channel assignment strategies

IEEE Communications Magazine • November 200786 0163-6804/07/$20.00 © 2007 IEEE

INTRODUCTION

The next generation of wireless communicationsystems are envisaged to provide high-speed,high-bandwidth ubiquitous connectivity to endusers through a converged network of differenttechnologies, such as third- and fourth-genera-tion (3G/4G) mobile cellular systems, IEEE802.11 (WiFi) based wireless local area networks

(WLANs), and emerging broadband wirelesstechnologies such as IEEE 802.16 (WiMAX).

One of the key components of such con-verged networks is the wireless mesh network(WMN), which is a fully wireless network thatemploys multihop ad hoc networking techniquesto forward traffic to/from the Internet. Unlikethe mobile ad hoc network (MANET), a meshnetwork uses dedicated nodes (called meshrouters) to build a wireless backbone to providemultihop connectivity between nomadic usersand the Internet gateways [1]. WMNs can pro-vide significant advantages in deploying cost-effi-cient, highly flexible, and reconfigurablebackhaul connectivity over large areas. As high-lighted in [2], WMNs have emerged as a promis-ing candidate for extending the coverage of WiFiislands and providing flexible high-bandwidthwireless backhaul for converged networks.

The wireless backbone, consisting of wirelessmesh routers equipped with one or more radiointerfaces, highly affects the capacity of the meshnetwork. This has a significant impact on the over-all performance of the system, thus generatingextensive research in order to tackle the specificchallenges of the WMN. (See [3] for a survey).

Current state-of-the-art mesh networks, whichuse off-the-shelf 802.11-based network cards, aretypically configured to operate on a single chan-nel using a single radio. This configurationadversely affects the capacity of the mesh due tointerference from adjacent nodes in the network,as identified in [3]. Various schemes have beenproposed to address this capacity problem, suchas modified medium access control (MAC) pro-tocols adapted to WMNs [4], the use of channelswitching on a single radio [5, 6], and directionalantennas [3]. While directional antennas andmodified MAC protocols make the practicaldeployment of such solutions infeasible on awide scale, the main issue in using multiplechannels with a single radio is that dynamicchannel switching requires tight time synchro-nization between the nodes.

ABSTRACT

Next-generation wireless mobile communica-tions will be driven by converged networks thatintegrate disparate technologies and services.The wireless mesh network is envisaged to beone of the key components in the converged net-works of the future, providing flexible high-bandwidth wireless backhaul over largegeographical areas. While single radio meshnodes operating on a single channel suffer fromcapacity constraints, equipping mesh routerswith multiple radios using multiple nonoverlap-ping channels can significantly alleviate thecapacity problem and increase the aggregatebandwidth available to the network. However,the assignment of channels to the radio inter-faces poses significant challenges. The goal ofchannel assignment algorithms in multiradiomesh networks is to minimize interference whileimproving the aggregate network capacity andmaintaining the connectivity of the network. Inthis article we examine the unique constraints ofchannel assignment in wireless mesh networksand identify the key factors governing assign-ment schemes, with particular reference to inter-ference, traffic patterns, and multipathconnectivity. After presenting a taxonomy ofexisting channel assignment algorithms forWMNs, we describe a new channel assignmentscheme called MesTiC, which incorporates themesh traffic pattern together with connectivityissues in order to minimize interference in multi-radio mesh networks.

WIRELESS MESH NETWORKS

Habiba Skalli, IMT Lucca Institute for Advanced Studies

Samik Ghosh and Sajal K. Das, The University of Texas at Arlington

Luciano Lenzini, University of Pisa

Marco Conti, Italian National Research Council (CNR)

Channel Assignment Strategies forMultiradio Wireless Mesh Networks:Issues and Solutions

This work was done whilethe first author was visit-ing Center for Research inWireless Mobility andNetworking (CReW-MaN), The University ofTexas at Arlington.

SKALLI LAYOUT 10/18/07 2:33 PM Page 86

IEEE Communications Magazine • November 2007 87

Equipping each node with multiple radios isemerging as a promising approach to improvingthe capacity of WMNs. First, the IEEE 802.11b/gand IEEE 802.11a standards provide 3 and 12nonoverlapping (frequency) channels, respective-ly, which can be used simultaneously within aneighborhood (by assigning nonoverlappingchannels to radios). This then leads to efficientspectrum utilization and increases the actualbandwidth available to the network. Second, theavailability of cheap off-the-shelf commodityhardware also makes multiradio solutions eco-nomically attractive. Finally, the spatio-temporaldiversity of radios operating on different fre-quencies with different sensing-to-hearing ranges,bandwidth, and fading characteristics can beleveraged to improve the capacity of the network.

