a mac layer protocol for wireless networks with asymmetric...

Available online at www.sciencedirect.com

Ad Hoc Networks 6 (2008) 424–440

www.elsevier.com/locate/adhoc

A MAC layer protocol for wireless networkswith asymmetric links

Guoqiang Wang, Damla Turgut *, Ladislau Boloni,Yongchang Ji, Dan C. Marinescu

School of Electrical Engineering and Computer Science, University of Central Florida, Orlando, FL 32816-2450, USA

Received 5 June 2006; received in revised form 25 February 2007; accepted 17 March 2007Available online 24 March 2007

Abstract

We introduce AsyMAC, a MAC layer protocol for wireless networks with asymmetric links and study a protocol stackconsisting of AsyMAC and the A4LP routing protocol. The two protocols are able to maintain connectivity where thestandard IEEE 802.11 MAC protocol coupled with either AODV or OLSR routing protocols may loose connectivity.A comparative study shows that AsyMAC improves on two previously proposed protocols’ accuracy in determiningthe nodes to be silenced to prevent collisions.Published by Elsevier B.V.

Keywords: Asymmetric link; AsyMAC; Heterogeneous MANET

1. Introduction

Asymmetric links are present in wireless net-works for a variety of physical, logical, operational,and legal reasons:

(a) The transmission range is limited by the node

hardware. The hardware properties of thenode (for instance, the antenna or the RFcircuits) determine the maximum transmissionrange. The different transmission ranges of thenodes lead to asymmetric links, which cannot

1570-8705/$ - see front matter Published by Elsevier B.V.

doi:10.1016/j.adhoc.2007.03.004

* Corresponding author. Tel.: +1 407 823 6171; fax: +1 407 8235835.

E-mail address: [email protected] (D. Turgut).

be avoided except by physically changing thenodes’ hardware components, for instance byinstalling a different antenna.

(b) Power limitation. Different nodes may havedifferent power constraints. For instance, nodeA may have sufficient power reserves and atransmission range enabling it to reach nodeB; however, node B has limited power, andeither (i) cannot reach node A, or (ii) maychoose not to reach node A to save power.The two scenarios influence the design of theprotocols in different ways. In the secondscenario, node B is capable to reach node Aand we could exploit this capability for shorttransmissions when necessary, e.g., during anetwork setup phase.

mailto:[email protected]

G. Wang et al. / Ad Hoc Networks 6 (2008) 424–440 425

(c) Interference. Node A can reach node B andnode B can reach node A, but if node B wouldtransmit at a power level sufficient to reachnode A, it would interfere with node C whomight be a licensed user of the spectrum. Thisscenario is critical for transmitters whichattempt to opportunistically exploit unusedparts of the licensed spectrum (such as unusedtelevision channels). Even if operating in theunlicensed bands, dynamic spectrum manage-ment arrangements might have given the pri-ority to node C, thus node B needs to refrainfrom sending at a power level above a giventhreshold.

(d) Stealth considerations. Node A and node Battempt to communicate and they wish to hidethe existence or the exact location of node Bfrom node O. One way to achieve this is torestrict the transmission power of node B tothe minimum and/or transmit on frequencieswhich make location detection more difficult.This is especially important in military/battle-field applications where low probability ofdetection (LPD) is an important consideration[20,21].

(e) Dynamic spectrum management. In the emerg-ing field of software defined radios, the nodescan transmit virtually in any band across thespectrum, but they need to share the spectrumwith devices belonging to licensed operators aswell as devices with limited flexibility. Onceany of the reasons discussed previously forcea link to be unidirectional additional con-straints, e.g., the need for a reverse pathbetween some pairs of nodes may cause otherlinks to change their status and operate in aunidirectional mode, even when there is noother reason for unidirectionality.

Inability of some MAC protocols to exploit theasymmetry of some of the communication channelscould lead to an inefficient bandwidth utilization,or, in the worst case, to inability to connect someof the nodes. To exploit the asymmetric links, theprotocols must be able to deliver the acknowl-edgements back to the sender in a direction oppositeto the direction of the asymmetric link. Further-more, the problem of hidden nodes appears moreoften and in more complex forms than in the caseof symmetric links. Depending whether the routingprotocol of the wireless ad hoc network is ableto handle asymmetric links, the MAC protocol

might need to hide the existence of asymmetric linkswith a symmetric overlay. The challenge for a MAClayer protocol able to exploit asymmetric links is tosolve the hard problems mentioned above, whilekeeping the cost incurred lower than the benefitsobtained from the utilization of the asymmetriclinks.

MAC protocols for asymmetric links were pre-viously proposed by Poojary et al. [15], Fujii et al.[6] and others. In this paper, we introduce a newprotocol, AsyMAC (asymmetric MAC) that uses ageometric analysis of the hidden node problem inthe presence of asymmetric links for a more precisedetermination of the nodes which need to besilenced during a transmission. Informally, a hiddennode is one that can interfere with the reception of adata packet without the knowledge of the sender. Asa note, there is a difference between the concept of aprotocol, as the collection of features necessary toimplement networking at a certain layer, and algo-rithm, which refers to the implementation of a spe-cific functionality. In this paper, when we refer to aprotocol, we concentrate on the subset of function-ality necessary to implement the asymmetric links,thus the terms algorithm and protocol will be usedinterchangeably.

The paper is organized as follows. Related workis presented in Section 2. Section 3 presents Asy-MAC protocol in every aspects. Section 4 describesthe simulation environment and presents the resultsof the simulation study of the effect of network load,network mobility and number of nodes. We con-clude in Section 5.

2. Related work

MAC layer protocols allow a group of users toshare a communication medium in a fair, stable,and efficient way. A MAC layer protocol for wire-less ad hoc networks must address several specificproblems:

1. Mobility – the connection between nodes canbecome unstable because of the independentmovement of the nodes.

2. Higher error rates – a wireless channel has ahigher bit error rate (BER) than a wired network.

3. Inability to detect collisions during some periodsof time – wireless transceivers work in a half-duplex mode; nodes do not ‘‘listen’’ when ‘‘talk’’and do not ‘‘talk’’ when ‘‘listen’’. The sender isunable to detect the collision and the receiver is

426 G. Wang et al. / Ad Hoc Networks 6 (2008) 424–440

unable to notify the sender of the collision duringthe transmission of a packet. Collision avoidance

is almost mandatory.

Carrier sensing multiple access (CSMA) [10]requires every node to sense the channel beforetransmitting, and if the channel is busy, refrain fromtransmitting a packet. CSMA reduces the possibilityof collisions in the vicinity of the sender. Multipleaccess collision avoidance (MACA) [9] and its vari-ant MACAW [2] are alternative medium access con-trol schemes for wireless ad hoc networks that aimto solve the hidden node problem by reducing thepossibility of collisions in the vicinity of the receiver.

The floor acquisition multiple access (FAMA) [7]protocol consists of both carrier sensing and a colli-sion avoidance handshake between sender andreceiver of a packet. Once the control of the channelis assigned to one node, all other nodes in the net-work should become silent. Carrier sensing multipleaccess based on collision avoidance (CSMA/CA),the combination of CSMA and MACA, is consid-ered a variant of FAMA protocols. The IEEE802.11 standard [8] is the best-known instance ofCSMA/CA.

In a wireless network with symmetric links, a hid-

den node is one out of range of the sender, but in the

range of the receiver. The solution provided bythe 802.11 MAC to the hidden node problem isthe RTS/CTS handshake mechanism. Xu et al.[11] analyzes the effectiveness of RTS/CTS hand-shake mechanism, and indicates that some of thehidden nodes may not be covered by the receiverdue to the fact that it requires much lower powerto interrupt a packet reception than to successfullydeliver a packet.

