a
Broadband and Wireless Networking (BWN) Lab, School of Electrical and Computer Engineering,
Georgia Institute of Technology, Atlanta, GA 30332, USA
b
Kiyon, Inc., 4225 Executive Square, Suite 290, La Jolla, CA 92037, USA
Received 1 June 2004; received in revised form 1 November 2004; accepted 20 December 2004
Abstract
Wireless mesh networks (WMNs) consist of mesh routers and mesh clients, where mesh routers have minimal mobil-
ity and form the backbone of WMNs. They provide network access for both mesh and conventional clients. The inte-
gration of WMNs with other networks such as the Internet, cellular, IEEE 802.11, IEEE 802.15, IEEE 802.16, sensor
networks, etc., can be accomplished through the gateway and bridging functions in the mesh routers. Mesh clients can
be either stationary or mobile, and can form a client mesh network among themselves and with mesh routers. WMNs
are anticipated to resolve the limitations and to significantly improve the performance of ad hoc networks, wireless local
area networks (WLANs), wireless personal area networks (WPANs), and wireless metropolitan area networks
(WMANs). They are undergoing rapid progress and inspiring numerous deployments. WMNs will deliver wireless ser-
vices for a large variety of applications in personal, local, campus, and metropolitan areas. Despite recent advances in
wireless mesh networking, many research challenges remain in all protocol layers. This paper presents a detailed study
on recent advances and open research issues in WMNs. System architectures and applications of WMNs are described,
followed by discussing the critical factors influencing protocol design. Theoretical network capacity and the state-of-
the-art protocols for WMNs are explored with an objective to point out a number of open research issues. Finally, test-
beds, industrial practice, and current standard activities related to WMNs are highlighted.
2004 Elsevier B.V. All rights reserved.
Keywords: Wireless mesh networks; Ad hoc networks; Wireless sensor networks; Medium access control; Routing protocol; Transport
protocol; Scalability; Security; Power management and control; Timing synchronization
*
Corresponding author. Tel.: +1 425 442 5039.
E-mail addresses: ian@ece.gatech.edu (I.F. Akyildiz), wxudong@ieee.org (X. Wang).
1389-1286/$ - see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.comnet.2004.12.001
ARTICLE IN PRESS
sented in Section 16. Current status of standard not exist in these nodes. In addition, mesh clients
activities in WMNs is highlighted in Section 17. usually have only one wireless interface. As a con-
The paper is concluded in Section 18. sequence, the hardware platform and the software
for mesh clients can be much simpler than those
for mesh routers. Mesh clients have a higher vari-
2. Network architecture ety of devices compared to mesh routers. They can
be a laptop/desktop PC, pocket PC, PDA, IP
WMNs consist of two types of nodes: mesh rou- phone, RFID reader, BACnet (building automa-
ters and mesh clients. Other than the routing capa- tion and control networks) controller, and many
bility for gateway/repeater functions as in a other devices, as shown in Fig. 2.
conventional wireless router, a wireless mesh rou- The architecture of WMNs can be classified
ter contains additional routing functions to sup- into three main groups based on the functionality
port mesh networking. To further improve the of the nodes:
flexibility of mesh networking, a mesh router is
usually equipped with multiple wireless interfaces • Infrastructure/Backbone WMNs. The architec-
built on either the same or different wireless access ture is shown in Fig. 3, where dash and solid
technologies. Compared with a conventional wire- lines indicate wireless and wired links, respec-
less router, a wireless mesh router can achieve the tively. This type of WMNs includes mesh rou-
same coverage with much lower transmission ters forming an infrastructure for clients that
power through multi-hop communications. connect to them. The WMN infrastructure/
Optionally, the medium access control (MAC) backbone can be built using various types of
protocol in a mesh router is enhanced with better radio technologies, in addition to the mostly
scalability in a multi-hop mesh environment. used IEEE 802.11 technologies. The mesh rou-
In spite of all these differences, mesh and con- ters form a mesh of self-configuring, self-healing
ventional wireless routers are usually built based links among themselves. With gateway func-
on a similar hardware platform. Mesh routers tionality, mesh routers can be connected to
can be built based on dedicated computer systems the Internet. This approach, also referred to as
(e.g., embedded systems) and look compact, as
shown in Fig. 1. They can also be built based on
general-purpose computer systems (e.g., laptop/
desktop PC).
Mesh clients also have necessary functions for
mesh networking, and thus, can also work as a
router. However, gateway or bridge functions do
infrastructure meshing, provides backbone for • Client WMNs. Client meshing provides peer-to-
conventional clients and enables integration of peer networks among client devices. In this type
WMNs with existing wireless networks, through of architecture, client nodes constitute the
gateway/bridge functionalities in mesh routers. actual network to perform routing and configu-
Conventional clients with Ethernet interface ration functionalities as well as providing end-
can be connected to mesh routers via Ethernet user applications to customers. Hence, a mesh
links. For conventional clients with the same router is not required for these types of net-
radio technologies as mesh routers, they can works. The basic architecture is shown in Fig.
directly communicate with mesh routers. If dif- 4. In Client WMNs, a packet destined to a node
ferent radio technologies are used, clients must in the network hops through multiple nodes to
communicate with the base stations that have reach the destination. Client WMNs are usually
Ethernet connections to mesh routers. formed using one type of radios on devices.
Infrastructure/Backbone WMNs are the most Moreover, the requirements on end-user devices
commonly used type. For example, community is increased when compared to infrastructure
and neighborhood networks can be built using meshing, since, in Client WMNs, the end-users
infrastructure meshing. The mesh routers are
placed on the roof of houses in a neighborhood,
which serve as access points for users inside the
homes and along the roads. Typically, two
types of radios are used in the routers, i.e., for
backbone communication and for user commu-
nication, respectively. The mesh backbone com-
munication can be established using long-range
communication techniques including direc-
tional antennas. Fig. 4. Client WMNs.
ARTICLE IN PRESS
must perform additional functions such as rout- (LOS) links. To meet these requirements, the
ing and self-configuration. mesh-style multi-hopping is indispensable [85],
• Hybrid WMNs. This architecture is the combi- which achieves higher throughput without sac-
nation of infrastructure and client meshing as rificing effective radio range via shorter link dis-
shown in Fig. 5. Mesh clients can access the net- tances, less interference between the nodes, and
work through mesh routers as well as directly more efficient frequency re-use.
meshing with other mesh clients. While the • Support for ad hoc networking, and capability of
infrastructure provides connectivity to other self-forming, self-healing, and self-organization.
networks such as the Internet, Wi-Fi, WiMAX, WMNs enhance network performance, because
cellular, and sensor networks; the routing capa- of flexible network architecture, easy deploy-
bilities of clients provide improved connectivity ment and configuration, fault tolerance, and
and coverage inside the WMN. The hybrid mesh connectivity, i.e., multipoint-to-multi-
architecture will be the most applicable case in point communications [128]. Due to these fea-
our opinion. tures, WMNs have low upfront investment
requirement, and the network can grow gradu-
ally as needed.
3. Characteristics • Mobility dependence on the type of mesh nodes.
Mesh routers usually have minimal mobility,
The characteristics of WMNs are explained as while mesh clients can be stationary or mobile
follows: nodes.
• Multiple types of network access. In WMNs,
• Multi-hop wireless network. An objective to both backhaul access to the Internet and peer-
develop WMNs is to extend the coverage range to-peer (P2P) communications are supported
of current wireless networks without sacrificing [75]. In addition, the integration of WMNs with
the channel capacity. Another objective is to other wireless networks and providing services
provide non-line-of-sight (NLOS) connectivity to end-users of these networks can be accom-
among the users without direct line-of-sight plished through WMNs.
ARTICLE IN PRESS
• Enterprise networking. This can be a small net- utilization of enterprise networks. WMNs can
work within an office or a medium-size network grow easily as the size of enterprise expands.
for all offices in an entire building, or a large WMNs for enterprise networking are much
scale network among offices in multiple build- more complicated than at home because more
ings. Currently, standard IEEE 802.11 wireless nodes and more complicated network topolo-
networks are widely used in various offices. gies are involved. The service model of enter-
However, these wireless networks are still iso- prise networking can be applied to many other
lated islands. Connections among them have public and commercial service networking sce-
to be achieved through wired Ethernet connec- narios such as airports, hotels, shopping malls,
tions, which is the key reason for the high cost convention centers, sport centers, etc.
of enterprise networks. In addition, adding • Metropolitan area networks. WMNs in metro-
more backhaul access modems only increases politan area have several advantages. The phys-
capacity locally, but does not improve robust- ical-layer transmission rate of a node in WMNs
ness to link failures, network congestion and is much higher than that in any cellular net-
other problems of the entire enterprise network. works. For example, an IEEE 802.11g node
If the access points are replaced by mesh rou- can transmit at a rate of 54% Mbps. Moreover,
ters, as shown in Fig. 8, Ethernet wires can be the communication between nodes in WMNs
eliminated. Multiple backhaul access modems does not rely on a wired backbone. Compared
can be shared by all nodes in the entire network, to wired networks, e.g., cable or optical net-
and thus, improve the robustness and resource works, wireless mesh MAN is an economic
ARTICLE IN PRESS
is lower than 1.0 Mbps). The reason for low sca- • Security. Without a convincing security solu-
lability is that the end-to-end reliability sharply tion, WMNs will not be able to succeed due
drops as the scale of the network increases. In to lack of incentives by customers to subscribe
WMNs, due to its ad hoc architecture, the cen- to reliable services. Although many security
tralized multiple access schemes such as TDMA schemes have been proposed for wireless LANs,
and CDMA are difficult to implement due to they are still not ready for WMNs. For
their complexities and a general requirement instance, there is no centralized trusted author-
on timing synchronization for TDMA (and ity to distribute a public key in a WMN due to
code management for CDMA). When a distrib- the distributed system architecture. The existing
uted multi-hop network is considered, accurate security schemes proposed for ad hoc networks
timing synchronization within the global net- can be adopted for WMNs, but several issues
work is difficult to achieve [62]. Thus, distrib- exist:
uted multiple access schemes such as CSMA/ – Most security solutions for ad hoc networks
CA are more favorable. However, CSMA/CA are still not mature enough to be practically
has very low frequency spatial-reuse efficiency implemented.
