delay, delay variation, packetcollision, and packet loss.

4.2.4. Black burst (BB) contention

Sobrinho and Krishnakumar [92,93] introducedBB contention scheme in. This scheme is distrib-

uted, can be overlaid on the IEEE 802.11 standardand relies on carrier sensing. The scheme operatesas follows: Normal data nodes use a longer inter-frame spacing than rt nodes. This automaticallybiases the system in favor of the rt nodes. Insteadof sending their packets when the channel becomesidle for a predetermined amount of time, rt nodes

jam the channel with pulses of energy (which aretermed the black bursts) whose length is propor-tional to the contention delay experienced by thenode. This delay is measured from the instant anattempt is made to access the channel until theBB transmission is started.

To uniquely identify all the BB pulses sent bydifferent rt nodes, they all differ in length by atleast one black slot. Following each BB transmis-sion, a node senses the channel for an observationperiod to determine whether its own BB was thelongest or not. If so, the node goes ahead withits data transmission. Otherwise it has to wait forthe channel to be idle before it can send anotherBB. In essence, the scheme seems to achieve a dy-namic TDM transmission structure without expli-

cit slot assignments or synchronization. Itguarantees that rt packets are transmitted withoutcollisions and with a higher priority over others. Ithas also been shown that BB contention enforces around robin discipline among rt nodes (if there ismore than one) and achieves bounded rt delays.

Immediate access when

medium is idle >= AIFS[TC] +

Slot Time

Busy Medium

Contention Window from

[1, CW[TC]+1]

Slot TimeDefer Access

Select Slot and decrement backoff as long

as medium stays idle

AIFS[TC]

+ Slot Time

DIFS

PIFS

SIFSBackoff Window Next Frame

Fig. 8. The EDCF channel access scheme.



21/33

The BB contention scheme thus provides someQoS guarantees to rt multimedia traffic as com-pared to simple carrier sense networks. Applica-

tions considered are those like voice and videothat require more or less periodic access to thechannel during long periods of time denominatedsessions. One of the main considerations in suchapplications is the end-to-end delay. This trans-lates to requiring a bounded packet delay at thedata link layer. However, this scheme does notconsider hidden terminal problem.

4.2.5. Elimination by sieving (ES-DCF) and

deadline bursting (DB-DCF)

Pal et al. [94,95] proposed two variants of theIEEE 802.11 DCF that offer guaranteed timebound delivery for rt traffic, by using deterministiccollision resolution algorithms. Interestingly, theyalso employ black burst features.

The ES-DCF has three phases of operationelimination, channel acquisition and collision reso-lution. In elimination phase, every node is assigneda grade based on the deadlines and priority of itspackets as in [88]. A closer deadline results in alower numerical grade, which translates to lowerthan DIFS channel-free wait times. Therefore,

the grade of the packet improves if it remains inthe queue for a longer time. In the channel acqui-sition phase, the node transmits RTS packet to ini-tiate the channel acquisition, when the channel hasbeen free for the requisite amount of time, asdecided by the grade of its data packet. If it re-ceives a CTS packet in return, the channel is con-sidered acquired successfully. Otherwise, the thirdphase of collision resolution is initiated by sendingout a BB (as in [92,93]). The length of the BB cor-responds to the node identification (Id) number.

Higher Id numbers are given to the nodes that gen-erate a lot of rt data. The node that sends out thelongest burst accesses the channel at the subse-quent attempt.

In the DB-DCF, the first phase is for BB con-tention wherein the lengths of the BB packets areproportional to the urgency (i.e., relative dead-lines) of the rt packet. This is followed by phasesfor channel acquisition and collision resolution,which are similar to the corresponding phases inES-DCF.

Both schemes assign channel-free wait timelonger than DIFS for nrt nodes, such that thesenodes are allowed to transmit only when the other

rt nodes have no data waiting to be sent. However,the results of the simulations carried out by theauthors indicate that ES-DCF is more useful whenhard rt traffic is involved, and DB-DCF performsbetter in the case of nodes with soft rt packets. Dueto the use of BB and longer (than DIFS) channel-free wait time for nrt traffic, these schemes cannotbe directly overlaid on any existing IEEE 802.11DCF implementation.

4.2.6. Multiple access collision avoidance with

piggyback reservations (MACA/PR)

Lin and Gerla [96] proposed MACA/PR archi-tecture to provide efficient rt multimedia supportover ad hoc networks. MACA/PR is an extensionof IEEE 802.11 [2325] and FAMA [15]. Thearchitecture includes a MAC protocol, a reserva-tion protocol for setting up rt connections and aQoS aware routing scheme. We will discuss onlythe MAC protocol here.