We can define a hidden node in wireless ad hocnetworks with asymmetric links as a node out of

s r k

RTSDATA

CTSACK

CTS

Fig. 1. (a) Hidden node problem in a ‘‘classical’’ wireless network withnode is a node out of the range of the source and in the range of the renode r. (b) Hidden node problem in a heterogeneous wireless network wsender and whose range covers the receiver. k is a hidden node for tra

the range of the sender and whose range covers the

receiver (see Fig. 1b). Thus, a hidden node is hiddenfrom the sender and possibly from the receiver aswell. The RTS/CTS handshake mechanism is nota solution for such networks since a CTS packetmay not be able to reach hidden nodes.

Several solutions to the hidden node problem inwireless ad hoc networks with asymmetric linksare discussed in the literature. Poojary et al. [15]propose that a node rebroadcasts a CTS packet ifit is received from a low-power node. To decreasethe probability of collisions, each node waits a ran-dom number (1, . . . , 6) of SIFS (short inter-framespacing) periods before transmitting a CTS packet.Fujii et al. [6] made several improvements relativeto [15]: (i) not only CTS but also RTS packets arerebroadcasted; (ii) nodes with a CTS packet torebroadcast, first sense the channel and transmitonly if the channel is not busy; and (iii) only high-power nodes rebroadcast RTS or CTS packets.The solutions proposed by [15,6] can lead to ineffi-cient use of the channel if nodes are misclassified

as hidden nodes. In such situations, nodes thatcould have been active are silenced due to misclassi-fication, severely degrading the channel utilization.Refs. [15,6] routinely assume routing over symmet-ric links so that the sender is able to receive bothCTS and ACK packets. In the presence of asymmet-ric links, however, the sender might not receive theCTS or ACK packets, thus the sender cannot triggerthe transmission of DATA packets, and does notknow whether a transmission was successful or not.

Bao et al. [1] propose a collision-free dynamicchannel access scheduling algorithm PANAMA.Two scheduling algorithms are proposed for net-works with unidirectional links, NAMA-UN thatis node activation oriented and supports broadcasttraffic efficiently, and PAMA-UN that is link activa-

s r kj

CTS XCTS

RTSDATA

CTSACK

mobile nodes. All links are assumed to be bidirectional. A hiddenceiver node. k is a hidden node for a transmission from node s toith mobile nodes. A hidden node is a node out of the range of the

nsmission from node s to node r.


tion oriented and is more suitable for relaying uni-cast traffic. The channel access is allocated forNAMA-UN and PAMA-UN alternatively, witheach scheduling algorithm lasting for a fixedamount of time. In PANAMA, the sender node isable to detect the hidden node that also attemptsto relay traffic to the receiver. The winner of a con-tention is the node with higher priority. When thelink from the hidden node to the receiver is unidirec-tional, the hidden node may not be aware of the sen-der. In these cases, the hidden node is automaticallyconsidered as having a higher priority.

The protocols considered previously are based onthe modification of the MAC protocol. In contrast,the sub routing layer (SRL) project [17,16] handlesasymmetric networks by adding an intermediarylayer between the MAC and network layers. Thislayer partially isolates the routing protocol fromthe MAC layer, although it still allows the routingprotocol to directly contact the MAC layer. Forunidirectional links, reverse paths are computedusing the reverse distributed Bellman–Ford algo-rithm. The SRL implementation also signals thedetection of new neighbors and the loss of (unidirec-tional) links.

A MAC layer protocol able to utilize asymmetriclinks should be stacked together with routing proto-cols that can utilize asymmetric links as well. A4LP[19] is a location-aware and power-aware routingprotocol designed for ad hoc networks with asym-metric links. In A4LP, neighbors are re-classifiedas In-bound, Out-bound, and In/Out-bound neigh-bors due to the asymmetry of links. A4LP iscomposed by a neighbor discovery protocol, a pathdiscovery protocol, and a path maintenance mecha-nism. A4LP proposes an advanced flooding tech-nique – m-limited forwarding. Receivers can re-broadcast a packet only if its fitness value exceedsa predefined threshold, specified by the sender.The fitness function used by m-limited forwardingcan be tuned to minimize the power consumption,maximize the stability of the routes, minimize theerror rates or the number of retransmissions. Byavoiding a full broadcast, m-limited forwardingreduces the cost of path discovery. A4LP is a hybridad hoc routing protocol, combining features of bothpro-active and on-demand protocols. The routes toIn-, Out-, and In/Out-bound neighbors are main-tained by periodic neighbor update and immediatelyavailable upon request, while the routes to othernodes in the network are obtained by a path discov-ery protocol.

In the following sections, we introduce a newMAC layer protocol for ad hoc networks withasymmetric links (AsyMAC). AsyMAC currentlyworks with A4LP, since they share the process ofneighbor discovery and neighbor maintenance.

3. The asymmetric MAC (AsyMAC) protocol

3.1. Topological considerations

The handling of the hidden nodes is an essentialproblem for wireless MAC protocols operating inthe presence of asymmetric links. We introducetopological concepts necessary to define a hiddennode of a network with asymmetric links.

The connection between two nodes is describedby the Boolean reachability function Rði; j; tÞ whichcan be interpreted as follows: a node i can send apacket to node j at time t if and only ifRði; j; tÞ ¼ true. A link between two nodes is sym-metric if Rði; j; tÞ ¼ Rðj; i; tÞ ¼ true. Note that thereachability is a time varying function; the connec-tion can be affected by various channel conditions,fading, the mobility of the node or the mobility ofthe obstacles in the field.

We assume that every node is aware of the cur-rent values of the reachability function betweenitself and the neighboring nodes (in both direction).In the A4LP/AsyMAC protocol stack, it is the roleof the neighbor discovery protocol of A4LP to findthese values and keep them up-to-date. Althoughneighbor discovery is a common feature of ad hocrouting protocols, most protocols will not detectoutbound neighbors, because the confirmation mes-sage will not reach back to the originating node.Asymmetric routing protocols, such as A4LP havea provision to route back the confirmation messageseven in the absence of a direct link, thus allowingthe discovery of the full asymmetric reachabilitymatrix. In the following definitions we omit the timeparameter even though all the sets are variable intime, a fact which needs to be considered by the pro-tocols relying on them.

We define a series of topological concepts relatedto communication in the presence of asymmetriclinks and illustrate for the simple scenario inFig. 2; a sender node s sends a packet to the receivernode r in the vicinity of nodes 1, . . . , 9. The circlescentered at s and r show the transmission ranges ofthe sender and the receiver, respectively. The reach-ability information of other nodes is shown by direc-ted lines; to avoid cluttering the figure we do not

s r

12

3

4

6

7

8

9

5

Fig. 2. An illustration for topology concepts. The transmissionranges of the sender s and the receiver r are reflected by the circlescentered at them. The partial reachability information of othernodes is shown by directed lines.


include the links not relevant to the scenario. In thissimple scenario we assume that the asymmetric linksare caused by the nodes having different transmis-sion ranges and the transmission range is a disk; thisis not necessarily true in real life scenarios, and ourdefinitions do not assume a unit disk model.

Definition 1. A set of m nodes i1; i2; . . . im 2 N are inan m-party proxy set if each node can reach theother m � 1 nodes either directly or through asubset of the other m � 2 members.

For instance, in the scenario in Fig. 2 the three-party proxy sets are {r, 1,6}, {r, 2,6}, {r, 2,7},{r, 3,7}, {r, 3,8}, {r, 4,8}, and {r, 4,9}.

Definition 2. Call the vicinity of node i, Vi the set ofall nodes that could be reached from node i.