[2], which significantly limits the scalability of – The network architecture of WMNs is differ-
CSMA/CA-based multi-hop networks. To ent from a conventional ad hoc network,
improve the scalability of WMNs, designing a which causes differences in security mecha-
hybrid multiple access scheme with CSMA/CA nisms.
and TDMA or CDMA is an interesting and
As a consequence, new security schemes ranging
challenging research issue.
from encryption algorithms to security key distri-
• Mesh connectivity. Many advantages of WMNs
bution, secure MAC and routing protocols, intru-
originate from mesh connectivity which is a crit-
sion detection, and security monitoring need to be
ical requirement on protocol design, especially
developed.
for MAC and routing protocols. Network self-
• Ease of use. Protocols must be designed to
organization and topology control algorithms
enable the network to be as autonomous as pos-
are generally needed. Topology-aware MAC
sible, in the sense of power management, self-
and routing protocols can significantly improve
organization, dynamic topology control, robust
the performance of WMNs.
to temporary link failure, and fast network-sub-
• Broadband and QoS. Different from other ad hoc
scription/user-authentication procedure. In
networks, most applications of WMNs are
addition, network management tools need to
broadband services with various QoS require-
be developed to efficiently maintain the opera-
ments. Thus, in addition to end-to-end transmis-
tion, monitor the performance, and configure
sion delay and fairness, more performance
the parameters of WMNs. These tools together
metrics such as delay jitter, aggregate and per-
with the autonomous mechanisms in protocols
node throughput, and packet loss ratios, must
will enable rapid deployment of WMNs.
be considered by communication protocols.
• Compatibility and inter-operability. It is a
desired feature for WMNs to support network
access for both conventional and mesh clients. 6. Capacity of WMNs
Thus, WMNs need to be backward compatible
with conventional client nodes; otherwise, the The capacity of WMNs is affected by many
motivation of deploying WMNs will be signifi- factors such as network architecture, network
cantly compromised. Integration of WMNs topology, traffic pattern, network node density,
with other wireless networks requires certain number of channels used for each node, trans-
mesh routers to have the capability of inter- mission power level, and node mobility. A clear
operation among heterogeneous wireless understanding of the relationship between net-
networks. work capacity and the above factors provides
ARTICLE IN PRESS
a guideline for protocol development, architec- Nevertheless, this implication has motivated
ture design, deployment and operation of the other research work such as [83,93], where a hy-
network. brid network architecture is considered to improve
the capacity of ad hoc networks. In the hybrid
6.1. Capacity analysis architecture, nodes only communicate with nearby
nodes. If they need to communicate with nodes
In the last decade, much research has been car- with many hops away, base stations or access
ried out to study the capacity of ad hoc networks points are used to relay packets via wired net-
which can be adopted to investigate the capacity works. The hybrid architecture can improve
of WMNs. capacity of ad hoc networks, however, it may still
For a stationary multi-hop network, it has been not be favored by many applications because
shown that the optimum transmission power level wired connections between base stations do not ex-
of a node is reached when the node has six neigh- ist in many ad hoc networks.
boring nodes [81]. With this value, an optimum The implication given in [54] can also be re-
tradeoff is achieved between the number of hops flected in [130]. The scheme proposed in [130] in-
from source to destination and the channel spa- creases the network capacity of ad hoc networks
tial-reuse efficiency. This result is useful for infra- by utilizing the node mobility. When a node needs
structure WMNs with minimal mobility. When to send packets to another node, it will not send
the mobility is a concern as in hybrid WMNs, no until the destination node is close to the source
theoretical results are reported so far. Some exper- node. Thus, via the node mobility, a node only
imental studies have been performed in [16], where communicates with its nearby nodes. This scheme
the simulation results of a stationary network val- has a limitation: The transmission delay may be-
idate the theoretical results of [81]. come large and the required buffer for a node
Analytical lower and upper bounds of network may be infinite.
capacity are given in [54]. From the analytical re- The analytical approach in [54] has significantly
sults, it follows that the throughput capacity per driven the progress in capacity research of ad hoc
node reduces significantly when the node density networks. However, it contains limitations. The
increases. An important implication is derived in networking protocols have not been fully captured
[54] as a guideline to improve the capacity of ad by the analysis. For example, power control mech-
hoc networks: A node should only communicate anisms, commonly used to improve the network
with nearby nodes. To implement this idea, two capacity, is not considered in the analysis. As an-
major schemes are suggested in [54]: other example, the characteristics of ad hoc rout-
ing protocols have not been totally captured in
• Throughput capacity can be increased by the analysis. In any routing protocol, the route
deploying relaying nodes. for packets does not necessary follow the path
• Nodes need to be grouped into clusters. along the straight-line segment between the source
and destination as given in the analysis, because
Thus, communications of a node with another the routing protocol determines a path according
node that is not nearby must be conducted to certain metrics such as the number of hop
through relaying nodes or clusters. However, these counts, link quality, etc. [41].
schemes have limitations. In the first scheme, a As a result, the applicability of the theoretical
very large number of relaying nodes are needed results on practical network architectures still re-
in order to increase the throughput by a significant mains unclear. A close match between the theoret-
percent. This will definitely increase the overall ical results in [54] and IEEE 802.11 based ad hoc
cost of a network. In the second scheme, clustering networks is reported in [90]. However, this study
nodes in ad hoc networks or WMNs is not a pre- relies on the assumption that the traffic pattern
ferred approach, because it is difficult to manage in a large ad hoc network tends to be local and
clusters in a distributed system. thus, nodes usually communicate with nearby
ARTICLE IN PRESS
nodes. This assumption is not always valid in a 2. Despite much research progress has been made
network unless it is intentionally designed so. in network capacity analysis of ad hoc net-
Most of the existing analytical approaches are works, WMNs have not been fully explored
based on asymptotic analysis [54]. The upper or due to the differences between WMNs and ad
lower capacity bounds derived from these ap- hoc networks. The research results about net-
proaches do not reveal the exact capacity of an work capacity and optimum node density of
ad hoc network with a given number of nodes, in ad hoc networks may not directly be applicable
particular when the number is small. Recently, to WMNs. For example, in [54], the network
an analytical approach is proposed in [75] to study architecture in the analysis does not match that
the exact capacity of WMNs. The analysis is sim- of WMNs, because both stationary and mobile
plified by taking advantage of the low mobility ad hoc nodes exist in WMNs.
feature of WMNs. However, the analytical model 3. Important techniques of increasing capacity of
in [75] contains three assumptions that are not nec- WMNs have not been considered in the analyt-
essarily valid. ical models for ad hoc networks. For example,
multi-channels per radio or multi-radios per
• The traffic in all nodes is sent to a single gate- node will be applied in WMNs. Then, a critical
way which is not the case in WMNs. question that arises is: what is the optimum
• Each node receives an equal share of the band- number of channels or radios for each network
width to achieve fairness. However, this node. Although the analytical model in [54]
assumption is not valid if the network nodes allows multi-channels in a node, it does not con-
have different distances between them. tain a scheme to find the optimum number of
• The unidirectional traffic case is mentioned to channels. When other advanced techniques such
be easily extendable to the bidirectional traffic as directional antennas, multi-input multi-out-
case. However, the network capacity becomes put (MIMO) systems, are considered, new ana-
totally different if bidirectional traffic is lytical models are required.
considered.
• The network architecture considered is actually
still an ad hoc network. Furthermore, only a
specific MAC protocol very similar to CSMA/ 7. Physical layer
CA with RTS/CTS is considered. However,
CSMA/CA is not the only MAC solution for 7.1. Advanced physical layer techniques
mesh networks. For example, the IEEE
802.11e or a TDMA MAC can achieve higher Physical layer techniques advance fast as RF
throughput than CSMA/CA, because of the and circuit design for wireless communications
existence of contention free periods (CFP). evolve. Most of existing wireless radios are able
to support multiple transmission rates by a combi-
6.2. Open research issues nation of different modulation and coding rates
[67,68]. With such modes, adaptive error resilience
Many research issues still exist in the capacity can be provided through link adaptation
analysis of WMNs for several reasons: [115,31,52]. It should be noted that under a fre-
quency selective fading environment, a link adap-
1. The theoretical results on the capacity of either tation algorithm cannot take signal-to-noise ratio
ad hoc networks or WMNs are still based on (SNR) or carrier-to-interference ratio (CIR) as a
some simplified assumptions, as explained single input from the physical layer, because
before. The derivation of new results by consid- SNR or CIR alone does not adequately describe
ering critical factors such as transmission power the channel quality [88].
levels, traffic patterns, optimal routing path, In order to increase the capacity of wireless net-
etc., is still a challenging research topic. works, various high-speed physical techniques
ARTICLE IN PRESS
have been invented. For example, orthogonal fre- ratio combining [104]. When strong interference
quency multiple access (OFDM) has significantly is present, diversity processing alone is insufficient
increased the speed of IEEE 802.11 from 11 Mbps to receive signals with high quality. To resolve this
to 54 Mbps. A much higher transmission rate can issue, adaptive antenna array processing is used to
be achieved through ultra-wide band (UWB) tech- shape the antenna beamform so as to enhance the
niques. However, UWB is only applicable to short- desired signals while to nullify the interfering sig-
distance applications such as wireless personal nals. The technique for adaptive antenna process-
area networks (WPANs). If a transmission speed ing is called optimum combining. It assumes that
as high as that of UWB is desired in a wider area part information of the desired signal can be ac-
network such as WLANs or WMANs, new physi- quired through a training sequence.
cal layer techniques are needed. Antenna diversity and smart antenna tech-
In order to further increase capacity and miti- niques are also applicable to WMNs and other
gate the impairment by fading, delay-spread, and ad hoc networks. However, their performance in
co-channel interference, multiple-antenna systems WMNs or any other ad hoc networks needs more
have been used for wireless communication evaluation. Examples of analyzing smart antenna
[104,21]. Considering communications between systems for MANETs are reported in [18,117].
nodes A and B in Fig. 12, node A is assumed to Due to complexity and cost, a fully adaptive smart
have M antennas for transmission and N antennas antenna system is only used in base stations of cel-
for reception, while in node B there are K antennas lular networks. On-going research and develop-
for transmission and L antennas for reception. ment efforts are still needed to implement fully
Different values of M, N, K, L result in various adaptive smart antenna system in a mobile termi-
multiple-antenna systems. If multiple antennas nal. For WMNs, low-cost is a challenging issue.
are in the receiver but single channel in the trans- As a consequence, directional antennas have been
mitter (i.e., K = 1, M = 1 and either L > 1 or actively researched in the area of ad hoc networks.