Although multiradio mesh nodes have thepotential to significantly improve the perfor-mance of mesh networks, efficient channelassignment is a key issue in guaranteeing net-work connectivity while still mitigating theadverse effects of interference from the limitednumber of channels available to the network. AWMN node needs to share a common channelwith each of its communication-range neighborswith which it wishes to set up a virtual link1 orconnectivity. However, to reduce interference, anode should minimize the number of neighborswith which it will share a common channel.There is thus a trade-off between maximizingconnectivity and minimizing interference, asillustrated in Fig. 1. In this figure the maximumconnectivity that can be achieved is shown inFig. 1a. In both Fig. 1b and 1c, there are fourchannels available but only three can be assignedto the radios in Fig. 1b that maximize connectivi-ty. On the other hand, all four can be exploitedsimultaneously if the only goal is to minimizeinterference, as shown in Fig. 1c.

The above issues provided us with the moti-vation to undertake a systematic study of differ-ent channel assignment schemes for WMNs, andexamine their relative strengths and weaknesses,as outlined later. Based on our research, weidentify the key characteristics pertinent to chan-nel assignment in WMNs and present an innova-tive scheme called MesTiC for channelassignment in multiradio mesh networks. MesTiCis a static centralized channel assignment schemebased on a ranking function that takes into

account traffic, number of hops from the gate-way, and number of radio interfaces per node.We summarize the performance of MesTiC interms of network throughput on a comparativesimulation platform. We compare the differentCA schemes. We then conclude the article.

A TAXONOMY OF CHANNELASSIGNMENT SCHEMES FOR WMNSChannel assignment (CA) in a multiradio WMNenvironment consists of assigning channels to theradio interfaces in order to achieve efficientchannel utilization and minimize interference.The problem of optimally assigning channels inan arbitrary mesh topology has been proven tobe NP-hard based on its mapping to a graph-col-oring problem [7]. Therefore, channel assignmentschemes predominantly employ heuristic tech-niques to assign channels to nodes in the net-work. The performance bottleneck associatedwith channel assignment in WMNs has beenextensively studied in the literature. In this sec-tion we present a taxonomical classification ofvarious CA schemes for mesh networks. Figure 2presents the taxonomy on which the rest of thesection is based. Specifically, the proposed CAschemes can be divided into three main cate-gories — fixed, dynamic, and hybrid — depend-ing on the frequency with which the CA schemeis modified. In a fixed scheme the CA is almostconstant, while in a dynamic scheme it is continu-ously updated to improve performance. A hybridscheme applies a fixed scheme for some inter-faces and a dynamic one for others. In the fol-lowing we analyze these three categories and giveexamples of CA schemes from each category.

FIXED CHANNEL ASSIGNMENT SCHEMESFixed assignment schemes assign channels tointerfaces either permanently or for long timeintervals with respect to the interface switchingtime. Such schemes can be further subdividedinto common channel assignment and varyingchannel assignment.

Common Channel Assignment — This is thesimplest scheme. In CCA [8] the radio interfacesof each node are all assigned the same set ofchannels. The main benefit is that the connectiv-

n Figure 1. Trade-off between connectivity and interference: a) single channel scenario; b) maximum con-nectivity; c) minimum interference.

1 channel1 interface

(a)

4 available channels2 interfaces/node

(b)

4 available channels2 interfaces/node

(c)

1 A virtual link betweentwo nodes is defined as apossible direct communi-cation link between them.

In a fixed scheme

the CA is almost

constant, while in a

dynamic scheme it is

continuously

updated to improve

performance.

A hybrid scheme

applies a fixed

scheme for some

interfaces and a

dynamic one for

others.


IEEE Communications Magazine • November 200788

ity of the network is the same as that of a single-channel approach, while the use of multiplechannels increases network throughput. Howev-er, the gain may be limited in scenarios wherethe number of nonoverlapping channels is muchgreater than the number of network interfacecards (NICs)2 used per node. Thus, although thisscheme presents a simple CA strategy, it fails toaccount for the various factors affecting channelassignment in a WMN.

Varying Channel Assignment — In the VCAscheme, interfaces of different nodes may beassigned different sets of channels [7, 9]. Howev-er, the assignment of channels may lead to net-work partitions and topology changes that mayincrease the length of routes between the meshnodes. Therefore, in this scheme, assignmentneeds to be carried out carefully. Below we pre-sent the VCA approach through two existingalgorithms in this subcategory.