V i ¼ fjjRði; jÞg: ð1ÞIn our scenario, the vicinity of the receiver node r

is Vr = {1,2,3,4,5}.

Definition 3. Call Hsr the set of hidden nodes of a

transmission Tsr. Hsr includes nodes that are notreachable from the sender, but from which thereceiver is reachable:

Hsr ¼ fkj:Rðs; kÞ ^ Rðk; rÞg: ð2ÞNote that Hsr are the hidden nodes for the trans-

mission of the DATA packets, while Hrs are the hid-den nodes for the transmission of ACK packets.

In our scenario, the hidden nodes of the trans-mission from source node s to receiver node r areHsr = {2,3,4,6,7,8,9}.

Definition 4. Call P3i the three-party proxy set

coverage of node i. P3i is the set of nodes whichare either reachable by node i directly or participatein a three-party proxy set with node i and a thirdnode.

P3i ¼ fkjRði; kÞ _ 9jðRði; jÞ ^ Rðj; kÞ ^ Rðk; iÞÞg:ð3Þ

In the scenario of Fig. 2 the three-party proxy setcoverage of node r is P3r = {1, 2,3,4,5,6,7,8,9}.

Definition 5. Call H3sr the hidden nodes of atransmission Tsr in the three-party proxy set cover-age of node r. The set H3sr includes hidden nodescovered by P3r.

H3sr ¼ Hsr \ P3r: ð4Þ

In our scenario, the hidden nodes in the three-party proxy set coverage of r are H3sr = {2,3,4,6,7,8,9}.

Definition 6. Call XH3sr the extended hidden nodes

of a transmission Tsr in three-party proxy setcoverage of node r. The set XH3sr includes nodesin H3sr not covered by Vr.

XH3sr ¼ H3sr � V r: ð5Þ

In the scenario of Fig. 2, the extended hiddennodes of the transmission from source node s toreceiver node r are XH3sr = {6, 7,8,9}.

Definition 7. Call XHR3sr the extended hidden nodesrelay set of a transmission Tsr in three-party proxyset coverage of node r. XHR3sr includes all nodes inP3r that could relay traffic from node r to nodesbelonging to XH3sr.

XHR3sr ¼ fjjj 2 V r ^ 9k2XH3srðRðj; kÞÞg: ð6ÞThe extended hidden nodes relay set of the trans-

mission from s to r on the example scenario isXHR3sr = {1,2,3,4}.

Definition 8. Call mXHR3sr a minimal extended

hidden nodes relay set of a transmission Tsr inthree-party proxy set coverage of node r. mXHR3sr

includes a subset of nodes from XHR3r

(mXHR3r � XHR3r) such that (i) the node r canrelay traffic to any node in XH3sr through somenodes from mXHR3sr and (ii) the removal of anynode from mXHR3sr makes some nodes in XH3sr

unreacheable from r.


8k2XH3sr9j2mXHR3srðRðj; kÞÞ ð7Þand

8j02mXHR3sr9k2XH3sr 9= j2mXHR3sr�fj0gðRðj; kÞÞ: ð8Þ

Note that mXHR3sr may not be unique, and dif-ferent minimal extended hidden nodes relay setscould contain a different number of nodes. Call{mXHR3sr} the set that contains all possible setsof mXHR3sr.

For instance, in our scenario there are two possi-ble minimal extended hidden nodes relay sets:mXHR3sr = {2,4} and mXHR3sr = {1,3,4}. Alsonote that the two sets have a different number ofnodes.

Definition 9. Call MXHR3sr the minimum extended

hidden nodes relay set of a transmission Tsr in three-party proxy set coverage of node r. MXHR3sr is theinstance of mXHR3sr with the smallest number ofnodes. Call {MXHR3sr} the set that contains allpossible sets of MXHR3sr.

In our scenario, we need to simply pick the small-est of the possible mXHR3sr sets, which in our casewill be MXHR3sr = {2, 4}.

We note that all the definitions provided aboveare constructive, providing their own implementationmethodology. Every set is defined based on the cas-cade of definitions preceding it, and all of them canbe reduced to the reachability matrix Rði; jÞ.

3.2. Determination of the sets in AsyMAC

The sets Vr and P3r of node r are the directresults of the neighbor discovery protocol ofA4LP. Based on which, we can determine the setsin AsyMAC.

1. Hsr includes all the hidden nodes of a transmis-sion Tsr, which might be outside of the three-party proxy coverage of node r (P3r), thus thecomplete set of nodes of Hsr may not be foundand is not maintained.

2. The members of H3sr can be found by removingfrom the set P3r the nodes that can be reached bythe other peer of the transmission. Note that inA4LP/AsyMAC, the reachability information oftwo neighbors of a node can be calculated basedon their locations and transmission ranges.

3. XH3sr is obtained by XH3sr = H3sr � Vr.4. XHR3sr includes all nodes in Vr that can reach a

node in set XH3sr.

5. The calculation of {mXHR3sr} is described inAlgorithm 1.

Algorithm 1 (Calculation of {mXHR3sr}).

1: {mXHR3sr} = U;2: List the complete permutation of XHR3sr, call

it P.3:4: /* Find mXHR3sr for each permutation

Pi 2 P. */5: for all permutations Pi 2 P do

6: mXHR3sr = U;7: T = XH3sr;8: found = false;9: while Pi5U § found = false do

10: remove the next node p from Pi, Pi =Pi � {p};

11: mXHR3sr = mXHR3sr [ {p};12: for all nodes t 2 T do

13: if Rðp; tÞ then

14: T = T � {t};15: end if

16: if T = U then

17: found = true;18: break;19: end if

20: end for

21: end while

22: add mXHR3sr to {mXHR3sr};23: end for

24:25: /* Remove all sets M from {mXHR3sr} if there

exists M 0 2 {mXHR3sr} such that M 0 �M. */26: for all M 2 {mXHR3sr} do

27: for all M 0 2 {mXHR3sr} do

28: if M 0 �M then

29: remove M from {mXHR3sr};30: break;31: end if

32: end for

33: end for34:35: return{mXHR3sr};

6. {MXHR3sr} includes the set(s) in {mXHR3sr}with the smallest cardinality. During the processof constructing {MXHR3sr}, we can ignore theminimal extended hidden nodes set whose cardi-nality already exceeds the achieved minimum


value, which becomes our incentive to improvethe algorithm. The calculation of {MXHR3sr} isdescribed in Algorithm 2.

Algorithm 2 (Calculation of {MXHR3sr}).

1: {MXHR3sr} = U;2: List the complete permutation of XHR3sr, call

it P.3: min_cardinality = MAX;4:5: /* Find mXHR3sr for each permutation

Pi 2 P. */6: for all permutations Pi 2 P do

7: mXHR3sr = U;8: T = XH3sr;9: found = false;

10: while Pi5U § found = false § |mXHR3sr|< min_cardinality do

11: remove the next node p from Pi, Pi =Pi � {p};

12: mXHR3sr = mXHR3sr [ {p};13: for all nodes t 2 T do

14: if Rðp; tÞ then

15: T = T � {t};16: end if17: if T = U then

18: found = true;19: break;20: end if

21: end for

22: end while

23:24: /* if |mXHR3sr| is less than the current

achieved minimum cardinality, updatemin_cardinality and remove all elements from{MXHR3sr}.