N > 1), techniques such as antenna diversity and A mechanically or electronically steerable or
adaptive/smart antennas can be used for a multi- switched directional antenna system can be tuned
antenna system. They have been proposed for to a certain direction. By using directional trans-
point-to-multipoint one-hop cellular networks. mission, interference between network nodes can
Antenna diversity is based on the fact that signals be mitigated and thus, network capacity can be im-
received from uncorrelated antennas have inde- proved [124,118]. Directional antenna can also im-
pendent fading. Thus, it has high probability that prove energy efficiency [123]. However, directional
at least one good signal can be received at the re- antennas bring challenges to the MAC protocol
ceiver. Antenna uncorrelation is usually achieved design [144,105,82,36].
through space, polarization, or pattern diversity, If multiple antennas are in the transmitter and
and the processing technologies for diversity in- single antenna in the receiver, i.e., N = 1, L = 1
clude switch diversity, equal gain, and maximum and either K > 1 or M > 1, antenna diversity or
smart antenna cannot be applied unless the chan-
nel state information (CSI) is available. However,
usually partial information of channel state is
M L
available at the transmitter. To achieve diversity
.. .. under this situation, a commonly used technique
. .
Node Node is space–time coding (STC) [12], where signals
.. .. transmitted at different antennas in different sym-
A . . B
.. bol periods are processed with a certain coding
. .. technique. The received signals are then combined
N K .
at the receiver through an appropriate algorithm
such as maximum likelihood detection (MLD).
Fig. 12. Multiple-antenna systems. STC is a promising technique that achieves second
ARTICLE IN PRESS
order diversity without bandwidth expansion censed) have not been utilized efficiently. Measure-
[104]. To date, if CSI is not available, no solution ments by the FCC show that around 70% of
has been developed yet for smart antennas at a allocated spectrum is not utilized [45,89]. In addi-
transmitter. Schemes such as [35] still assume that tion, the time scale of spectrum occupancy can
CSI is perfectly known. vary from milliseconds to hours [45]. Therefore,
If multiple antennas are in both the transmitter abundant spectrum is still available for wireless
and the receiver, i.e., M > 1, L > 1 or K > 1, N > 1, communication. Furthermore, in a large scale ad
the multiple-antenna system is an MIMO system, hoc network, the complexity is beyond human
where both diversity and simultaneous transmis- planning, and thus, conventional static frequency
sions exist. Thus, MIMO can potentially increase planning becomes impossible [96]. To achieve
the system capacity by three times or even more much better spectrum utilization and viable fre-
[94]. Currently MIMO is being adopted into IEEE quency planning, frequency agile [97] or cognitive
802.11n [64]. Depending on where the MIMO pro- radios [89] are being developed to dynamically
cessing is placed, MIMO systems can be catego- capture this unoccupied spectrum. The FCC has
rized into three types: receiver processing only, recognized the promising future of this technique
transmitter processing only, and both transmitter and pushes to enable it to a full realization. In or-
and receiver processing MIMO systems. The pro- der to implement cognitive radios, software de-
cessing techniques can be based on maximum like- fined radio (SDR) is one of the most convenient
lihood detection (MLD), vertical Bell Lab Layered platforms [46] because programmability exists in
Space–Time (V-BLAST) [51], singular value all components of a radio such as programmable
decomposition (SVD) [13], and space–time coding. RF bands, channel access modes, and channel
So far only few results have been reported on the modulations [102]. SDR is not a mature technique
research of applying STC and MIMO to WMNs yet, although testbeds are available now [139].
as well as other ad hoc networks. However, for the long term, SDR will be a key
Since multiple channels are usually available in technique for wireless communications. It cannot
the frequency band of a wireless radio, they can be only realize the cognitive radios, but can also eas-
used to increase the capacity. A single-transceiver ily implement all other advanced physical tech-
radio can use different channels by channel switch- niques such as adaptive modulation and coding,
ing on the time axis according to the needs of high- MIMO system [139], controller for smart and
er layer protocols. For a multi-transceiver radio, directional antennas, multi-channel radio, and
simultaneous transmissions in different channels multi-radio systems.
can be supported. Multiple transceivers can be eas-
ily implemented in a base station of cellular net- 7.2. Open research issues
works. However, with the concern of cost and
system complexity, a wireless radio with multiple Open issues in the physical layer are twofold.
transceivers has not become a mature technique First, it is necessary to further improve the trans-
yet, although IEEE 802.11 chipsets with multiple mission rate and the performance of physical layer
transceivers are already available [44]. In some sit- techniques. New wideband transmission schemes
uations, the system capacity of a network node can other than OFDM or UWB are needed in order
be improved by using multiple radios each with to achieve higher transmission rate in a larger area
single or multiple channels. Since each radio con- network. Multiple-antenna systems have been re-
tains both MAC and physical layers, in order to searched for years. However, their complexity
make a multi-radio network work as a single node, and cost are still too high to be widely accepted
a virtual MAC protocol is usually required to for WMNs. An example of low-cost directional
coordinate the communication in all radios [3]. antenna implementation is reported in [76]. Fre-
For a wireless network, the frequency band is a quency agile techniques are still in the early phase.
very precious resource. However, many of existing Many challenging issues need to be resolved before
allocated frequency bands (both licensed and unli- they can be accepted for commercial use [89].
ARTICLE IN PRESS
Second, to best utilize the advanced features • Network self-organization is needed for the
provided by physical layer, higher layer protocols, MAC. MAC protocol should have the knowl-
especially MAC protocols, need to be carefully de- edge about network topology which can help
signed. Otherwise, the advantages brought by such better cooperation between neighboring nodes
physical layer techniques will be significantly com- and nodes in multi-hop distances. This can
promised. For directional and smart antennas, significantly improve the MAC performance in
many MAC protocols have been proposed for ad a multi-hop environment. In some circum-
hoc networks [144,105,82,36]. A MAC protocol stances, network self-organization based on
for MIMO systems is studied in [126]. However, power control can optimize network topology
for multi-antenna systems, an efficient MAC pro- [91], minimize the interference between neigh-
tocol to achieve significant throughput improve- boring nodes, and thus, improve the network
ment is still needed, as will be discussed in capacity.
Section 8. Communication protocols for cognitive • Mobility affects the performance of MAC.
radios remain an open issue. Significant research Mobility dynamically changes network con-
efforts are needed to make cognitive-radio based figuration, and thus, may significantly impact
WMNs become practical. the performance of the MAC protocol. In
order to be adaptive to mobility or even to
utilize the mobility [130], the network nodes
8. MAC layer need to exchange network topology infor-
mation.
MAC protocols for WMNs have the following
differences compared to classical counterparts for These differences must be considered in order to
wireless networks: design a scalable MAC for WMNs.
The scalability of WMNs can be addressed by
• MAC for WMNs is concerned with more than the MAC layer in two ways. The first way is to en-
one hop communication. Classical MAC proto- hance existing MAC protocols or propose new
cols are limited to one-hop communication MAC protocols to increase end-to-end throughput
while the routing protocol takes care of multi- when only single channel is available in a network
hop communication. This assumption makes node. The second way is to allow transmission on
protocol design easier, since MAC and routing multiple channels in each network node. In the fol-
are transparent to each other. However, this lowing, existing single-channel and multi-channel
method does not work well in WMNs, because MAC protocols are studied separately. Since IEEE
data transmission and reception at a node is not 802.11 is a widely accepted radio technique for
only affected by nodes within one hop but WMNs, most of the following discussions are fo-
within two or more hops away. The hidden- cused on the IEEE 802.11 MAC, i.e., CSMA/CA
node issue in a multi-hop wireless LAN is such with RTS/CTS.
an example.
• MAC is distributed and cooperative and works 8.1. Single-channel MAC
for multipoint-to-multipoint communication. In
WMNs, no centralized controller is available. There are three approaches in this case:
The MAC function is accomplished in a distrib-
uted way, i.e., the MAC protocol must ensure • Improving existing MAC protocols. Currently
all nodes to cooperate in transmission. In addi- several MAC protocols have been proposed
tion, any network node with mesh networking for multi-hop ad hoc networks by enhancing
capability is able to communicate all its neigh- the CSMA/CA protocol [27,116]. These schemes
boring mesh nodes. Thus, multipoint-to-multi- usually adjust parameters of CSMA/CA such as
point communications can be established contention window size and modify backoff
among these nodes. procedures. They may improve throughput for
ARTICLE IN PRESS
one-hop communications. However, for multi- erative MAC with TDMA or CDMA. The
hop cases such as in WMNs, these solutions other is the compatibility of TDMA (or
still reach a low end-to-end throughput, because CDMA) MAC with existing MAC protocols.
they cannot significantly reduce the probabil- For example, in IEEE 802.16, the original
ity of contentions among neighboring nodes. MAC protocol is a centralized TDMA scheme.
As long as contention occurs frequently, A distributed TDMA MAC for IEEE 802.16
whichever method is taken to modify backoff mesh is still being researched. In WMNs based
and contention resolution procedures, the end- on IEEE 802.11, how to design a distributed
to-end throughput will be significantly reduced TDMA MAC protocol overlaying CSMA/CA
due to the accumulating effect on the multi- is an interesting but challenging problem [134].
hop path. For distributed TDMA or CDMA MAC proto-
• Cross-layer design with advanced physical layer cols, network self-organization based on topol-
techniques. Two major schemes exist in this cat- ogy control and/or power control must also be
egory: MAC based on directional antenna considered.
[82,36] and MAC with power control [112].
The first set of schemes eliminates exposed 8.2. Multi-channel MAC
nodes if antenna beam is assumed to be perfect.