Centralized Channel Assignment — Basedon Hyacinth, a multichannel wireless mesh net-work architecture, a central ized channelassignment algorithm for WMNs (C-HYA) isproposed in [7], where traffic is mainly direct-ed toward gateway nodes (i.e. the traffic isdirected to/from the Internet). Assuming thetraffic load is known, this algorithm assignschannels, thus ensuring the network connectiv-ity and bandwidth limitations of each link. Itfirst estimates the total expected load on eachvirtual link based on the load imposed by eachtraffic flow. Then the channel assignment algo-rithm visits each virtual link in decreasingorder of expected traffic loads and greedilyassigns it a channel. The algorithm starts withan initial estimation of the expected trafficload and iterates over both channel assign-ment and routing until the bandwidth allocat-ed to each virtual link matches its expectedload. While this scheme presents a method forchannel allocation that incorporates connectiv-ity and traffic patterns, the assignment ofchannels on links may cause a ripple effect [7]whereby already assigned l inks have to berevisited, thus increasing the time complexityof the scheme.

A Topology Control Approach — In [9] thenotion of a traffic independent channel assign-ment scheme is proposed to enable an efficientand flexible topology formation, ease of coordi-nation, and to exploit the static nature of meshrouters to update the channel assignment onlarge timescales. A polynomial time greedyheuristic called Connected Low InterferenceChannel Assignment (CLICA) is presented in[9] that computes the priority for each meshnode and assigns channels based on the connec-tivity graph and conflict graph. However, thealgorithm can override the priority of a node toaccount for the lack of flexibility in terms ofchannel assignment and to ensure network con-nectivity. Thus, while this scheme overcomes linkrevisits, it does not incorporate the role of trafficpatterns in channel assignment for WMNs.

DYNAMIC CHANNEL ASSIGNMENT SCHEMESDynamic assignment strategies allow any inter-face to be assigned any channel, and interfacescan frequently switch from one channel to anoth-er. Therefore, when nodes need to communicatewith each other, a coordination mechanism hasto ensure they are on a common channel. Forexample, such mechanisms may require all nodesto periodically visit a predetermined rendezvouschannel [5] to negotiate channels for the nextphase of transmission. In the Slotted SeededChannel Hopping (SSCH) mechanism [6], eachnode switches channels synchronously in a pseu-do-random sequence so that all neighbors meetperiodically in the same channel.

The benefit of dynamic assignment is the abil-ity to switch an interface to any channel, therebyoffering the potential to use many channels withfew interfaces. However, the key challengesinvolve channel switching delays (typically on theorder of milliseconds in commodity 802.11 wire-less cards), and the need for coordination mecha-nisms for channel switching between nodes.

A Distributed Channel Assignment Scheme— A set of dynamic and distributed channelassignment algorithms is proposed in [10, 11],which can react to traffic load changes in orderto improve the aggregate throughput and achieveload balancing. Based on the Hyacinth architec-ture, the algorithm (D-HYA) builds on a span-ning tree network topology in such a way thateach gateway node (the node directly connectedto the wired network) is the root of a spanningtree, and every mesh node belongs to one ofthese trees. The channel assignment problemconsists of:• Neighbor-to-interface binding (i.e., it selects

the interface to communicate with everyneighbor), where the dependence amongthe nodes is eliminated in order to preventripple effects in the network [7]

• Interface-to-channel binding (i.e., it selectsthe channel to assign to every interface),where the goal is to balance the load amongthe nodes and relieve interferenceFinally, channels are dynamically assigned to

the interfaces based on their traffic information.The tree-topology constraint of the schemeposes a potential hindrance in leveraging multi-path routing in mesh networks.

n Figure 2. Taxonomy of channel assignment schemes in wireless mesh net-works.

LLP IACA

CLICAC-HYA

RZV CCM*

* The common channel mechanism as in [5]

SSCH D-HYACCA VCA

Hybrid CA Dynamic CAFixed CA

CA

Coordinationmechanisms

2 Note: In this article weuse NIC and radio inter-face interchangeably.



HYBRID CHANNEL ASSIGNMENT SCHEMES

Hybrid channel assignment strategies combineboth static and dynamic assignment propertiesby applying a fixed assignment for some inter-faces and a dynamic assignment for other inter-faces [8, 12, 13]. Hybrid strategies can be furtherclassified based on whether the fixed interfacesuse a common channel [13] or varying channel[8, 12] approach. The fixed interfaces can beassigned a dedicated control channel [10] or adata and control channel [13], while the otherinterfaces can be switched dynamically amongchannels. Hybrid assignment strategies areattractive because, as with fixed assignment, theyallow for simple coordination algorithms, whilestill retaining the flexibility of dynamic channelassignment.