25: if found = true then

26: if |mXHR3sr| < min_cardinality then

27: min_cardinality = |mXHR3sr|;28: {MXHR3sr} = U;29: end if30: add mXHR3sr to {MXHR3sr};31: end if

32: end for

33:34: return {MXHR3sr};

3.3. Accuracy metrics for node classification

We introduce a set of metrics characterizing theability of a MAC protocol to silence nodes which

could cause collisions. Ideally, an algorithm shouldsilence all nodes that have the potential to be hiddennodes, as well as nodes that could potentially beaffected by the transmission Tsr. Assume there existsan algorithm I which constructs the set of all thenodes that should be silenced during a transmissionTsr:

SsrðIÞ ¼ Hsr [ H rs [ V s [ V r: ð9ÞIn practice, the set of nodes silenced by an algo-

rithm F , SsrðFÞ, might contain nodes that aresilenced unnecessarily (misclassified) and might lacknodes which should have been silenced (missednodes).

Call MiscsrðFÞ the misclassification ratio of analgorithm F for a transmission Tsr. MiscsrðFÞ mea-sures the ratio of nodes that are incorrectly silencedby F

MiscsrðFÞ ¼jSsrðFÞ � SsrðIÞj

jSsrðIÞj: ð10Þ

Call MisssrðFÞ the miss ratio of an algorithm Ffor a transmission Tsr. Misssr(F ) measures the ratioof nodes which are not silenced by the algorithm F ,although they should have been

MisssrðFÞ ¼jSsrðIÞ � SsrðFÞj

jSsrðIÞj: ð11Þ

Let MiscðFÞ and MissðFÞ be the average mis-

classification ratio and average miss ratio of an algo-rithm F , respectively. The averages are computedover a network N

MiscðFÞ ¼P8s;r2NRðs;rÞjSsrðFÞ � SsrðIÞjP8s;r2NRðs;rÞjSsrðIÞj

; ð12Þ

and

MissðFÞ ¼P8s;r2NRðs;rÞjSsrðIÞ � SsrðFÞjP8s;r2NRðs;rÞjSsrðIÞj

: ð13Þ

3.4. A solution to the hidden node problem

In a wireless ad hoc network with asymmetriclinks, the sender may not be able to receive theCTS or ACK packets from the receiver. In such acase a DATA packet, or the next frame cannot besent. The IEEE 802.11 protocol assumes that allthe connections are symmetric. Our protocol relaxesthis assumption, asymmetric links can be used pro-vided that they are part of a three-party proxy set

[19].

s r

j2

CTS XCTSDATA

j

CTSACK

TCTSTACK

j1

RTS

k1

k2

XRTS

RTS

Fig. 3. Routing over asymmetric links in a heterogeneouswireless ad hoc network. Node s is the sender, r is the receiver,the link from node s to r is asymmetric, and node j is the proxynode that can relay traffic to s for r. Nodes k1 and k2 are hiddennodes for transmissions Trs and Tsr, respectively. Nodes j1 and j2are the proxy nodes that can relay traffic from s to k1 and from r

to k2, respectively.


Our protocol retains the use of RTS, CTS,DATA and ACK frames defined in IEEE 802.11standard. In addition, we introduce four newframes: XRTS (Extended RTS), XCTS (ExtendedCTS), TCTS (Tunneled CTS), and TACK (Tun-neled ACK).

An ideal MAC layer protocol should be basedupon a scheme that delivers the RTS and CTS pack-ets to all hidden nodes in Hrs and Hsr, respectively.However, such a scheme can be impractical because(i) a node may not have knowledge of all its In-bound neighbors; (ii) the number of hops neededto reach an In-bound neighbor might be large, thusthe time penalty and the power consumptionrequired for the RTS/CTS diffusion might outweighthe benefits of a reduced probability of collision.

Our solution is to send RTS and CTS packets tothe nodes in H3rs and H3sr respectively. In this way,a considerable number of nodes that are misclassi-fied as ‘‘hidden’’ nodes by [15], referred to as proto-col A, and [6], referred to as protocol B, are allowedto transmit. Note that our approach does not iden-tify all hidden nodes, but neither methods A or Bare able to identify all hidden nodes.

3.5. Node status

In IEEE 802.11, when a node overhears a RTS ora CTS packet, it becomes silent and cannot send anypacket until its NAV expires. This way, nodes in therelay set cannot send XRTS/XCTS as they shouldbe in a silent state after overhearing the RTS/CTSpacket. To resolve this dilemma, we replace thesilent state with a quasi silent state, in which a nodeis allowed to send control packets, except RTS andCTS.

The medium access control model proposed in thispaper classifies a node as either idle, active, quasi

silent, or silent. When a node is idle, it is able to sendor receive any type of packets. When a node is active,it is either sending or receiving a packet. When a nodeis in the quasi silent state, it can either receive packetsor send any packet type except RTS, CTS, or DATA.When a node is in the silent state, the node can receivepackets but cannot send any packet.

3.6. Medium access control model

The medium access control (MAC) model of ourprotocol is based upon an extended four-way hand-shake (Fig. 3). For short data frames, there isno need to initiate a RTS/CTS handshake (see

Fig. 4a and b). For long data frames, we recognizeseveral phases (see Fig. 4c and d):

1. Sensing. The sender s senses the medium. If it doesnot detect any traffic for a DIFS period, the senderstarts the contention phase; otherwise, it backs offfor a random time before it senses again.

2. Contention. The sender s generates a randomc 2 [0, contention window] slot time. The senders starts a transmission if it does not detect anytraffic for c time.

3. RTS transmission. The sender s sends a RTSpacket to the receiver r. The RTS packet specifiesthe NAV(RTS), link type of Lsr and MXHR3rs.The link type field is used to determine whethersymmetric or asymmetric medium access controlmodel is used.

4. CTS transmission. The receiver r checks whetherthe link is symmetric or not. If link Lsr is symmet-ric, node r sends a CTS packet back to node s;otherwise, node r sends a TCTS packet to nodes. A TCTS packet specifies both the proxy nodeand the receiver r. The proxy node forwards theTCTS packet to the original sender s after receiv-ing it. A CTS/TCTS packet can be sent only aftersensing a free SIFS period. Instead of MXHR3sr,MXHR3rs �MXHR3sr is specified in the CTS/TCTS packet so that every extended hidden noderelay is included only once. Thus, the duration ofXCTS/XRTS diffusion phase can be reduced.

Node s

Node r

DATA

ACK

DIFS

SIFS t

t

CW

Node s

Node r

Node j

DATA

ACK

TACK

DIFS

SIFS

SIFS

t

t

t

CW

Node s

Node r

Node j

Node j2

RTS

XCTS

CTS

XCTS

CW

Node j1 XRTS

DATA

ACK

DIFS

SIFS SIFS

SIFS

random SIFSs

random SIFSs

random SIFSs

t

t

t

t

t

6SIFS

Node s

Node r

Node j

Node j2

RTS

XCTS

CTS

XCTS

CW

Node j1 XRTS

DATA

ACK

TACK

DIFS

SIFS SIFS

SIFS

SIFS

random SIFSs

random SIFSs

random SIFSs

t

t

t

t

t

TCTS

SIFS

6SIFS

Fig. 4. The medium access control model for the MAC layer protocol introduced in this paper for the scenario in Fig. 3. (a) The mediumaccess control model of proposed MAC protocol for short data frames over a symmetric link. (b) The medium access control model ofproposed MAC protocol for short data frames over an asymmetric link. (c) The medium access control model of proposed MAC protocolfor long data frames over a symmetric link. (d) The medium access control model of proposed MAC protocol for long data frames over anasymmetric link.


5. XRTS/XCTS diffusion. All nodes that overhear aRTS/CTS/TCTS packet enters a quasi silent

state. After the CTS transmission phase, allextended hidden node relays that are either spec-ified in RTS or CTS/TCTS starts contention forbroadcasting XRTS/XCTS to its neighbors.When a node captures the medium, all othernodes back-off for a random number of(1, . . . , 4) SIFS periods, and continue the conten-tion until the XRTS/XCTS diffusion phase fin-ishes. An XRTS/XCTS diffusion phase lasts for6 SIFS periods, after which all nodes except theproxy node become silent.