However, due to the directional transmission, A multi-channel MAC can be implemented on
more hidden nodes are produced in this case. several different hardware platforms, which also
Thus, new solutions must be developed to impacts the design of the MAC. A multi-channel
reduce the number of hidden nodes. Moreover, MAC may belong to one of the following
MAC protocols based on directional antennas categories:
also face other difficulties such as cost, system
complexity, and practicality of fast steerable • Multi-channel single-transceiver MAC. If the
directional antennas. The second set of schemes cost and compatibility are the concern, one
is developed for the purpose of reducing power transceiver on a radio is a preferred hardware
consumptions [33,147,131]. These schemes platform. Since only one transceiver is avail-
reduce exposed nodes problem, especially in a able, only one channel is active at a time in each
dense network, and thus, improve the spectrum network node. However, different nodes may
spatial-reuse factor in WMNs [2]. However, operate on different channels simultaneously in
hidden nodes still exist and may become worse order to improve system capacity. To coordi-
because lower transmission power level reduces nate transmissions between network nodes
the possibility of detecting a potential interfer- under this situation, protocols such as the
ing node [72]. multi-channel MAC in [122] and the seed-slot-
• Proposing innovative MAC protocols. In order ted channel hopping (SSCH) scheme [14] are
to fundamentally resolve the issue of low end- needed. SSCH is actually a virtual MAC proto-
to-end throughput in a multi-hop ad hoc envi- col, since it works on top of IEEE 802.11 MAC
ronment such as WMNs, innovative solutions and does not need changes in the IEEE 802.11
are necessary. Determined by their poor scala- MAC.
bility in an ad hoc multi-hop network, random • Multi-channel multi-transceiver MAC. In this
access protocols such as CSMA/CA are not an scenario, a radio includes multiple parallel RF
efficient solution. Thus, revisiting the design of front-end chips and baseband processing mod-
MAC protocols based on TDMA or CDMA ules to support several simultaneous channels.
is an important research topic [78,134]. To date, On top of the physical layer, there is only one
few TDMA or CDMA MAC protocols have MAC layer to coordinate the functions of mul-
been proposed for WMNs. This is probably tiple channels. Engim multi-channel wireless
because of two factors. One is the complexity LAN switching engine [44] belongs to this cate-
and cost of developing a distributed and coop- gory. However, how to design an efficient MAC
ARTICLE IN PRESS
protocol for this type of physical layer platform channel is not always appropriate. Using pend-
is still an open research topic. ing packets as a metric to select a channel
• Multi-radio MAC. In this scenario, a network achieve better performance [122].
node has multiple radios each with its own • The MMAC eliminates multi-channel hidden
MAC and physical layers. Communications in nodes, but it also generates many exposed nodes
these radios are totally independent. Thus, a vir- because of using RTS/CTS and ATIM/ATIM-
tual MAC protocol such as the multi-radio unifi- ACK (for default channel) procedures.
cation protocol (MUP) [3] is required on top of
MAC to coordinate communications in all chan- In MUP, there are multiple wireless network
nels. In fact one radio can have multiple chan- interface cards (NICs) on each node. Channels
nels. However, for simplicity of design and on all NICs are orthogonal and fixed. The major
application, a single channel is used in each radio. functions of MUP [3] include:
To illustrate typical issues existing in multi- • Discovering neighbors. After the discovering
channel MAC protocols, here we explain two pro- procedures, neighbors are classified into MUP-
tocols in detail: multi-channel MAC (MMAC) enabled and legacy nodes.
[122] and multi-radio unification protocol (MUP) • Selecting a NIC based on one-hop round trip time
[3]. (RTT) measurements. MUP selects the NIC
As far as MMAC is concerned, there are three with the shortest RTT between a node and its
main functions: neighbors.
• Utilizing the selected NIC for a long period. This
• Maintaining data structure of all channels in each period is determined by a random process and
node. Channels of a node are classified into in the order of 10–20 s.
three types depending on its status of allocation. • Switching channels. After the random time
• Negotiating channels during ad hoc traffic indica- period, all NICs are measured again through
tion message (ATIM) window. Negotiations are one-hop probe messages. If an NIC has a cer-
done through a pre-defined channel known to tain amount of quality improvement than the
all nodes. existing NIC, then it is selected for sending
• Selecting a channel. The criterion is to use a packets.
channel with the lowest count of source–desti-
nation pairs that have selected the channel. Several issues still need to be investigated
further:
Several problems have not been solved in the
MMAC [122]. • Hidden node issue is not effectively solved. The
channel quality measurement is based on one-
• It is assumed that RTS/CTS always work in hop RTT. However, measurements based on
IEEE 802.11 DCF. In reality, RTS/CTS is an shortest RTT do not guarantee that there exists
optional function of DCF, and it may cause a no hidden nodes. For example, suppose nodes
high standard. A and C are hidden from each other and node
• Global synchronization in the network is diffi- B is a neighbor of both A and C. Then the same
cult to achieve in an ad hoc network with a channel can be selected by nodes A and C. The
large number of hops and nodes. RTS/CTS mechanism can be applied to reduce
• The channel switching time may be much larger the collision probability in this situation, but it
than 224 ls [122]. A larger channel switching causes very high overhead.
time will significantly degrade the performance • NIC switching mechanism is not justified. MUP
of a multi-channel MAC protocol [3]. allocates a random time period for each selected
• Channel selection criterion based on the lowest NIC. Performance of this scheme cannot be
number of source–destination pairs for each guaranteed, because the time of having the best
ARTICLE IN PRESS
quality in a NIC is not randomized but related radios such as IEEE 802.11, 802.16, 802.15, etc.,
to the wireless channel characteristics and inter- can seamlessly work together. Reconfigurable/soft-
ference from nodes using the same channel. ware radios and the related radio resource man-
• Packet re-ordering is needed after NIC switch- agement schemes may be the ultimate solution to
ing. MUP relies on TCP to handle this issue. these bridging functions.
However, this will cause low end-to-end Multi-channel MAC protocols for radios with
throughput in a multi-hop network like WMNs. multiple transceivers have not been thoroughly ex-
plored, possibly due to the relatively high cost of
In addition, fixed channel assignment on each such radios. However, as the cost goes down, a
NIC also limits the flexibility of MUP. multi-channel multi-transceiver MAC will be a
rather promising solution for WMNs. To really
8.3. Open research issues achieve spectrum efficiency, a multi-channel
MAC protocol must include the single-channel
To the best of our knowledge, the scalability is- solution that can fundamentally resolve the scala-
sue in multi-hop ad hoc networks has not been bility issue of WMNs. Otherwise, the throughput
fully solved yet. Most of existing MAC protocols per node per channel will still be very low. How
based on CSMA/CA solve partial problems of to apply the innovative single-channel solution to
the overall issue, but raise other problems. Thus, a multi-radio or multi-channel system is another
how to fundamentally improve the scalability of research problem.
CSMA/CA is a challenging problem. A distributed Most of the existing research efforts in MAC
TDMA MAC overlaying CSMA/CA [134] ad- are focused on capacity, throughput, and fairness.
dressed one possible solution to this problem. However, many applications need to support
For networks based on techniques other than broadband multimedia communication in WMNs.
CSMA/CA, both TDMA and CDMA can be ap- Thus, the development of MAC protocols with
plied to WMNs, if a distributed scheme can be multiple QoS metrics such as delay, packet loss ra-
developed to locally eliminate the difficulties of tios, and delay jitter is an important topic for
implementing TDMA or CDMA in an ad hoc net- WMNs.
work. Many other interesting research problems Another challenge related to MAC is the imple-
related to the scalability issue of ad hoc networks mentation, because both software and firmware
still remain to be solved. When advanced tech- may be involved when a MAC protocol is to be
niques such as MIMO and cognitive radios are modified. As an example in IEEE 802.11 MAC,
used in the physical layer, novel MAC protocols although chipset manufacturers have put efforts
need to be proposed to utilize the agility provided to pull up more functions in the firmware into
by the physical layer. the driver level as software, many timing critical
Due to the differences between WMNs and ad functions remain in the firmware. Such a ‘‘thin’’
hoc networks, a scalable MAC protocol for ad MAC solution provides little flexibility in modify-
hoc networks may not be effective to WMNs. In ing MAC protocols. To avoid modifying firmware,
WMNs, mesh routers and mesh clients hold signif- one approach is to design a MAC without cou-
icantly different characteristics such as mobility, pling with firmware. For example, the virtual
power constraints, etc. Same distributed solution MAC protocols do not require any modification
may not work for both mesh routers and clients. in firmware or hardware. However, in some cir-
Thus, a MAC protocol for WMNs must consider cumstances key functions in the firmware need to
both scalability and heterogeneity between differ- be modified in order to significantly improve the
ent network nodes. performance of the MAC protocol. Changing the
Some mesh routers in WMNs are responsible firmware is a doable but not a viable solution
for integration of various wireless technologies. due to its cost and complexity. A solution to this
Thus, advanced bridging functions must be devel- problem is to choose a more flexible MAC proto-
oped in the MAC layer so that different wireless col architecture. To our knowledge, there are sev-
ARTICLE IN PRESS
eral IEEE 802.11 chipset manufacturers that have Such differences imply that the routing protocols
eliminated firmware in their MAC implementation designed for ad hoc networks may not be appro-
architecture. With such an architecture, a true soft priate for WMNs.
MAC [134] or even a programmable MAC can be Based on the performance of the existing rout-
implemented. When software radios become ma- ing protocols for ad hoc networks and the specific
ture enough for commercial use, more flexible requirements of WMNs, we believe that an opti-
and powerful MAC protocols can be easily mal routing protocol for WMNs must capture
developed. the following features:
constraint of power consumption in mesh rou- WMNs. A new performance metric, called the
ters, a much simpler routing protocol can be weighted cumulative expected transmission time
developed for mesh routers than existing ad (WCETT) is proposed for the routing protocol.
hoc routing protocols. However, for mesh cli- WCETT takes into account both link quality met-
ents, the routing protocol must have the full ric and the minimum hop-count. It can achieve
functions of ad hoc routing protocols. Conse- good tradeoff between delay and throughput be-
quently, it is necessary to design an efficient cause it considers channels with good quality and
routing protocol for WMNs that can adaptively channel diversity in the same routing protocol.
support both mesh routers and mesh clients. In WMNs, multi-channel per radio is another
alternative to improve the capacity. For this type
In the rest of this section, we discuss various of networks, the scheme proposed in [42] is not
routing protocols applicable to WMNs and applicable because significant differences exist be-
emphasize the open research issues. tween a multi-channel node and a multi-radio
node as explained in Section 8.