In the next two subsections we describe twohybrid schemes for CA.

Link Layer Protocols for Interface Assign-ment — In [8, 12] an innovative link layer inter-face assignment algorithm (LLP) is proposedthat categorizes available interfaces into fixedand switchable interfaces. Fixed interfaces areassigned, for long time intervals, specific fixedchannels, which can be different for differentnodes. On the other hand, switchable interfacescan be switched over short timescales amongnon-fixed channels based on the amount of datatraffic. By distributing fixed interfaces of differ-ent nodes on different channels, all channels canbe used, while the switchable interface can beused to maintain connectivity. Two coordinationprotocols based on hash functions and theexchange of Hello packets are proposed in [8] todecide which channels should be assigned to thefixed interface and manage communicationbetween nodes. In [12] the authors propose aCA scheme based on the second coordinationprotocol, but this scheme does not take intoaccount the traffic load in assigning the fixedchannels.

Interference-Aware Channel Assignment— The channel assignment problem in wirelessmesh networks in the presence of interferencefrom collocated wireless networks is addressedin [13]. The authors propose a dynamic central-ized interference-aware algorithm (IACA)aimed at improving the capacity of the WMNbackbone and minimizing interference. Thisalgorithm is based on an extension to the con-flict graph concept described in [9], called themultiradio conflict graph (MCG), where thevertices in the MCG represent edges betweenmesh radios instead of edges between meshrouters. To compensate for the drawbacks of adynamic network topology, the proposed solu-tion assigns one radio on each node to operateon a default common channel throughout thenetwork. This strategy ensures a common net-work connectivity graph [13], provides alternatefallback routes, and avoids flow disruption bytraffic redirection over a default channel. Thisscheme computes interference and bandwidthestimates based on the number of interferingradios, where an interfering radio is a simultane-ously operating radio that is visible to a mesh

router but external to its network. The channelassignment scheme works on a rank-based strat-egy where the rank for every available channelis based on interference and load. The load,however, is considered for external wireless net-works only.

MESTIC: A NEW CHANNELASSIGNMENT SCHEME

As highlighted earlier, the central goal of chan-nel assignment for multiradio mesh networks isto improve the aggregate throughput of the net-work, taking into account the effects of trafficand interference patterns, as well as maintainingtopological connectivity. Based on our observa-tions of the impact of traffic patterns and net-work connectivity on the performance of aWMN, below we propose an innovative schemecalled MesTiC, which stands for mesh-based traf-fic and interference aware channel assignment.It has the following important features:• MesTiC is a fixed, rank-based, polynomial

time greedy algorithm for centralized chan-nel assignment, which visits every nodeonce, thereby mitigating any ripple effect.

• The rank of each node is computed on thebasis of its link traffic characteristics, topo-logical properties, and number of NICs ona node.

• Topological connectivity is ensured by acommon default channel deployed on aseparate radio on each node, which canalso be used for network management.Fixed schemes alleviate the need for channel

switching, especially when switching delays arelarge as is the case with the current 802.11 hard-ware. In addition, MesTiC is rank-based, whichgives the nodes that are expected to carry heavyloads more flexibility in assigning channels.Finally, the use of a common default channelprevents flow disruption.

It should also be mentioned that the pro-posed scheme has been designed for a mesh net-work with a single gateway node, but could easilybe extended to multiple gateways with minormodifications to the basic scheme.

PROPOSED ALGORITHMThe central idea behind MesTiC is to assignchannels to the radios of a mesh node based onranks assigned a priori to the nodes. The rank ofa node, Rank(node), determines its priority inassigning channels to the links emanating fromit. The rank encompasses the dynamics of chan-nel assignment and is computed on the basis ofthree factors:• The aggregate traffic at a node based on the

offered load of the mesh network as com-puted in [7]

• The distance of the node, measured as theminimum number of hops from the gatewaynode

• The number of radio interfaces available ona nodeNote that the gateway node is assigned the

highest rank as it is expected to carry the mosttraffic. The rank for the remaining nodes isgiven by:

The channel

assignment scheme

works on a

rank-based strategy

where the rank for

every available

channel is based on

interference and

load. The load,

however, is

considered for

external wireless

networks only.