6. Data transmission. When the XRTS/XCTS diffu-sion phase finishes, the sender s starts sending

DATA packets to the receiver r after sensing afree SIFS period.

7. Acknowledgement. Once the receiver r success-fully received the DATA packet from the senders, it replies with an ACK if link Lsr is symmetric,or a TACK packet if link Lsr is asymmetric. AnACK/TACK packet can be sent only after sens-ing a free SIFS period. When the sender s

receives an ACK/TACK packet, it starts con-tending the medium for the next frame. Mean-while, the NAVs that are reserved for thistransmission should expire.

At any moment, if a node overhears a packetcontaining new NAV information, it compares it


with the currently stored NAV, and retains theNAV which expires later.

4. Simulation and case study

We have implemented the AsyMAC protocol inNS-2 [3,18], an object-oriented event-driven simula-tor developed at the Lawrence Berkeley NationalLaboratory, with the CMU wireless extensions[13]. As AsyMAC requires a routing protocol ableto handle asymmetric links, we paired it withA4LP routing protocol to form a complete ad hocnetworking stack. In our experiments, we comparethe A4LP/AsyMAC pair against the standard IEEE802.11 protocol coupled with AODV [14], a widelyused on-demand ad hoc routing protocol and themore recent OLSR [5] protocol.

The simulation results reflect the performance ofthe pair of the corresponding MAC and routingprotocols rather than the performance of theMAC or routing protocols alone. We had chosenthis experimental setup because it provides the mostinformative comparison of real scenarios. We can-not run a routing protocol which does not supportasymmetric links on top of AsyMAC. On the otherhand, A4LP can be run on top of MAC protocolswhich do not support asymmetric links. However,A4LP has a higher overhead than routing protocolswhich assume symmetric connections, thus anA4LP/802.11 combination would always performsomewhat worse than combination such as OLSR/802.11, because we can take advantage of the exis-tence of asymmetric links only if they are supportedthroughout the stack. Thus, the only reasonablechoices are to use either all symmetric protocols orall asymmetric-link aware ones in the full stack.

A possible study would involve the comparisonof between asymmetric stacks, by substituting for

0 41

5

90

40

150

55

80

80

50

(0,0)2

3

7295

110

100120

95

72

80

(200,60

(140,20)

(110,60)

(40,120)

(40,0)

(0,60)

72

Fig. 5. (a) The physical topology of the network, where node 0 and 4 arposition in the (x,y) format and the transmission range (underlined). Th(b) The logical topology of a network.

AsyMAC the protocols described by [6,15]. In Sec-tion 4.3, we have implemented the core decisionalgorithms of these protocols for a comparison ofthe classification accuracy. However, there is nopublicly available NS-2 implementation of theseprotocols, and a fully functional implementationof these protocols is beyond the scope of this paper.

First, we analyze the benefits of algorithms ableto take advantage of asymmetric links in the main-taining the connectivity of a network. Through thestudy of a specific scenario, we show that a protocolstack composed by AsyMAC and the A4LP routingprotocol is able to maintain connectivity where thestandard IEEE 802.11 MAC protocol coupled withAODV or OLSR loose connectivity.

Second, we perform a simulation study in whichwe measure the performance of the A4LP/AsyMACstack against the AODV/802.11 and OLSR/802.11stacks in a series of randomized mobile ad hoc net-work scenarios with realistic traffic source patterns.

Finally, we compare AsyMAC against two previ-ously proposed asymmetric MAC protocols interms of the accuracy of the hidden node classi-fication.

4.1. A connectivity scenario

In this section, we briefly discuss an examplewhen A4LP/AsyMAC uses asymmetric links toroute packets from each pair of nodes while bothAODV/802.11 and OLSR/802.11 fail to route pack-ets. The connectivity scenario is given in Fig. 5. Theinitial position of nodes is depicted in the graph (a),which shows also the transmission range and thedistance between the nodes. The graph (b) is a log-ical view of the above scenario. The nodes do notmove during the simulation. The forward andreverse routes are found and established by A4LP,

)

0 41

5

2

3

e exchanging packets. The numbers next to the nodes indicate thee numbers on the links represent the distance between the nodes.

Table 1The default values and the range of the parameters for oursimulation studies

Field Value Range

Simulation area 500 · 500 (m2)Number of nodes 8(C1), 16(C2), 24(C3), 32(C4) 30–110Ratio of nodes C1:C2:C3:C4 = 1:2:3:4Transmission ranges 200(C1),150(C2),

100(C3),50(C4) (m)Speed 1 (m/s) 1–10

(m/s)Pause time 15 (s)Simulation time 300 (s)Setup time 20 (s)Switching interval 10 (s)Number of CBR

sources10 4–40

CBR packet size 64 (bytes)CBR sending rate 512 (bps)CBR duration 5 (s)


and MAC layer acknowledgements are assured byAsyMAC. For instance, node 5 is a proxy node thatforwards CTS and ACK packets for a unidirec-tional transmission from node 1 to node 4 atMAC layer. In this scenario, the two far-most nodes0 and 4 are exchanging packets. The packets aresuccessfully delivered and acknowledged by A4LP/AsyMAC, while all packets are lost by AODV/802.11 or OLSR/802.11 during the transmission.

4.2. A study of alternative protocol stacks in a

mobile ad hoc network

The previous scenario illustrates the case whenthe A4LP/AsyMAC protocol maintained connectiv-ity, while the AODV/802.11 and OLSR/802.11stacks did not. However, these extreme cases mightbe relatively rare. In the following, we comparethese protocol stacks in a series of simulationsinvolving an ad hoc network with mobile nodes ina more realistic setup. To describe the movementof nodes in the system, we use the ‘‘random way-point’’ model [4]. Each node randomly picks a des-tination on the map, moves to the destination at aconstant speed, and then pauses for certain time,the pause time. After the pause time, it continuesthe movement following the same pattern. Thenodes are classified into four classes C1, C2, C3and C4 with different transmission ranges.

The traffic patterns are generated by constant bit

rate (CBR) sources sending UDP packets. EachCBR source resides at one node and generates pack-ets for another node. Each CBR source is active fora time interval called CBR duration. Our simulationallows a setup time to allow nodes gather certainrouting information before generating any traffic.After the setup time, the simulation time is dividedinto equal time slices, called switching intervals.During each switching interval, we generate CBRsources for different pairs of senders and receivers.Table 1 illustrates the default settings and the rangeof the parameters for our simulation experiments.

To construct 95% confidence intervals, eachexperiment was repeated 20 times for a pair of sce-nario and traffic pattern, the two elements affectingthe results of a performance study. This involves 200individual runs for the each of the three studies. Theaverage simulation time for a single experiment wasabout 3 h, for a total of 1800 h of computer time. Byobserving the evolution of the average values andthe calculated confidence intervals after 5, 10 and20 repetitions, we notice that at 20 repetitions the

values reach quiescence, and future repetitionswould provide only insignificant changes on theoverall shape of the graphs.

We are concerned with the impact of node mobil-ity, network load, and network density upon powerconsumption, packet loss ratio, and latency. For eachrandomly generated scenario and traffic patterns, werun simulation experiments covering AODV withIEEE 802.11, OLSR with IEEE 802.11, A4LP using3-limited forwarding with distance metric (A4LP-M3-F1/AsyMAC) with AsyMAC, and A4LP using3-limited forwarding with the metric proposed in[19] (A4LP-M3-F2) with AsyMAC.