9.1. Routing protocols with various performance
metrics 9.3. Multi-path routing for load balancing and fault
tolerance
The impact of performance metrics on a routing
protocol is studied in [41], where the link quality The main objective of using multi-path routing
source routing (LQSR) is proposed on the is to perform better load balancing and to provide
basis of DSR. LQSR aims to select a routing high fault tolerance [103]. Multiple paths are se-
path according to link quality metrics. Three lected between source and destination. Packets
performance metrics, i.e., the expected transmis- flow in one of these selected paths. When link is
sion count (ETX) [39], per-hop RTT, and per- broken on a path due to a bad channel quality
hop packet pair are implemented separately in or mobility, another path in the set of existing
LQSR. paths can be chosen. Thus, without waiting for set-
The performance of the routing protocol with ting up a new routing path, the end-to-end delay,
these three performance metrics is also compared throughput, and fault tolerance can be improved.
with the method using the minimum hop-count. However, the improvement depends on the avail-
For stationary nodes in WMNs, ETX achieves ability of node-disjoint routes between source
the best performance, while the minimum hop- and destination.
count method outperforms the three link quality A drawback of multi-path routing is its com-
metrics when nodes are mobile. The reason is that, plexity. Whether or not the multi-path routing
as the sender moves, the ETX metric cannot can be used for WMNs needs to be investigated
quickly track the change in the link quality. This depending on applications. Another problem is
result illustrates that the link quality metrics used that multi-path routing is infeasible if the shortest
in [41] are still not enough for WMNs when mobil- path is taken as the routing performance metric.
ity is concerned. Better performance metrics need Unless a large number of shortest paths are se-
to be developed, and routing protocols integrating lected, load distribution is almost the same as sin-
multiple performance metrics are necessary for gle shortest path routing [50]. Thus, how to design
WMNs. an effective multi-path routing protocol with
appropriate performance metrics is an interesting
9.2. Multi-radio routing research topic.
stead of addressing each of them, we describe the proposed in [58], which reduces signaling overhead
common principle of these routing protocols. by eliminating the periodic hello messages in other
In hierarchical routing, a certain self-organiza- greedy routing algorithms. However, all greedy
tion scheme is employed to group network nodes routing algorithms have a common problem:
into clusters. Each cluster has one or more cluster Delivery is not guaranteed even if a path exists be-
heads. Nodes in a cluster can be one or more hops tween source and destination. Partial flooding and
away from the cluster head. Since connectivity be- keeping the past routing information can help to
tween clusters are needed, some nodes can commu- guarantee delivery. However, these approaches in-
nicate with more than one cluster and work as a crease communication overhead and lose the state-
gateway. Routing within a cluster and routing be- less property of single-path greedy routing [48].
tween clusters may use different mechanisms. For In order to keep the stateless property and guar-
example, inter-cluster routing can be a proactive antee delivery, planar-graph based geographic
protocol, while intra-cluster routing can be on de- routing algorithms are proposed recently [23,38].
mand [120]. However, open issues still remain in these algo-
When the node density is high, hierarchical rithms. For example, in the face routing algorithm
routing protocols tend to achieve much better per- [23], the communication overhead is much higher
formance because of less overhead, shorter aver- than in the single-path greedy routing algorithm
age routing path, and quicker set-up procedure [48]. Thus, the face routing algorithm is mainly
of routing path. However, the complexity of main- used a recovery scheme when the greedy routing
taining the hierarchy may compromise the perfor- algorithm fails.
mance of the routing protocol. In WMNs,
hierarchical routing actually may face the imple- 9.6. Open research issues
mentation difficulty, because a node selected as a
cluster head may not necessarily have higher pro- Scalability is the most critical question in
cessing capability and channel capacity than the WMNs. Hierarchical routing protocols can only
other nodes. Unless being intentionally designed partially solve this problem due to their complexity
so, the cluster head may become a bottleneck. and difficulty of management. Geographic routing
Hierarchical routing provides a possible approach relies on the existence of GPS or similar position-
for scalability. However, whether or not these hier- ing technologies, which increases cost and com-
archical schemes can really solve the scalability plexity of WMNs. Moreover, the inquiry of
problem still remains a question. destination position produces additional traffic
load. Thus, new scalable routing protocols need
9.5. Geographic routing to be developed. Existing performance metrics
incorporated into routing protocols need to be ex-
Compared to topology-based routing schemes, panded. Moreover, how to integrate multiple per-
geographic routing schemes forward packets by formance metrics into a routing protocol so that
only using the position information of nodes in the optimal overall performance is achieved is a
the vicinity and the destination node [48]. Thus, challenging issue.
topology change has less impact on the geographic Routing for multicast applications is another
routing than other routing protocols. important research topic. Many applications of
Early geographic routing algorithms are actu- WMNs need multicasting capability. For example,
ally a type of single-path greedy routing schemes in a community or a city-wide network, video dis-
in which packet forwarding decision is made based tribution is a common application.
on the location information of current forwarding Cross-layer design between routing and MAC
node, its neighbors, and the destination node. Var- protocols is another interesting research topic. Pre-
ious greedy routing algorithms differ in the optimi- viously, routing protocol research was focused on
zation criterion applied in the forwarding decision. layer-3 functionality only. However, it has been
To improve power efficiency, a greedy algorithm is shown that the performance of a routing protocol
ARTICLE IN PRESS
may not be satisfactory in this case. Adopting mul- protocols that are an enhanced version of the clas-
tiple performance metrics from layer-2 into rout- sical TCP for wired networks. In an entirely new
ing protocols is an example. However, transport protocol [127], the reliable transport
interaction between MAC and routing is so close mechanism is designed from a fresh start, with
that merely exchanging parameters between proto- an objective to avoid fundamental problems in
col layers is not adequate. Merging certain func- TCP.
tions of MAC and routing is a promising
approach. 10.1.1. TCP variants
When multi-radio or multi-channel wireless The performance of classical TCPs degrades
mesh nodes are considered, new routing protocols significantly in ad hoc networks. In this section,
are needed for two reasons. First, the routing pro- we discuss various enhanced TCP protocols by
tocol not only needs to select a path in-between addressing the fundamental problems in TCP
different nodes, it also needs to select the most and the corresponding solutions.
appropriate channel or radio on the path. Second, One of the well-known reasons for TCP perfor-
cross-layer design becomes a necessity because mance degradation is that the classical TCPs do
change of a routing path involves the channel or not differentiate congestion and non-congestion
radio switching in a mesh node. Without consider- losses [142]. As a result, when non-congestion
ing cross-layer design, the switching process may losses occur, the network throughput quickly
be too slow to degrade the performance of WMNs. drops. Moreover, once wireless channels are back
The existing routing protocols treat all network to the normal operation, the classical TCP cannot
nodes in the same way. However, such solutions be recovered quickly. The protocol in [29] en-
may not be efficient for WMNs, because the mesh hances TCP through a feedback mechanism to dif-
routers in WMNs backbone and mesh clients have ferentiate between losses caused by congestion or
significant differences in power constraint and wireless channels. This concept can be adopted
mobility. More efficient routing protocols that to WMNs. However, how to design a loss differen-
take into account these differences are desired for tiation approach and accordingly modify the TCP
WMNs. for WMNs accordingly is subject to future study.
Link failure also degrades the TCP perfor-
mance. Link failure may occur frequently in mo-
10. Transport layer bile ad hoc networks since all nodes are mobile.
As far as WMNs are concerned, link failure is
To the best of our knowledge, no transport pro- not as critical as in mobile ad hoc networks, be-
tocol has been introduced specifically for WMNs cause the WMN infrastructure avoids the issue
to date, although several transport protocols have of single-point-of-failure. However, due to wireless
been developed for both wired and wireless net- channels and mobility in mesh clients, link failure
works in the last decade. In this section, we explain may still happen. To enhance TCP performance,
existing transport protocols with a focus on ad hoc congestion losses and link failure also need to be
networks, since WMNs share common features differentiated. Schemes similar to explicit link fail-
with ad hoc networks in spite of their differences. ure notification (ELFN) scheme [59] can perform
Then, we also discuss the research challenges. such differentiations.
TCP is critically dependent on ACK, so its per-
10.1. Protocols for reliable data transport formance can be severely impacted by network
asymmetry which is defined as the situation where
To date, a large number of reliable transport the forward direction of a network is significantly
protocols have been proposed for ad hoc net- different from the reverse direction in terms of
works. They can be classified into two types: bandwidth, loss rate, and latency [15]. In WMNs,
TCP variants and entirely new transport proto- TCP data and TCP ACK packets may take differ-
cols. TCP variants [15,29,59,92] include transport ent paths, and thus experience different packet loss
ARTICLE IN PRESS
rate, latency, or bandwidth. Even if the same path lid for WMNs, since WMNs will be integrated
is taken by TCP data and ACK packets, they still with the Internet and many other wireless net-
face network asymmetry problem, because the works. Transport protocols for WMNs must be
channel condition and bandwidth on the path var- compatible with TCPs in other networks.
ies from time to time. Consequently, TCP has poor
performance for wireless multihop ad hoc net- 10.2. Protocols for real-time delivery
works [110,141]. To solve the network asymmetry
problem, schemes such as ACK filtering, ACK To support end-to-end delivery of real-time
congestion control, etc., [15] have been proposed. traffic, UDP instead of TCP is usually applied as
However, a different network architecture is stud- a transport protocol. However, the simple mecha-
ied in [15], i.e., the effectiveness of these schemes nism of UDP cannot guarantee real-time delivery
in WMNs needs investigation. and may starve TCP connections in the same net-
In WMNs, mesh routers and mesh clients are work. Thus, additional protocols such as real-time
connected as an ad hoc network, so dynamic protocol (RTP) and real-time transport protocol
change of routing path is common. Considering (RTCP) are needed to work over UDP. On top
mobility, variable link quality, traffic load, and of RTP/RTCP, rate control protocol (RCP) is also
other factors, the change may be frequent and needed for congestion control.
may cause large variations in RTT. This will de- To date, many RCP protocols have been pro-
grade the TCP performance, because the normal posed for wired networks. They can be classified
operation of TCP relies on a smooth measurement into two types: additive-increase multiplicative-de-
of RTT [1]. How to enhance a TCP so that it is ro- crease (AIMD)-based or equation-based. How-
bust to large RTT variations has not been thor- ever, these protocols are not applicable to
oughly studied for both mobile ad hoc networks wireless networks due to existence of packet errors
and WMNs. and link failures. Thus, differentiation between
losses caused by congestion or wireless channels
10.1.2. Entirely new transport protocols need to be taken into account with RCP. Various
As discussed before, many fundamental prob- loss differentiation algorithms (LDAs) with con-
lems exist in TCP. Therefore, some researchers gestion control are studied in [28], where only
have started to develop entirely new transport pro- one wireless link is considered on the path between
tocols for ad hoc networks. sender and receiver. It is shown in [28] that the hy-
In [127], the ad hoc transport protocol (ATP) is brid LDA is the most effective. However, this re-
proposed for ad hoc networks. Transmissions in sult may not be applicable to WMNs, since
ATP are rate-based, and quick start is used for ini- multiple wireless links are on the path between re-
tial rate estimation. The congestion detection is a ceiver and sender.
delay-based approach, and thus ambiguity be- An analytical rate control scheme is proposed in
tween congestion losses and non-congestion losses [6] for end-to-end transmission of real-time traffic
is avoided. Moreover, in ATP, there is no retrans- over both wired and wireless links. However, the
mission timeout, and congestion control and reli- scheme has not considered factors such as the ad
ability are decoupled. By using an entirely new hoc architecture and mobility as in WMNs. There-
set of mechanisms for reliable data transport, fore, whether this scheme is applicable to WMNs
ATP achieves much better performance (e.g., de- needs to be researched further.