(1)

Clearly, the aggregate traffic flowing througha mesh node has an impact on the channelassignment strategy. The rationale behind thisobservation stems from the fact that if a noderelays more traffic, assigning it a channel ofleast interference will increase the networkthroughput. Thus, aggregate traffic in thenumerator in Eq. 1 increases the rank of a nodewith its traffic. In addition, due to the hierarchi-cal nature of a mesh topology, the nodes near-est the gateway should have a higher preference(rank) in channel assignment, as they are more

likely to carry more traffic. At the same time,the number of radios on a node gives flexibilityin channel assignments and should inverselyaffect its priority (i.e., the lower the number ofradios, the higher the priority in channel assign-ment). The aggregate traffic (total traffictraversing a node) is a key factor in computingthe rank of the node. Such measure is subject totemporal variability due to the randomness ofthe wireless channel, routing protocols andapplication layer traffic profiles. We envisagethat the traffic characterizations aggregatedfrom a large number of network flows changeover longer periods of time, whereas MesTiCcan reassign channels based on new traffic char-acteristics.

Once the rank of each node has been com-puted, the algorithm traverses the mesh net-work in decreasing order of Rank(node),assigning channels to the radios as described inFig. 3. In this figure the algorithm starts by cal-culating a fixed rank for every node (I), andthen every node is visited in decreasing order(II). If two nodes have already been assignedat least one common channel, by default thereis a link between these nodes (II.1). If not, forevery possible unassigned link, the one thatcarries the higher traffic is assigned first (II.2)in the following manner: if the node visitedstill has an assigned radio, the least used chan-nel is assigned to one of its free radios and alink is established with its neighbor (II.2.a).Otherwise, if all the visited node’s radios havealready been assigned, the least used channelamong those already assigned to its radios isassigned to the link (II.2.b). For a detaileddescription of the MesTiC algorithm, pleaserefer to [14].

We illustrate the working principle of Mes-TiC by considering a simple example in Fig. 4awhere the input connectivity graph and esti-mated link traffic (estimated traffic between anode and its neighbors) are shown. In additionthe network is configured with three channelsand two interfaces per node. Assuming thatnode b is the gateway node, the rank of theremaining nodes, in decreasing order, is d, a,c. The algorithm starts by visiting node b first,assigning channel C1 to the link between b-a(which carries the highest traffic of 120), andthen moving on to assign channel C2 to link b-d. Now, while assigning a channel to link b-c,it has to choose between C1 and C2. However,as C1 carries more traffic than C2, it assignsC2 to link b-c. Similarly, at node d, it assigns apreviously unassigned channel C3 to link d-c,and as C3 carries less traffic than C2 (90 + 80= 170) or C1 (120), it assigns C3 to link d-a.The algorithm proceeds until all l inks andradios are assigned channels, as shown in Fig.4b.

In this manner MesTiC assigns channels tothe radio interfaces of the nodes in a WMN,while the connectivity of the network is ensuredthrough a separate radio on a default channel.The cost dynamics of 802.11-based hardware andthe availability of 12 nonoverlapping channels inthe IEEE 802.11a standard make a default con-nectivity scheme feasible under current scenariosfor community mesh networks.

Rank nodeAggregate Traffic node

hops( )

( )

min=

ffrom the gateway node

number of radios

( )

* (( )node

n Figure 3. Flow diagram of MesTiC.

II.2.b. Assign the link a leastused channel within those

already assigned to thenode’s radios

II.2.a. Assign to a radio theleast used channel within

the vicinity

II.1. Assign a channel to theincident link, if the node and

one of its neighbors areassigned a channel in

common

II. Visit every node inthe rank order

If the node has anunassigned radio

Else: while the nodehas an unassigned

incident link

I. Order all the nodes according to Formula (1)

Else

II.2. Pick a neighbor withwhom the node has themost traffic in the traffic

matrix



PERFORMANCE STUDY

In this section we study the performance of theproposed channel assignment scheme, MesTiC,in terms of overall throughput on a wirelessmesh network. We present details of the simula-tion platform and results of a comparison withthe traffic-aware centralized scheme based onthe Hyacinth architecture [7], C-HYA.