4.2.1. The influence of network load

The effect of the network load upon the packetloss ratio for two standard protocol stacksAODV/IEEE 802.11, OLSR/IEEE 802.11 and forA4LP-M3-F1/AsyMAC and A4LP-M3-F2/Asy-MAC is summarized by the graphs in Fig. 6. Theratio of the packets lost by AODV/802.11 is roughlytwice the rate of the packets lost by the other proto-cols. The major reason is that flooding, an inefficientbroadcast solution, is used in AODV/802.11 forfinding a route. Among the other protocols,A4LP-M3-F2/AsyMAC performs the best, followedby OLSR/802.11, which delivers more packets thanA4LP-M3-F1/AsyMAC for similar scenarios andtraffic patterns. OLSR/802.11 is able to deliverpackets only via symmetric links, thus packets aredropped if at least one asymmetric link is on thecritical path; however, A4LP/AsyMAC is able to

1 2 3 4 5 6 7 8 9 1010

15

20

25

30

35

40

45

50

55

60

Network Load (Kbps)

Pack

et L

oss

Rat

io (%

)

AODV / IEEE 802.11OLSR / IEEE 802.11A4LP—M3—F1 / AsyMACA4LP—M3—F2 / AsyMAC

Fig. 6. Packet loss ratio versus network load. The ratio ofpackets lost by AODV/802.11 is roughly twice the ratio ofpackets lost by the other protocols. Among the other protocols,A4LP-M3-F2/AsyMAC performs the best, followed by OLSR/802.11, which delivers more packets than A4LP-M3-F1/AsyMACfor similar scenarios and traffic patterns.

1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

Network Load (Kbps)

Aver

age

Late

ncy

(ms)

AODV / IEEE 802.11OLSR / IEEE 802.11A4LP—M3—F1 / AsyMACA4L—PM3—F2 / AsyMAC

Fig. 7. Average latency versus network load. The average latencyof AODV/802.11 is much higher than the other protocols.Among the other protocols, OLSR/802.11 has the shortestlatency.


deliver those packets. Our experiment also showsthe metric we proposed in [19] (A4LP-M3-F2), acombined metric with distance, power level andclass information, provides better performance thanthe distance only metric (A4LP-M3-F1) in heteroge-neous mobile ad hoc networks.

In our study, the measured values have relativelylarge confidence intervals, and most of these confi-dence intervals overlap. This means that we donot have 95% confidence that for any particularexperimental instance the given protocol will per-form better than the other protocol. Indeed, if thereare no (or very few) asymmetric links, the symmetricprotocols will likely outperform the asymmetricones, due to the higher overhead of the asymmetricprotocol. Unfortunately, the range of the measur-able values for metrics such as packet loss is verywide – in some scenarios there might be no packetloss, in other ones, many of packets are lost. Thisvariability is reflected in relatively large confidenceintervals. We believe that often when the averagevalue of packet loss is lower for one of the proto-cols, the protocol will perform in average better thanthe other ones.

The effect of the network load upon the averagelatency for two standard protocol stacks AODV/IEEE 802.11, OLSR/IEEE 802.11 and for A4LP-M3-F1/AsyMAC and A4LP-M3-F2/AsyMAC issummarized by the graphs in Fig. 7. The averagelatency of AODV/802.11 is much higher than that

of the other protocols. AODV is a reactive protocolwhich finds routes only when needed. A4LP is ahybrid protocol, routes to non-neighbors are stilldiscovered when needed, however, routes to certainIn-, Out-, and In/Out-bound neighbors are main-tained proactively in a routing table; this fact con-tributes to the reduction of the average packetdelivery latency.

OLSR/802.11 has the lowest average packetdelivery latency, followed by A4LP-M3-F2/Asy-MAC, and A4LP-M3-F1/AsyMAC. Note, however,that the average packet delivery latency is basedonly on the delivered packets. OLSR/802.11 dropsmore packets than A4LP-M3-F2/AsyMAC; theseare the packets which require a protocol able to dealwith asymmetric links. The packets that could bedelivered by A4LP-M3-F2/AsyMAC but not byOLSR/802.11 generally have higher latency, andthis could explain why the average packet deliverylatency of A4LP-M3-F2/AsyMAC is higher thanthat of OLSR/802.11.

4.2.2. The influence of network mobility

The average packet loss ratio versus node mobil-ity is summarized in Fig. 8. With the networkmobility increasing, the performances of A4LP-M3-F2/AsyMAC and OLSR/802.11 are degraded,while the performance of AODV/802.11 fluctuatesbetween 35% and 45%. AODV/802.11 performsthe worst in case of ad hoc networks with lowmobility, but it outperforms the other protocolsfor highly mobile ad hoc networks. The reason for

1 2 3 4 5 6 7 8 9 1010

20

30

40

50

60

70

Node Mobility (m/s)

Pack

et L

oss

Rat

io (%

)


Fig. 8. Packet loss ratio versus node mobility. With the networkmobility increasing, the performances of A4LP-M3-F2 andOLSR/802.11 are degraded while the performance of AODV/802.11 fluctuates between 35% and 45%. AODV/802.11 performsthe worst in case of ad hoc networks with low mobility, but itoutperforms the other protocols for highly mobile ad hocnetworks.

30 40 50 60 70 80 90 100 11010

20

30

40

50

60

70

80

90

Number of Nodes

Pack

et L

oss

Rat

io (%

)


Fig. 10. Packet loss ratio versus number of nodes. A4LP-M3-F2/AsyMAC delivers most packets, followed by OLSR/802.11,A4LP-M3-F1/AsyMAC and AODV/802.11 for similar scenariosand traffic patterns. The packet loss ratio decreases when thenumber of nodes increases.


this is that for ad hoc networks with relatively highmobility, cached routes and neighbor informationbecomes stale rapidly, which degrades the perfor-mance of proactive (OLSR) or hybrid (A4LP) pro-tocols but not reactive (AODV) protocols.However, A4LP-M3-F2/AsyMAC always outper-forms OLSR/802.11 and A4LP-M3-F1/AsyMACat any network mobility in terms of packet lossratio.

1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500

Node Mobility (m/s)

Aver

age

Late

ncy

(ms)


Fig. 9. Average latency versus node mobility. The averagelatency of AODV/802.11 is much higher than the other protocolsthat perform similarly.

Fig. 9 presents average packet delivery latencyversus network mobility. AODV, which is an on-demand protocol, shows about the same, relativelylong, latency irrespective of the mobility of thenodes. For A4LP/AsyMAC and OLSR/802.11the latency is increasing with the mobility, as theprotocols need additional overhead to keep theirtopology information up-to-date. At the mobilityof about 10 m/s, AODV/802.11, A4LP-M3-F2/Asy-MAC and A4LP-M3-F1/AsyMAC show about the

30 40 50 60 70 80 90 100 1100

500

1000

1500

2000

2500

Number of Nodes

Ave

rage

Lat

ency

(m

s)


Fig. 11. Average latency versus number of nodes. The averagelatency of AODV/802.11 is much higher than the other protocols.The packet latency tends to decrease as the number of nodesincreases for A4LP/AsyMAC and OLSR/802.11.


same latency. In these tests, OLSR/802.11 outper-forms A4LP/AsyMAC because the amount oftopology data it needs to maintain is lower, beingrestricted to the symmetric links only. This latencyadvantage comes at the cost of ignoring asymmetriclinks and therefore, potentially disconnecting nodeswhich would maintain connectivity with the A4LP/AsyMAC solution.

4.2.3. The influence of the number of nodes

In the following set of experiments, we vary thenumber of nodes moving in the measurement area.As the nodes have a limited range, when the numberof nodes is too low, some nodes might looseconnectivity.