lay, throughput, and fairness) than the TCP To date, few rate control schemes are available
variants. for mobile ad hoc networks. Recently, an adaptive
Despite its advantages, an entirely new trans- detection rate control (ADTFRC) scheme has
port protocol is not favored by WMNs due to been proposed for mobile ad hoc networks in
the compatibility issue. ATP [127] assumes that [49], where an end-to-end multi-metric joint detec-
the wireless network can be stand-alone. While this tion approach is developed for TCP-friendly rate
may be true for mobile ad hoc networks, it is inva- control schemes. However, to really support real-
ARTICLE IN PRESS
time delivery for multimedia traffic, the accuracy tics of these networks may be significantly hetero-
of the detection approach is still insufficient. In geneous due to different network capacity and
addition, all non-congestion packet losses due to behaviors of error control, MAC, and routing pro-
different problems are processed in the same way tocols. Such heterogeneity renders the same TCP
[49]. This may degrade the performance of the rate ineffective for all networks. Applying different
control scheme. TCPs in these networks will make the integration
To date, no RCP has been proposed for WMNs. be complicated and costly. As a consequence, pro-
In addition, no effective RCPs for ad hoc networks posing an adaptive TCP is the most promising
can be adopted and tailored for WMNs. Thus, solution for WMNs. An adaptive transport proto-
RCP for WMNs is a new research area. col is proposed in [7] for an integrated network of
wireless LANs, cellular networks, Internet back-
10.3. Open research issues bone, and satellite networks. However, due to
the hybrid ad hoc and infrastructure architecture,
In addition to the above mentioned open re- an integrated WMN is much different from the
search issues, there exist several other problems integrated network in [7]. Consequently, new
which need investigation. adaptive transport protocols need to be proposed
In order to reduce the impact of network asym- for an integrated WMN.
metry on TCP performance, cross-layer optimiza- For real-time delivery, no existing solution from
tion is a challenging but effective solution, since ad hoc networks can be adopted and tailored for
all problems of TCP performance degradation the use of WMNs. Thus, brand-new RCPs need
are actually related to protocols in the lower lay- to be developed considering the features of
ers. For example, it is the routing protocol that WMNs. In addition, new loss differentiation
determines the path for both TCP data and schemes must be developed to work together with
ACK packets. To avoid asymmetry between data RCPs. Since WMNs will be integrated with vari-
and ACK packets, it is desired for a routing proto- ous wireless networks and the Internet, adaptive
col to select an optimal path for both data and rate control protocols are also needed for WMNs.
ACK packets but without increasing overhead.
We also know that the link layer performance di-
rectly impacts packet loss ratio and network asym- 11. Application layer
metry. Thus, in order to reduce the possibility of
network asymmetry, the MAC layer may need to Applications determine the necessity to deploy
treat TCP data and ACK packets differently. In WMNs. Thus, it is always a key step to find out
addition, error control schemes may need to be en- what existing applications can be supported by
hanced in the MAC layer. WMNs and what new applications need to be
It is also important that the enhanced TCP has developed.
minimal impact on existing TCP. For WMNs, it is
common that a network node will communicate 11.1. Applications supported by WMNs
with other network nodes outside of mesh net-
works such as the Internet. Thus, from an end Since numerous applications can be supported
node to another end node, both wireless and wired by WMNs, it is infeasible to have a complete list
links may exist, which requires the enhanced TCP of them. Here, depending on the functions for
in WMNs and can work together with classical WMNs, we categorize applications of WMNs into
TCPs for wired networks. For example, the inter- several classes:
mediate-layer concept of ATCP [92] can be
adopted for WMNs. • Internet access. Various Internet applications
Besides the Internet, WMNs will also be inte- provide important timely information to peo-
grated with various wireless networks such as ple, make life more convenient, and increase
IEEE 802.11, 802.16, 802.15, etc. The characteris- work efficiency and productivity. For example,
ARTICLE IN PRESS
email, search engine like Google, on-line actions in the lower layers to provide perfect support
like eBay, on-line purchase, chatting, video for the application layer. For example, as per-
streaming, etc., have become an indispensable ceived by the application layer, packet loss
part of life. Thus, people are interested to sub- may not always be zero, packet delay may be
scribe the Internet. In a home or small/medium variable with a large jitter, etc. Such problems
business environment, the most popular net- may fail certain applications working smoothly
work access solution is still DSL or cable in a wired network, especially those with time-
modem along with IEEE 802.11 access points. critical constraints. Therefore, algorithms in
However, comparing with this approach, the application layer must be developed to
WMNs have many potential advantages: lost improve the performance of real-time Internet
cost, higher speed, and easy installation. There- applications over WMNs.
fore, Internet access will greatly motivate the 2. To study application protocols for distributed
development of WMNs. information sharing in WMNs. For example,
• Distributed information storage and sharing for wired networks, application protocols are
within WMNs. For this type of applications, available for peer-to-peer information sharing,
backhaul access to the Internet is not necessary. on-line gaming, etc. However, WMNs have
Users of these applications communicate within much different characteristics than wired net-
WMNs. A user may want to store high-volume works. Whether these protocols achieve a satis-
data in disks owned by other users, download factory performance in WMNs need to be
files from other usersÕ disks based on peer-to- investigated. In case the answer is negative,
peer networking mechanism, and query/retrieve new application protocols need to be developed.
information located in distributed database 3. To discover unique applications that utilize the
servers. Users within WMNs may also want to advantages of WMNs. Such applications must
chat, talk on the video phones, and play games bring tremendous benefits to customers. More-
with each other. To have these applications over, their functions cannot be accomplished
work at the end-users, certain protocols must by other existing networks. In such a way,
exist in the application layer. WMNs will be enabled to be a unique network-
• Information exchange across multiple wireless ing solution instead of just another option of
networks. Again, this type of applications does wireless networking. For example, if wireless
not need backhaul access to the Internet. For sensor networks are integrated with WMNs,
example, when a cellular phone talks to a Wi- software tools can actually be developed for
Fi phone through WMNs, no Internet is needed. users in a home networking environment to
Similarly, a user on a Wi-Fi network may expect remotely monitor, configure, and control all
to monitor the status in various sensors in a electronic devices, which makes home automa-
wireless sensor network. All these applications tion become a reality.
must be supported by new algorithms or soft-
ware in the application layer of the end-users.
12. Protocols for network management
1. To make existing Internet applications work Mobility management consists of two impor-
under the architecture of WMNs. Due to ad tant tasks: location and handoff management
hoc and multi-hop wireless network architec- [10]. Location management handles location regis-
ture of WMNs, there is no way for protocols tration and call delivery, while handoff manage-
ARTICLE IN PRESS
ment is responsible for handoff initiation, new con- ment schemes are closely coupled with MAC
nection generation, and data flow control for call protocols. Moreover, since connectivity affects
handoff. The mobility management schemes devel- performance of a routing protocol, power manage-
oped for cellular [10] or mobile IP networks [11] ment is also crucial for the network layer.
could be useful for WMNs. However, the central- In contrast to mesh routers, mesh clients may
ized scheme is generally not applicable on WMNs expect protocols to be power efficient. For exam-
which are based on distributed and ad hoc archi- ple, some mesh clients are IP phones or even sen-
tecture. Thus, distributed mobility management sors; power efficiency is the major concern for
is a preferred solution for WMNs. Mobility man- them. Thus, it is quite possible that some applica-
agement schemes of ad hoc networks are mainly tions of WMNs require power management to
comprised of two types: distributed [56] and hier- optimize both power efficiency and connectivity,
archical mobility management [125,30]. These which results in a complicated problem.
schemes may not perform well for WMNs due to
the specific features of WMNs. More specifically, 12.3. Network monitoring
the backbone of WMNs does not have high mobil-
ity as mobile nodes in ad hoc networks, but con- Many functions are performed in a network
nections between all mesh routers are wireless. management protocol. The statistics in the MIB
Mesh clients may constantly roam across different (management information base) of mesh nodes,
mesh routers. These features also render the mobil- especially mesh routers, need to be reported to
ity management schemes for cellular networks one or several servers in order to continuously
ineffective for WMNs. As a result, new mobility monitor the network performance. Data process-
management schemes need to be developed for ing algorithms in the performance monitoring
WMNs. software on the server analyze these statistical data
Location service is a desired feature in WMNs. and determine potential abnormality. In case any
Location information can enhance the perfor- abnormal symptom is detected, the server reacts
mance of MAC and routing protocols. It can help to take responses, e.g., triggering an alarm. Based
to develop promising location-related applica- on the statistical information collected from MIB,
tions. Proposing efficient algorithms for location data processing algorithms can also accomplish
service is still an open research topic. many other functions such as network topology
Mobility management is closely related to mul- monitoring. The network topology of WMNs is
tiple layers of network protocols. The development not always fixed due to mobility in mesh clients
of multi-layer mobility management schemes as in or possible failures in some mesh routers. Thus,
[43] is an interesting topic. monitoring the network topology is a desired fea-
ture for WMNs.
12.2. Power management A few network management protocols [121]
have been proposed for ad hoc networks. How-
The goal of power management for WMNs var- ever, the efficiency of these schemes needs to be im-
ies. Usually, mesh routers do not have a constraint proved for a large scale mesh network. In addition,
on power consumption; power management aims in order to accurately detect abnormal operation
to control connectivity, interference [86], spectrum of WMNs, effective data processing algorithms
spatial-reuse, and topology [91]. If a single channel are needed. Also, how to quickly determine net-
is used in each network node, the interference work topology is still an open question.
among the nodes directly impacts the spectrum
spatial-reuse factor. Reducing transmission power
level decreases the interference and increases the 13. Security
spectrum spatial-reuse efficiency [86]. However,
more hidden nodes may cause performance degra- Security is always a critical step to deploy
dation in MAC protocols. Thus, power manage- and manage WMNs. Virtual private networking
ARTICLE IN PRESS
(VPN) is possible over wireless LANs. It is usually tional exchange, must be used. Rational exchange
implemented with standard key encryption algo- ensures that a misbehaving party cannot gain any-
rithms for tunneling such as IPSec to provide se- thing from misbehavior, and thus, will not have
cure virtual paths along the shared networks. any incentives to misbehave [26].