In MACA/PR, nodes maintain a special reser-vation table that tells them when a packet is dueto be transmitted. The first data packet in an rt

data stream sets up reservations along the entirepath by using the standard RTSCTS approach.Both these control packets contain the expectedlength of the data packet. As soon as the firstpacket makes such a reservation on a link, a trans-mission slot is allocated at the sender and the nextreceiver node at appropriate time intervals (usuallyin the next time cycle) for the subsequent packet ofthat stream. The sender also piggybacks the reser-vation information for the subsequent data packetin the current data packet. The receiver notes this

reservation in its reservation table, and also con-firms this through the ACK packet. Neighboringnodes overhearing the data and ACK packets, be-come aware of the subsequent packet transmissionschedule, and back off accordingly. The ACK onlyserves to renew the reservation, as the data packetis not retransmitted even if the ACK is lost due tocollision. If the sender consecutively fails to receiveACK N times, it assumes that the link cannot sat-isfy the bandwidth requirement and notifies theupper layer (i.e., QoS routing protocol). Since



22/33

there is no RTSCTS exchange after the first datapacket, collision prevention of rt packets isthrough the use of the reservation tables. For nrt

data packet, MACA/PR uses IEEE 802.11 DCF.Using simulations, the authors have demon-strated that this asynchronous scheme is able toachieve a lower end-to-end delay than otherschemes that require time synchronization suchas Cluster Token and Cluster TDMA. However,since the cluster based schemes use code separa-tion, they can achieve higher aggregate throughputefficiency. Another reason for lower throughputachieved by MACA/PR is that multiple reserva-tion tables need to be kept current at all times sothat the sending node can consult them beforetransmission. This introduces an overhead on thenetwork as the tables are exchanged frequentlyamong neighbors.

4.2.7. Asynchronous QoS enabled multi-hop MAC

Ying et al. [97] proposed an asynchronous pro-tocol based on the IEEE 802.11 DCF, that sup-ports constant bit-rate (CBR) and variable bitrate (VBR) rt traffic, and regular nrt datagramtraffic. In the case of an nrt data transmission,the regular RTSCTSDATAACK sequence is

employed between the sender and the receiver.The acknowledgments sent in response to nrt andrt packets are called D-ACK and R-ACK, respec-tively. Similarly, the nrt and rt data packets aretermed as D-PKT and R-PKT, respectively. Inthe case of rt traffic, though, there is no RTSCTS exchange for the data packets sent after thefirst R-PKT (similar to the MACA/PR scheme[96]). In other words, the R-ACK packet reservesthe transmission slot for the next rt data packet.The scheme requires every node to maintain two

reservation tables, Rx RT and Tx RT. The former(latter) informs the node when neighbors expectincoming (to transmit) rt traffic. These estimatesare recorded in the corresponding tables basedon the overhearing of R-PKT and R-ACK pack-ets. In essence, before sending any RTS, nodeslook for a common free slot based on the entriesin the reservation tables so as not to interfere withrt transmissions already in the queue in the neigh-borhood. Similarly, if a node receives an RTS, itperforms the same checks before responding with

a CTS packet. After a successful RTSCTS ex-change, data is sent out, and an ACK is expected.If an ACK is missed, the node starts to backoff

(using BEB) and uses the IEEE 802.11 contentionwindows for the same.This scheme allows for bounded delays in rt

traffic but depends on the overhearing of R-PKTand R-ACK packets within each nodes transmis-sion range to avoid hidden node problem. Boththe receiver and transmitter nodes check theirown tables, thereby eliminating the overhead ofexchanging table information. Using simulations,the authors have demonstrated that this schemeachieves lower delays for rt traffic than BB Con-tention, MACA/PR and DFS [99] schemes. Thepacket loss rates are also relatively small.

Sheu et al. [98] have proposed the DynamicBandwidth Allocation/Sharing/Extension (DBASE)protocol that also uses a reservation table for sup-porting rt traffic. A unique feature of this scheme isthat bandwidth allocation can change dynamicallyover time, which allows efficient support of CBRas well as VBR traffic. The scheme achieves veryhigh throughput and low packet loss probabilityfor rt-packets even at heavy traffic load, and out-performs the IEEE 802.11 DCF [2325] and DFS

[99] schemes. DBASE, however, assumes that allthe nodes can hear one another and it may be dif-ficult to extend it to the (multi-hop) ad hoc net-works with hidden terminals. Overall, DBASE isa quite different scheme than the other two (previ-ously discussed) reservation based schemes.

4.2.8. Distributed fair scheduling (DFS)

Vaidya et al. [99] proposed the DFS scheme toensure that different flows sharing a common wire-less channel are assigned appropriate bandwidth

corresponding to their weights or priorities. DFSis derived from the IEEE 802.11 DCF and requiresno central coordinator to regulate access to themedium. The fundamental idea of DFS is thateach packet is associated with start and finish time-stamps. A higher priority packet is assigned asmaller finish-tag and shorter backoff periods.This approach ensures that any flow that haspackets of higher priority will consistently haveshorter backoff times, thereby achieving a higherthroughput.



23/33

In DFS, the start and finish times for packetsare calculated on the basis of the Self-Clocked FairQueuing (SCFQ) algorithm proposed by Golestani

[100]. Following the idea of SCFQ, every node alsomaintains a local virtual clock. DFS does not,however, REMOVE short-term unfairness in cer-tain cases. The authors observe that

medium access control protocols for ad hoc wireless networks

Documents