Fig. 10 illustrates the packet loss ratio versus thenumber of nodes. For similar scenarios and traffic

40 50 60 70 80 90 100 110 1200

2

4

6

8

10

12

14

16

Number of nodes

Mis

clas

sifie

d no

des

/ tra

nsm

issi

on

Protocol AProtocol BAsyMAC

40 50 60 70 800

1

2

3

4

5

6

7

8

9

Number of

Inco

rrec

t sile

ncin

g ch

oice

s / t

rans

mis

sion


Fig. 12. (a) The average misclassified nodes/transmission as a functmisclassify nodes in a static network. (b) The average missed nodes/transnumber of nodes. (c) The average number of incorrect silencing decisio

patterns, A4LP-M3-F2/AsyMAC delivers mostpackets, followed by OLSR/802.11, A4LP-M3-F1/AsyMAC, and AODV/802.11. As the number ofnodes in the network increases, the network connec-tivity increases as well, thus the packet loss ratiodecreases. Fig. 10 shows that the packet loss ratiodecreases from roughly 40% to about 10% as thenumber of nodes increases from 30 to 110.

Fig. 11 shows the average packet delivery latencyversus the number of nodes. The average latency ofAODV/802.11 is much higher than the other proto-cols. For A4LP/AsyMAC and OLSR/802.11 thepacket latency tends to decrease as the number ofnodes increases. As the number of nodes in the net-work increases, more neighbors and routes arefound during the neighbor information exchangeprocess, thus the packet delivery latency decreases.

40 50 60 70 80 90 100 110 1200

0.5

1

1.5

2

2.5

Number of nodes

Mis

sed

node

s / t

rans

mis

sion


90 100 110 120 nodes

ion of the number of nodes. The AsyMAC protocol does notmission for protocols A, B, and our approach, as a function of thens per transmission for protocols A, B, and for our approach.


4.3. The accuracy of hidden node classification

A node is misclassified as hidden if it is silencedby the algorithm while it should not be silenced.Misclassification reducing bandwidth utilizationbecause it leads to unnecessary silencing of nodeswhich could have been transmitting. A node ismissed by the algorithm if it was not silencedalthough it should have been. Missed nodes leadto collisions. The more accurate is a protocol inclassifying the nodes, the better the bandwidth utili-zation. A useful measure of the global performanceof an algorithm is the number of incorrect silencingdecisions per transmission – defined as the sum ofmisclassified and missed nodes.

We compare the accuracy of the classification ofour proposed AsyMAC protocol with the accuracyof two well known protocols which are performingthe same classification [6,15]. As a note, the basicIEEE 802.11 protocol does not perform any classifi-cation of nodes. The simulation environment is anarea of 500 · 500 m. We populate our environmentwith a heterogeneous collection of nodes belongingto the four main classes of wireless nodes C1, C2,C3, and C4 (see [12,19]). The transmission rangesare normally distributed random variables withthe mean 100, 75, 50, and 25 m, respectively andthe standard deviations for each class is 5 m. Thesimulation scenarios are created using a set of 40–120 nodes including an equal number of nodes foreach class, uniformly distributed in the area. Foreach generated scenario, we repeat the experiment1000 times. The displacement of nodes are distrib-uted around an initial position and the standarddeviation is 20% of its transmission range.

The results of the simulation are shown inFig. 12. The graph (a) shows the number of misclas-sified nodes per transmission. The AsyMAC algo-rithm does not misclassify nodes in a staticnetwork, because in the process of three-party proxyset formation, the nodes whose transmission rangedoes not reach the current node are filtered out.However, misclassified nodes can appear with theAsyMAC protocol if the nodes are highly mobileand the current configuration does not reflect theone detected when the three-party proxy set wasestablished. The graph (b) shows the missed nodesper transmission. Here the AsyMAC protocol per-forms worse than the other two protocols consid-ered, as it is considering only the three-partyproxy sets, and ignores possible higher order proxysets. However, the number of missed nodes is very

small for all the three protocols. Graph (c) showsthe number of incorrect silencing decisions pertransmission. Here, the AsyMAC protocol emergeswith the lowest number of incorrect decisions, as itsbetter performance at misclassification compensatesfor the lower performance in regards to missednodes.

5. Conclusions

In this paper, we argue that asymmetry of thetransmission ranges in wireless networks is a realityand should be treated as such. This asymmetrymakes reliable communication more difficult andcomplicates medium access control, as well as net-work layer protocols.

The models of traditional multiple access net-works assume that all nodes share a single commu-nication channel and have access to the feedback(success, idle slot, collision) from any transmission.In this case, splitting algorithms allow sharing of thecommunication channel in a cooperative environ-ment with reasonable efficiency and fairness. Thisis no longer the case for wireless networks withasymmetric or unidirectional links, where the senderand the receiver do not share the feedback channeland hidden nodes may interfere with a transmission.

In case of networks with asymmetric links, hid-den nodes may be out of the reach of both the sen-der and the receiver, but their transmissions mayinterfere with the reception of a packet by theintended destination. The problem of hidden nodesis further complicated because the feedback fromthe receiver in an RTS/CTS exchange may have topass through several relay stations before reachingall the nodes expected to be silent.

Some of the solutions proposed in the literaturereduce the probability of a collision by requiring alarger than necessary set of nodes to be silent. Inturn, this has negative effects upon the communica-tion latency and the overall network throughput.We propose a MAC layer protocol, AsyMAC,which reduces the number of nodes that have tobe silent but, as all the other schemes proposed,may miss some of the nodes which should have beenclassified as ‘‘hidden’’.

IEEE 802.11 assumes symmetric links betweeneach pair of nodes while AsyMAC does not. Fortraffic over asymmetric links, AsyMAC relies on aproxy node in the three-party proxy set to relayacknowledgements back to the sender so that thereliability is assured. Our MAC protocol reduces


average packet loss ratio and average packet deliv-ery latency as asymmetric links are comprehensivelyutilized which dominate routing in heterogeneousad hoc networks.

We conducted a simulation experiment using theNS-2 simulator and compared the performance ofAODV/IEEE 802.11, OLSR/IEEE 802.11, A4LPusing 3-limited forwarding with distance metric(A4LP-M3-F1) with AsyMAC, and A4LP using 3-limited forwarding with the metric proposed in[19] (A4LP-M3-F2) with AsyMAC. It is reportedthat A4LP/AsyMAC performs much better thanAODV/IEEE 802.11 in terms of average packetloss ratio and average packet delivery latency, inrelatively stable ad hoc networks. A4LP-M3-F2with AsyMAC incurs a lower average packet lossratio compared to OLSR/IEEE 802.11. Our simula-tion results also indicate that the fitness functionproposed in [19] is better than the traditional dis-tance function used in heterogeneous ad hocnetworks.

Our future work is dedicated to remove thedependency of AsyMAC from A4LP, and providetransparent interface to routing protocols so thatit could be the underlying MAC protocol for anyrouting protocol in heterogeneous wireless ad hocnetworks with asymmetric links.

Acknowledgments

The authors thank anonymous reviewers for theirinsightful comments which have helped improve thequality of this paper. The research reported in thispaper was partially supported by National ScienceFoundation grants ACI0296035 and EIA0296179.

References

[1] L. Bao, J. Garcia-Luna-Aceves, Channel access scheduling inad hoc networks with unidirectional links, in: Proceedings ofWorkshop on Discrete Algorithms and Methods for MobileComputing and Communications (DIALM), 2001, pp. 9–18.

[2] V. Bharghavan, A. Demers, S. Shenker, L. Zhang,MACAW: A media access protocol for wireless LAN’s, in:Proceedings of ACM SIGCOMM’94, 1994, pp. 221–225.

[3] L. Breslau, D. Estrin, K. Fall, S. Floyd, J. Heidemann, A.Helmy, P. Huang, S. McCanne, K. Varadhan, Y. Xu, H. Yu,Advances in network simulation, IEEE Computer 33 (5)(2000) 59–67.