Security in terms of authentication and authori- The key management is one of the most impor-
zation is not a big issue for wireless LANs; some tant tasks for network security. However, the key
wireless LAN commercial system implementations management for WMNs becomes much more dif-
provide authentication, authorization, and ficult, because there is no central authority, trusted
accounting (AAA) services directly over the wire- third party or server to manage security keys. Key
less LAN access point or via gateways to take care management in WMNs need to be performed in a
of this issue. AAA is usually performed through a distributed way. A self-organization scheme was
centralized server such as RADIUS (remote proposed in [63] to distribute and manage the secu-
authentication dial-in user service). However, the rity keys. In this self-organizing key management
centralized scheme is not scalable in WMNs. Sim- system, certificates are stored and distributed by
ilar to mobile ad hoc networks, WMNs still lack users themselves. When the public keys of two
efficient and scalable security solutions because users need to be verified, they first merge the local
their security is easier to be compromised due to certificate repositories and then find the appropri-
[143,25]: vulnerability of channels and nodes in ate certificate chains within the merged reposito-
the shared wireless medium, absence of infrastruc- ries that can pass this verification.
ture, and dynamic change of network topology. To enhance security of WMNs, two strategies
The attacks may advertise routing updates in [60] need to be adopted. Either to embed security
and [145] for DSR and AODV, respectively. An- mechanism into network protocols such as secure
other type of attacks is packet forwarding, i.e., routing and MAC protocols or to develop security
the attacker may not change routing tables, but monitoring and response systems to detect attacks,
the packets on the routing path may be lead to a monitor service disruption, and respond quickly to
different destination that is not consistent with attacks. To date, many secure protocols have been
the routing protocol. Moreover, the attacker proposed [60,145]. However, their role of defend-
may sneak into the network, and impersonate a ing attacks is very limited, because schemes located
legitimate node and does not follow the re- in a single protocol layer cannot solve problems in
quired specifications of a routing protocol [37]. other layers. However, security attacks in a net-
Some malicious nodes may create wormhole and work may come simultaneously from different pro-
shortcut the normal flows among legitimate nodes tocol layers. Thus, a multi-protocol layer security
[61]. scheme is desired for network protocols.
Same types of attacks as in routing protocols For a security monitoring system, a cross-layer
may also occur in MAC protocols. For example, framework also needs to be developed. A frame-
the backoff procedures and NAV for virtual car- work of intrusion detection in ad hoc networks is
rier sense of IEEE 802.11 MAC may be misused proposed in [146]. However, how to design and
by some attacking nodes, which cause the network implement a practical security monitoring system,
to be always congested by these malicious nodes including cross-layer secure network protocols and
[55]. various intrusion detection algorithms, is a chal-
Attackers may sneak into the network by mis- lenging research topic.
using the cryptographic primitives [22]. In a cryp-
tographic protocol, the exchange of information
among users occurs frequently. The users employ 14. Timing synchronization
a fair exchange protocol which depends on a
trusted third party. However, this trusted party is Timing synchronization is critical for the per-
not available in WMNs due to lack of infrastruc- formance of network protocols, e.g., power man-
ture. Thus, another exchange scheme, called ra- agement schemes and MAC protocols.
ARTICLE IN PRESS
A power management scheme usually depends have to work together with the physical layer. In
on the synchronized time to determine when a net- addition, MAC, routing, and transport protocols
work node needs to be turned on/off or to enter also need to work collaboratively among them-
sleep mode according to certain performance crite- selves. Such interactions demand a cross-layer de-
ria. For WMNs, TDMA MAC protocols are usu- sign among different protocols.
ally difficult to implement because network nodes In WMNs, because of the ad hoc feature, net-
are not accurately synchronized. Without enough work topology constantly changes due to mobility
accuracy in timing synchronization, a large per- and link failures. Such dynamic network topology
centage of bandwidth must be used as guard time impacts multiple protocol layers. Thus, in order to
in TDMA MAC. improve protocol efficiency, cross-layer design be-
Synchronization among nodes has been speci- comes indispensable. For instance, a MAC proto-
fied as timing synchronization function (TSF) in col for WMNs may include a mechanism for
IEEE 802.11 for both infrastructure and ad hoc network topology control and self-organization.
modes. To improve performance of the standard Such information can be directly shared by a rout-
synchronization scheme, especially the scalability, ing protocol. To avoid broadcast storming in a
some solutions have been proposed in [62,87]. routing protocol, the underlying MAC protocol
However, the TSF in IEEE 802.11 standard and can optimize the procedure of transmitting signal-
in [62,87] cannot be directly applied to large ing messages initiated by routing protocols.
WMNs, because the synchronization information Cross-layer design can be performed in two
may take a long time. ways. The first way is to improve the performance
To date, timing synchronization for WMNs of a protocol layer by considering parameters in
lacks an effective approach to achieve satisfactory other protocol layers. Typically, parameters in
accuracy. Thus, it is necessary to develop better the lower protocol layers are reported to higher
power management schemes or MAC protocols layers. For example, the packet loss rate in the
that do not demand high accuracy of timing MAC layer can be reported to the transport layer
synchronization. so that a TCP protocol is able to differentiate con-
gestion from packet loss. As another example, the
physical layer can report the link quality to a rout-
15. Cross-layer design ing protocol as an additional performance metric
for the routing algorithms. The second way of
Traditionally, different protocol layers are re- cross-layer design is to merge several protocols
quired to be transparent from each other. This into one component. For example, in ad hoc net-
makes the protocol development and implementa- works, MAC and routing protocols can be com-
tion be a simple and scalable process. However, bined into one protocol in order to closely
the methodology of layered protocol design does consider their interactions. The advantage of the
not necessarily lead to an optimum solution for first way is that it does not totally abandon the
wireless networks. For example, the physical chan- transparency between protocol layers, while the
nel in a wireless environment is variable in terms of second way will totally lose this advantage. How-
capacity, bit error rate, etc. Although different ever, the second way can achieve much better per-
coding, modulation, and error control schemes formance by considering an optimized interaction
can be used to improve the performance of the between protocol layers.
physical channel, there is no way to guarantee Cross-layer design can significantly improve
fixed capacity, zero packet loss rate, or reliable network performance [20,84,32]. However, certain
connectivity as expected by higher layers. There- issues must be considered when carrying out cross-
fore, higher layer protocols will be inevitably af- layer protocol design [77]: cross-layer design have
fected by the unreliable physical channel. To risks due to loss of protocol layer abstraction,
further improve the performance of a wireless net- incompatibility with existing protocols, unforeseen
work, MAC, routing, and transport protocols impact on the future design of the network, and
ARTICLE IN PRESS
difficulty in maintenance and management. Thus, number of neighbors, but the radio link to each
certain guidelines need to be followed [77]. neighbor is often of marginal quality; finding the
best multi-hop routes through a rich mesh of mar-
ginal links turns out to be a challenge. The average
16. Testbeds and implementations TCP throughput and latency of all Roofnet nodes
to their nearest gateway were measured in April
Numerous testbeds have established to carry 2004. When 1 hop is considered for 18 nodes, the
out research and development for WMNs. average throughput and latency are 357.2 kbytes/
s and 9.7 ms. However, when 4 hops are consid-
16.1. Academic research testbeds ered for 7 nodes, the average throughput is only
47.3 kbytes/s and the average latency is 43.0 ms.
One of the earliest mesh network testbeds is The low multihop throughput reflects the problem
Carnegie-Mellon UniversityÕs mobile ad hoc net- typical in all 802.11 MAC based multihop
work testbed [95]. It consists of seven nodes: two networks.
stationary nodes, five car mounted nodes that University of Illinois at Urbana-Champaign has
drive around the testbed site, and 1 car mounted reported a 4-node multi-channel 802.11b testbed
roving node that enters and leaves the site. Packets [119]. Each node is equipped with 2 cards whose
are routed between the nodes using the DSR pro- channels were determined based on the load-aware
tocol [73] which also integrates the ad hoc network channel assignment algorithm. The multi-channel
into the Internet via a gateway. They experimented network achieves 2.63 times the throughput as
with the network behavior under different levels of compared to the single channel network. The num-
traffic load, including audio and video streams, ber of non-overlapping channels in 802.11b stan-
and designing protocol enhancements to provide dard, i.e., 3, is the limiting factor for this
these streams with QoS promises. Interesting re- performance. The performance, however, does
sults were observed in the experiments: not reach 3 times of the single-channel network
performance because of the inter-channel interfer-
• Local (i.e., link layer) retransmission algorithms ence that cannot be completely eliminated. Sub-
is a critical part of any multihop ad hoc net- stantial interference was observed between two
work. If the retransmission algorithms imple- 802.11b cards placed on the same machine despite
mented above the link layer, it must be operating on non-overlapping channels. In addi-
adaptive in order to accommodate network tion, the degradation due to inter-channel interfer-
congestion and periods of high contention in ence was found independent of the guard band.
the wireless channel. One way to reduce the interference is to use USB
• Delivering routing protocol control packets as cards instead of PCI/PCMCIA cards and place
rapidly as possible is important for high end- them side-by-side in similar configuration as in
to-end performance, and this implies that pack- Orinoco AP-1000 access points. Another possibil-
ets with routing implications should be sched- ity is to equip cards with external antennas and
uled for transmission ahead of users data place the external antennas slightly away from
packets. each other. Yet another option is to use the
upcoming Engim chipsets which solve the interfer-
MITÕs Roofnet is an experimental multi-hop ence problem at RF-level [44].
802.11b mesh network [4,5]. It consists of about California Institute for Telecommunications
50 wireless nodes to interconnect the Ethernet net- and Information Technology (Cal-(IT)2) has
works (with Internet gateways) in apartments in established a basic Wi-Fi MAC development plat-
Cambridge, MA. A primary feature of Roofnet form called CalRADIO-I [132]. This is a Wi-Fi re-
is that it requires no configuration or planning. search and development device that consists of a
One consequence of an unplanned network is that TI 5410 DSP, a 16-bit stereo CODEC, external
each node can route packets through any of a large Flash and SRAM memories, and support of a
ARTICLE IN PRESS
RF LAN module. It provides a convenient plat- The Broadband and Wireless Network (BWN)
form for development of RF radios from the phys- Lab at Georgia Institute of Technology has re-
ical layer up to the application layer. The key cently built a testbed of WMNs, as shown in Fig.
benefit of the board is that all aspects of the 13. The WMN, called BWN-Mesh, consists of 15
MAC are coded in C language and, therefore, IEEE 802.11b/g based mesh routers, among which
are altered easily for research in queueing, security, several of them are connected to the next genera-
power management, MIMO, cognitive radio, and tion Internet testbed (also available in the BWN
so forth. It also utilizes basic Symbol Technolo- Lab) as backhaul access to the Internet. The test-
giesÕ Wi-Fi test board as the base for modifying bed consists of laptops and desktops equipped
board and re-spinning to incorporate new features. with IEEE 802.11b and IEEE 802.11g cards lo-
CalRADIO-I is evolving into a CalRADIO-II de- cated in various rooms on the floor where the
velop platform with basic DSP board and multiple BWN Lab resides. By changing the topology of
RF front-end modules such as 802.11x, 802.16, cell the network, experiments investigating the effects
and general RF. This will eventually allow the of inter-router distance, backhaul placement and
capability of publishing standards in software/ clustering are performed along with mobility
firmware and hardware. experiments using the laptops in the testbed.