In order to build a common platform for acomparative study, we developed our simulationon a modified version of ns-2 [15] software,which incorporates support for multiple radiosand configurable routing protocols, such asdynamic source routing (DSR) and ad hoc ondemand distance vector routing (AODV). Thesimulation experiments were performed on a 5 ×5 grid topology3 where each node could poten-tially communicate with four neighbors. With arandomly generated traffic profile, the trafficbetween any source-destination pair is chosen inthe range [0–3] Mb/s. Ns-2 was configured toemulate the traffic profile by running constantbit rate (CBR) UDP flows. The conflict graphwas created based on the interference-to-com-munication ratio set to 2, and the experimentsreported in this article were performed basedon the DSR protocol. As mentioned earlier, thecentralized CA scheme based on C-HYAaccounts for the link traffic matrix in their chan-nel assignment algorithm. Moreover, their simu-lation analysis is based on a similar ns-2-basedplatform with similar settings. Thus, in this arti-cle we report our results based on comparisonswith C-HYA. However, our simulation platformcan easily be extended to incorporate differentrouting and channel assignment schemes formesh networks.

The WMN was simulated on ns-2 with thenumber of radios on each node set to 3, with 12nonoverlapping channels. The simulation wasperformed for 100 s for a given set of traffic pro-files, and ns-2 was configured to report theaggregate throughput obtained in the network.The experiments were conducted on the meshnetwork topology with channel assignments gen-erated by MesTiC, and repeated for the channelassignments generated by C-HYA. Figure 5reports the dynamics of the network in terms ofaggregate throughput. The figure highlights thatthe simulation stabilizes around 40 s from thestart of the simulation run, after which MesTiCreports a sustained higher aggregate throughputfor the mesh network.

Similarly, at the stable region, with MesTiCthere is enough bandwidth for a larger numberof flows in the system, with an average value of14 flows against an average of 9 flows in C-HYA[14]. Our extensive simulation results (notreported due to space constraints) conclude thatMesTiC provides a significant improvement inaggregate throughput over C-HYA for varioustopologies and traffic profiles [14].

Note that although the simulation experi-ments were performed with three radios pernode, MesTiC essentially operates its channelassignment scheme on two radios, with the thirdconfigured on a default channel for connectivity.Thus, even with a lower degree of freedom interms of radio flexibility, MesTiC was able to

improve the overall network performance interms of aggregate throughput.

COMPARISON OF CA SCHEMESThe most important features of the existing CAalgorithms for WMNs are summarized in Table1.

The key issues are connectivity, topology con-trol, interference minimization, and traffic pat-tern. C-HYA is a traffic-aware CA scheme.While its distributed version, D-HYA, alleviatesthe effect of link revisits, stringent restrictionswere imposed on the topology of the mesh net-work, thereby failing to leverage the advantagesof multipath routing in a mesh scenario. Whilethe goal of LLP and CLICA was to minimizeinterference, the effect of traffic patterns oninterference and thus on the CA scheme was nottaken into account. The effect of traffic in IACAwas considered, but only for traffic emanatingfrom external wireless networks. From another

n Figure 4. Example illustrating how MesTiC works: a) connectivity and linktraffic; b) channel assignment with MesTiC.

120

50

80

7090a c

b

Given: 3 channels2 interfaces/node

(a)

d

C1

C3

C2

C3C2a c

b

(b)

d

n Figure 5. Aggregate throughput dynamics of MesTiC vs. C-HYA.

Time (s)

Aggregate throughput dynamics of the WMN

1000

1

Agg

rega

te t

hrou

ghpu

t (M

b/s)

2

3

4

5

6

7

20 30 40 50 60 70 80

Simulation transient interval

90 100

C-HYAMesTiC

3 Although simulationscan be conducted on larg-er networks, we report ona 25-node mesh network,as community mesh net-works are envisaged tocontain typically 25–30mesh routers.



perspective, some algorithms, such as CLICA,considered topology control, which incurs over-heads in the channel assignment algorithm butalleviates the need for an additional interfacetuned to a common channel; while others (e.g.IACA) assume default connectivity using a sepa-rate common channel on a separate radio.

MesTiC is a fixe centralized scheme that takesinto account traffic load information whileassigning channels to radio interfaces. This isimportant because links that need to supporthigher traffic should be given more bandwidth.This means that these links should use a fre-quency channel shared by a fewer number ofnodes. Although C-HYA and D-HYA both con-sider traffic load, they suffer from ripple effectsand topology constraints, respectively. Moreover,MesTiC uses a default radio that creates a fixedtopology and alleviates the need for a mecha-nism to ensure connectivity, as is the case forCLICA.