[4] J. Broch, D.A. Maltz, D.B. Johnson, Y. Hu, J. Jetcheva, Aperformance comparison of multi-hop wireless ad hocnetwork routing protocols, in: Proceedings of Mobile Com-puting and Networking, 1998, pp. 85–97.

[5] T. Clausen, P. Jacquet, A. Laouiti, P. Muhlethaler, A.Qayyum, L. Viennot, Optimized link state routing protocolfor ad hoc networks, in: Proceedings of IEEE INMIC,December, 2001, pp. 62–68.

[6] T. Fujii, M. Takahashi, M. Bandai, T. Udagawa, I. Sasase,An efficient MAC protocol in wireless ad-hoc networks withheterogeneous power nodes, in: The 5th InternationalSymposium on Wireless Personal Multimedia Communica-tions (WPMC’2002), Hawaii, vol. 2, 2002, pp. 776–780.

[7] C. Fullmer, J.J. Garcia-Luna-Aceves, Floor acquisitionmultiple access (FAMA) for packet-radio networks, in:Proceedings of ACM SIGCOMM’95, 1995, pp. 262–273.

[8] IEEE std 802.11b-1999. Part 11: Wireless LAN mediumaccess control (MAC) and physical layer (PHY) specifica-tions, August 1999.

[9] P. Karn, MACA – a new channel access method for packetradio, in: Proceedings of the 9th ARRL Computer Net-working Conference, 1990, pp. 134–140.

[10] L. Kleinrock, F. Tobagi, Packet switching in radio channels:Part I – carrier sense multiple-access modes and theirthroughput-delay characteristics, IEEE Transactions onCommunications COM-23 (12) (1975) 1400–1416.

[11] K. Xu, M. Gerla, S. Bae, Effectiveness of RTS/CTShandshake in IEEE 802.11 based ad hoc networks, AdHoc Networks Journal 1 (1) (2003) 107–123.

[12] D.C. Marinescu, G.M. Marinescu, Y. Ji, L. Boloni, H.Siegel, Ad hoc grids: communication and computing in apower constrained environment, in: Proceedings of theWorkshop on Energy-Efficient Wireless Communicationsand Networks (EWCN), 2003, pp. 113–122.

[13] CMU Monarch extensions to ns. URL: http://www.monarch.cs.cmu.edu.

[14] C. Perkins, E. Royer, Ad hoc on-demand distance vectorrouting, in: Proceedings of the 2nd IEEE Workshop onMobile Computing Systems and Applications, 1999, pp. 99–100.

[15] N. Poojary, S.V. Krishnamurthy, S. Dao, Medium accesscontrol in a network of ad hoc mobile nodes with hetero-geneous power capabilities, in: Proceedings of IEEE ICC2001, vol. 3, 2001, pp. 872–877.

[16] V. Ramasubramanian, R. Chandra, D. Mosse, Providing abidirectional abstraction for unidirectional ad-hoc networks,in: Proceedings of INFOCOM 2002, vol. 3, 2002, pp. 1258–1267.

[17] V. Ramasubramanian, D. Mosse, Statistical analysis ofconnectivity in unidirectional ad hoc networks, in: Proceed-ings of the International Workshop on Ad Hoc Networking2002, Vancouver, 2002, pp. 109–115.

[18] VINT project, The ucb/lbnl/vint network simulator-ns(version 2), URL: http://www.isi.edu/nsnam/ns.

[19] G. Wang, Y. Ji, D.C. Marinescu, D. Turgut, A routingprotocol for power constrained networks with asymmetriclinks, in: Proceedings of the ACM Workshop on Perfor-mance Evaluation of Wireless Ad Hoc, Sensor, and Ubiq-uitous Networks (PE-WASUN), 2004, pp. 69–76.

[20] DARPA Advanced Technology Office – FCS Communica-tions program, URL: http://www.darpa.mil/ato/programs/fcs_comm.htm.

[21] Joint Tactical Radio System (JTRS), URL: http://jtrs.army.mil.

http://www.monarch.cs.cmu.edu

http://www.monarch.cs.cmu.edu

http://www.isi.edu/nsnam/ns

http://www.darpa.mil/ato/programs/fcs_comm.htm

http://www.darpa.mil/ato/programs/fcs_comm.htm

http://jtrs.army.mil

http://jtrs.army.mil


Guoqiang Wang received a B.S. degreefrom Southeast University, Nanjing,China, in 2001. He is currently a Ph.D.Candidate in the School of ElectricalEngineering and Computer Science atthe University of Central Florida,Orlando.

His research areas include ad hocrouting, grid computing, and parallelsimulation.

Damla Turgut is an Assistant Professorwith the School of Electrical Engineeringand Computer Science at University ofCentral Florida.

She received her B.S., M.S., and Ph.D.degrees from the Computer Science andEngineering Department of Universityof Texas at Arlington in 1994, 1996, and2002, respectively. She has been includedin the WHO’s WHO among students inAmerican Universities and Colleges in

2002. She has been awarded outstanding research award and hasbeen recipient of the Texas Telecommunication Engineering
Consortium (TxTEC) fellowship. She is a member of IEEE,member of the ACM, and the Upsilon Pi Epsilon honorarysociety. Her research interests include wireless networking,mobile computing, distributed systems, agents, and databases.
Ladislau Boloni is an Assistant Professorwith the School of Electrical Engineeringand Computer Science at University ofCentral Florida.

He received a Ph.D. degree from theComputer Sciences Department of Pur-due University in May 2000. He receiveda Master of Science degree from theComputer Sciences department of Pur-due University in 1999 and DiplomaEngineer degree in Computer Engineer-

ing with Honors from the Technical University of Cluj-Napoca,Romania in 1993. He received a fellowship from the Computer
and Automation Research Institute of the Hungarian Academyof Sciences for the 1994–95 academic year. He is a senior member
of IEEE, member of the ACM, AAAI and the Upsilon Pi Epsilonhonorary society. His research interests include autonomousagents, grid computing and wireless networking.

Yongchang Ji received the M.S. andPh.D. degrees in computer science fromUniversity of Science and Technology ofChina, Heifei, in 1996 and 1998, respec-tively. He was a Postdoctoral Researcherof computer sciences at Purdue Univer-sity, West Lafayette, IN. He is currentlya Postdoctoral Research Associate ofcomputer science at University of Cen-tral Florida, Orlando. He has publishedmore than 30 papers in professional

journals and referred conference proceedings. His main researchinterests include high-performance computing, Grid computing,
computational biology, parallel and distributed architecture,model, algorithm, and scalability.
Dan C. Marinescu joined the ComputerScience Department at University ofCentral Florida in August 2001 as Pro-fessor of Computer Science. He has beenan Associate and then Full Professor ofComputer Science at Purdue University,in West Lafayette, Indiana, since 1984.He is the Scientific Director of theI2.Lab, an organization supportinginterdisciplinary research in computerand information sciences (http://

I2lab.ucf.edu) at UCF.He is conducting research in parallel and distributed systems,

computational biology, quantum computing, parallel and dis-tributed computing, and has published more than 180 papers injournals and referred conference proceedings in these areas. He isthe author of ‘‘Internet-Based Workflow Management: Towardsa Semantic Web’’, published by Wiley in 2002 and has co-edited‘‘Process Coordination and Ubiquitous Computing’’. The book‘‘Approaching Quantum Computing’’, co-authored with Gabri-ela M. Marinescu was published in September 2004 by Prentice-Hall.

http://I2lab.ucf.edu

http://I2lab.ucf.edu

a mac layer protocol for wireless networks with asymmetric...

Documents