CONCLUSIONSIn this article we have identified the key chal-lenges associated with assigning channels toradio interfaces in a multiradio WMN. Afterpresenting a taxonomy of existing channelassignment schemes, we describe an innovativealgorithm, MesTiC, which incorporates trafficpatterns and network topology while rankingnodes for channel assignments. Since the prob-lem of defining optimal routes in a mesh net-work is closely linked to channel assignmentschemes, we are currently studying the impactof channel assignment schemes on the perfor-mance of routing protocols for wireless meshnetworks.

REFERENCES[1] R. Bruno, M. Conti, and E. Gregori, “Mesh Networks:

Commodity Multi-hop Ad Hoc Networks,” IEEE Com-mun. Mag., Mar. 2005, pp. 123–31.

[2] R. Karrer, A. Sabharwal and E. Knightly, “EnablingLarge-scale Wireless Broadband: The Case for TAPs,”2nd Wksp. Hot Topics in Wireless, Nov. 2003.

[3] I. Akyildiz, X. Wang, and W. Wang, “Wireless Mesh Net-works: A Survey,” Comp. Networks, vol. 47, no. 47,2005, pp. 445–87.

[4] X. Wang, W. Wang, and M. Nova, “A High Performance Sin-gle-Channel IEEE 802.11 MAC with Distributed TDMA,”Tech. rep., Kiyon, Inc. (submitted for patent application),Oct. 2004.

[5] J. So and N. Vaidya, “Multi-Channel MAC for Ad HocNetworks: Handling Multi-Channel Hidden Terminalsusing a Single Transceiver,” Proc. ACM Mobihoc, 2004,pp. 222–33.

[6] P. Bahl, R. Chandra and J. Dunagan, “SSCH: SlottedSeeded Channel Hopping for Capacity Improvement inIEEE 802.11 Ad-Hoc Wireless Networks,” Proc. ACMMobicom, 2004, pp. 216–30.

[7] A. Raniwala, K. Gopalan, and T. Chiueh, “CentralizedChannel Assignment and Routing Algorithms for Multi-channel Wireless Mesh Networks,” ACM Mobile Comp.and Commun. Rev., Apr. 2004, pp. 50–65.

[8] P. Kyasanur and N. Vaidya, “Routing and InterfaceAssignment in Multi-Channel Multi-Interface WirelessNetworks,” Proc. IEEE Conf. Wireless Commun. andNet. Conf., 2005, pp. 2051–56.

[9] M. Marina and S. R. Das, “A Topology Control Approachfor Utilizing Multiple Channels in Multi-Radio WirelessMesh Networks,” Proc. Broadnets, Oct 2005, pp. 381–90.

[10] A. Raniwala and T. Chiueh, “Evaluation of a WirelessEnterprise Backbone Network Architecture,” Proc. 12thHot-Interconnects, 2004.

[11] A. Raniwala, and T. Chiueh, “Architecture and Algorithmsfor an IEEE 802.11-Based Multi-Channel Wireless MeshNetwork,” Proc. IEEE INFOCOM, Mar 2005, pp. 2223–34.

[12] P. Kyasanur and N. Vaidya, “Routing and Link-layerProtocols for Multi-Channel Multi-Interface Ad HocWireless Networks,” Mobile Comp. and Commun. Rev.,vol. 10, no. 1, Jan. 2006, pp. 31–43.

[13] K. Ramachandran et al., “Interference Aware ChannelAssignment in Multi-Radio Wireless Mesh Networks,”Proc. IEEE INFOCOM, Apr. 2006.

n Table 1. Comparative study of the salient features of channel assignment schemes.

PropertyFIxed CA Hybrid CA Dynamic CA

CCA C-HYA [7] CLICA [9] MesTiC LLP [8] IACA [10] D-HYA [6]

Switchingtime

No switchingrequired




Switching over-head involved

Infrequentswitching

Infrequentswitching

Connectivity Ensured by theCA scheme

Ensured bythe CAscheme

Ensured bythe CAscheme

Ensured bydefault radio

Ensured bychannel switching

Ensured bydefault radio

Ensured by theCA scheme

Ripple effect No Yes No No No No No

Interferencemodel N/A Protocol

modelProtocolmodel

Protocolmodel Protocol model Trace driven Trace driven

Trafficpattern

Notconsidered Considered Not

considered Considered Not consideredConsideredfrom externalradios

Considered

Topologycontrol Fixed Fixed

CA schemedefines thetopology

Fixed Dynamicallychaging Fixed

No, topology isdefined by therouting tree

Controlphilosophy N/A Centralized Centralized Centralized Distributed Centralized Distributed



[14] H. Skalli et al., “Traffic and Interference Aware Chan-nel Assignment for Multiradio Wireless Mesh Net-works,” Tech. rep., Dec. 2006, http://www.imtlucca.it/_documents/publications/publication54-9369-Technical_report_2006.pdf

[15] Network Simulator ns-2, http://www.isi.edu/nsnam/ns/

BIOGRAPHIESHABIBA SKALLI ([email protected]) received her Master(2004) and Bachelor (2003) degrees in computer scienceand engineering from Al Akhawayn University, Ifrane,Morocco. She is now a Ph.D. student at IMT — Lucca Insti-tute (Italy) under the supervision of Prof. Luciano Lenzini(University of Pisa, Italy) and Dr. Marco Conti (IIT/CNR Pisa,Italy). During her Ph.D. she spent a visiting period at theCenter for Research in Wireless Mobility and Networking(CReWMaN) at the University of Texas at Arlington and didresearch under Prof. Sajal Das. Her current research inter-ests are in channel assignment and routing algorithms forwireless mesh networks.

SAMIK GHOSH ([email protected]) received a Bachelor ofTechnology (Honors) degree in computer science and engi-neering from Haldia Institute of Technology, India, in 2001.After working in the telecommunication software industryfor a year, he began his graduate studies at CreWMaN,Department of Computer Science and Engineering, theUniversity of Texas at Arlingto, where he is currently a sec-ond year doctoral student. His research interests lie inmobile communications, wireless mesh networks, anddeveloping mobile applications for handheld devices. Healso works in modeling and simulation of complex systemsin life sciences.

SAJAL K. DAS ([email protected]) is a Distinguished ScholarProfessor of Computer Science and Engineering and alsothe founding director of CReWMaN at the University ofTexas at Arlington. His current research interests includesensor networks, smart environments, resource and mobili-ty management in wireless networks, mobile and pervasivecomputing, wireless mesh networks, grid computing, bio-logical networking, applied graph theory, and game theo-ry. He coauthored the book Smart Environments:Technology, Protocols, and Applications (Wiley, 2005). Hehas published over 400 research papers in international

conferences and journals, and holds five U.S. patents. Hereceived Best Paper Awards at IEEE PerCom ’06, ACMMobiCom ’99, ICOIN ’02, ACM MSwiM ’00, and ACM/IEEEPADS ’97. He is also a recipient of UTA Outstanding FacultyResearch Award in Computer Science (2001 and 2003),College of Engineering Research Excellence Award (2003),University Award for Distinguished record of Research(2005), and UTA Academy of Distinguished Scholars Award(2006). He serves as Editor-in-Chief of Pervasive and MobileComputing, and Associate Editor of IEEE Transactions onMobile Computing, ACM/Springer Wireless Networks, andIEEE Transactions on Parallel and Distributed Systems. Hehas served as General or Technical Program Chair andTechnical Prgram Committee member of numerous IEEEand ACM conferences. He is a member of IEEE TCCC andTCPP Executive Committees.

LUCIANO LENZINI ([email protected]) holds a degree inphysics from the University of Pisa, Italy. He joined CNUCE,an institute of the Italian National Research Council (CNR),in 1970. In 1994 he joined the Department of InformationEngineering of the University of Pisa as a full professor. Hiscurrent research interests include the design and perfor-mance evaluation of MAC protocols for wireless networksand QoS provision in integrated and differentiated servicesnetworks. He is currently on the Editorial Boards of Com-puter Networks and the Journal of Communications andNetworks. He served as chairman of the 1992 IEEE Work-shop on Metropolitan Area Networks and the 2002 Euro-pean Wireless conference. He served as guest editor of theIEEE Journal on Selected Areas in Communications SpecialIssue on Analysis and Synthesis of MAC Protocols and theACM/Kluwer Wireless Networks Special Issue on the bestpapers of EW 2002. He has directed several national andinternational projects in the area of computer networking.

MARCO CONTI ([email protected]) is a research directorat IIT-CNR. He has published more than 180 researchpapers and two books: Metropolitan Area Networks(Springer, 1997) and Mobile Ad Hoc Networking (IEEE-Wiley, 2004). He has served as TPC chair of several confer-ences. including, IFIP-TC6 Networking 2002, IEEEWoWMoM 2005 and PerCom 2006, and ACM MobiHoc2006. He is Associate Editor of Pervasive and Mobile Com-puting (Elsevier), and Area Editor for IEEE Transactions onMobile Computing and Ad Hoc Networks.


channel assignment strategies

Documents