Moreover, experiments with existing protocols layer software runs unmodified over the ad hoc net-
(i.e., TCP, AODV, and IEEE 802.11g as transport, work. In MicrosoftÕs testbeds, both IPv4 and IPv6
routing, and MAC protocols) for BWN-Mesh test- can run over the ad hoc network. No modification
bed have demonstrated that these protocols do not to either network stack is required. Network layer
perform well in terms of end-to-end delay and functionality (for example ARP, DHCP, and
throughput in WMNs. Currently, the research is Neighbor Discovery) works fine. Second, the ad
focused on adaptive protocols for transport layer, hoc routing runs over heterogeneous link layers.
routing and MAC layers and their cross-layer de- MicrosoftÕs implementation supports Ethernet-like
sign [24]. These protocols are developed and eval- physical link layers (e.g., IEEE 802.11 and 802.3)
uated on the BWN-Mesh testbed. The approaches but the architecture accommodates link layers with
being explored in the BWN lab are not merely lim- arbitrary addressing and framing conventions. The
ited to Wi-Fi mesh networks but also applicable virtual MCL network adapter can multiplex sev-
for wireless sensor networks (WSNs) and wireless eral physical network adapters, so that the ad hoc
sensor and actor networks (WSANs). Thus, the network can extend across heterogeneous physical
BWN-Mesh testbed is integrated with the already links. Third, the design can support other ad hoc
existing BWN Sensor Network Testbed, which routing protocols as well.
consists of MICA motes, with TinyOS distributed A variety of research and development at Intel
software operating system, and light, temperature, are geared toward understanding and addressing
acoustic actuator, magnometer, and accelerometer the technical challenges of multi-hop mesh net-
sensors. In align with this effort, BWN Lab is also works. Early work at the Intel Research-Berkeley
trying to integrate the current Wi-Fi mesh net- Lab, affiliated with the University of California,
works with other wireless networks such as Wi- Berkeley, has resulted in small sensor motes [70]
MAX. Consequently, this integrated testbed will which form self-configuring, low-cost adaptive
enable the design and evaluation of protocols networks. Additional work within IntelÕs Network
applicable to heterogeneous wireless networks Architecture Lab is aimed at overcoming many of
including WMNs, next generation Internet, the challenges faced by mesh networks. Low-cost
WSNs, WSANs, and WiMAX. and low-power access point prototypes, or nodes,
have been developed to enable further research
16.2. Industrial practice on security, traffic characterization, dynamic rout-
ing and configuration, and QoS problems. A dem-
Microsoft Research Lab (MSR) implements ad onstration was discussed in various occasions [71],
hoc routing and link quality measurement in a soft- consisting of a collection of Centrino laptop com-
ware module called the mesh connectivity layer puters and IXP425 network processor based rou-
(MCL) [101]. Architecturally, MCL is a loadable ters running AODV and 802.11b MAC
Windows driver. It implements a virtual network protocols. The testbed results confirm that 802.11
adapter, so that the ad hoc network appears as MAC limits full exploitation of multihop through-
an additional (virtual) network link to the rest of put. As a means to enhance multihop throughput,
the system. MCL routes by using a modified ver- it advocates spatial-reuse through carrier sensing
sion of DSR called LQSR. MCL is a routing pro- threshold tuning [53]. Another potential solution
tocol well-suited for low mobility, unconstraint experimented was the concept of heterogeneous
power consumption and small diameter networks. networks: an 802.11 mesh network comprised of
The MCL driver implements an interposition layer four high-end nodes, such as Intel XScale based
between the link layer and the network layer. To nodes, is overlaid on a 50-sensor node (motes) net-
higher layer software, MCL appears to be just an- work scattered throughout a large conference
other Ethernet link, albeit a virtual link. To lower room. Sensor data can enter and exit the 802.11
layer software, MCL appears to be just another backbone at multiple interchanges (the XScale
protocol running over the physical link. This design nodes) in order to bypass the intermediate sensors.
has several significant advantages. First, higher This should enable faster trips across the network
ARTICLE IN PRESS
and result in improved performance since the num- tive routing techniques. To adapt the routing pro-
ber of nodes that the data has to pass through is tocol to a given radio platform, adaptive
much lower, leading to improved reliability and transmission protocol (ATP) is implemented to
lower energy consumption [70]. This experiment tightly bind the scalable routing protocol to the
has shown that, without the top level mesh net- underlying radio platform. MeshNetworks pro-
work (XScale nodes), the average data age dou- vides a software-only overlay solution that lets na-
bled, from approximately 10 s to 20–25 s, and the tive 802.11b clients in existing networks work in
network lifetime increased by 20%. mesh-mode. While it will not add any mobile
NortelÕs commercial roll out of the WMN prod- broadband capabilities beyond what 802.11b can
ucts [106] includes wireless access point (WAP) already support, it will extend the range and link
which is a dual radio system supporting a robustness of existing Wi-Fi networks by allowing
2.4 GHz access link and a 5 GHz transit link, mesh-style multi-hopping. Security features in
equipped with smart antennas. Along with Nor- MeshNetworksÕ QDMA-based systems include a
telÕs wireless gateway routers and a network man- hardware firewall on a chip that makes it impossi-
agement system, WAPs can be used to wirelessly ble for a client to access somebody elseÕs packets.
backhaul data traffic to the wired network or serv- This feature will not be available in retrofitted
ers for an enterprise or carrier network. The WAP 802.11 networks using the software overlay tech-
transit link employs an 802.11a standard physical nology. However, other approaches such as route
layer, and an inter-AP meshing protocol above diversity will help resolve security issues.
the MAC layer for transit link discovery, establish- Tropos Networks employs a cellular Wi-Fi net-
ment, monitoring, maintenance, and re-establish- work architecture to support ‘‘infrastructure
ment. It also performs automatic assignment of mesh’’ networking [129], using its a layer-3 net-
channels according to local conditions and access work operating system (NOS) called Tropos
point neighborhood. The transit link uses an ele- Sphere, that runs on standard 802.11 hardware
vated dual-polar antenna with multiple degrees and software. Tropos Sphere operates on every
of freedom: beam, frequency, polarization, and (small sized) Tropos Wi-Fi cell and contains the
burst time. Having multiple beams alleviates key communications, path selection, and security
deployment difficulties. The WAP access link uses functions that allow the Wi-Fi cells to inter-oper-
an 802.11g standard physical layer with an ele- ate and form a completely wireless network like
vated, dual-polar, diversity switched antennas. Ac- a wireless routed LAN. Tropos uses a lightweight
cess Link coverage is typically less than the reach control protocol for supporting a large number
of a transit link. The user devices use standard of Wi-Fi cells. It uses a proprietary predictive path
802.11b/g NIC and software to access the network. optimization protocol to improve end-user
The Nortel mesh network is a sophisticated, car- throughput and continuously optimize perfor-
rier class quality but expensive system. Multiple mance to compensate for the changing RF envi-
field trials have been reported since late 2003. ronment. This protocol is called predictive
MeshNetworksÕ initial attempt to commercializ- wireless routing protocol (PWRP), which is analo-
ing mesh technology was its quadrature division gous to traditional wired routing protocols such as
multiple access (QDMA) radio platform. [99]. open shortest path first (OSPF). However, PWRP
The QDMA radio is designed for mobile ad hoc does not use routing tables or rely on hop-count
broadband networking. It uses multi-channel only to select transmission paths. Rather, it com-
MAC and PHY to overcome the effects of Doppler pares packet error rates and other network condi-
shifting, rapid Raleigh fading and multipath tions to determine the best path at a given
encountered in a mobile system. The MeshNet- moment. Since the system is largely a layer-3 solu-
worksÕ scalable routing protocol [100] is imple- tion that relies on the standard 802.11 MAC pro-
mented above QDMA radios. The scalable tocol for a large mesh network, many of the
routing technology utilizes a hybrid ad hoc routing throughput performance impairments remain
algorithm that combines both proactive and reac- unresolved.
ARTICLE IN PRESS
PacketHopÕs core technology was developed in function router. Without KiyonÕs software, a client
the SRI International lab [107]. It consists of a net- device running standard 802.11 station software
work controller, performing gateway, QoS, secu- can originate and terminate traffic. Mobility of cli-
rity, and roaming functions, a network ent devices is supported in both cases. The third
management system, and the Windows software option is a form of hierarchical network, in which
for ad hoc mesh routing. This is largely a layer-3 a number of standard 802.11 access points serve as
solution that runs on 802.11 and multi-mode broad- the access layer for client devices. Each of these ac-
band radios. PacketHop is in collaboration with cess points is attached to one of the backbone rou-
Nortel to complement NortelÕs ‘‘infrastructure ters via an Ethernet connection.
mesh’’ solution with its ad hoc meshing capability.
Kiyon is in conduction field trials with building
automation and ‘‘small office home office’’ 17. Standard activities
(SOHO) customers of its broadband wireless mesh
routers [80]. Its technology is a layer-2/3 solution 17.1. IEEE 802.11 mesh networks
that implements a hybrid CSMA/CA and distrib-
uted TDMA MAC protocol atop an 802.11g/a Currently, IEEE 802.11 wireless networks can
physical layer. This is tightly coupled with a mul- achieve a peak rate of 11 Mbps (802.11b) and
ti-metric ‘‘Attribute Routing’’ protocol, aiming at 54 Mbps (802.11a/g). Also under development is
high and steady multihop throughput in a mesh a high-bandwidth extension to the current Wi-Fi
network. As the new generation 802.11 radios standard. Researchers expect 802.11n to increase
adopt the soft MAC approach, e.g., Atheros, the speed of Wi-Fi connections by 10–20 times.
Broadcom and more recently Intel, KiyonÕs en- Although many home users will not benefit from
hanced MAC/routing protocols can be imple- the additional speed right away, because of limita-
mented in host software and downloaded into tions on their cable or DSL connections, enter-
these standard 802.11 chipsets at runtime. Kiyon prises are hoping the technology will allow them
supports both infrastructure and client mesh, sta- to reduce the burden of laying and maintaining
tionary or mobile wireless networks. Ethernet cabling throughout the building.
In a Kiyon mesh network, a number of routers There are many academic testbeds and commer-
form a broadband backbone of the network. Each cial deployment of mesh networks using IEEE
of the routers is equipped with KiyonÕs routing 802.11 wireless LAN technology. However, mesh
and MAC protocol software plus a standard IEEE networking is at the same stage as wireless LANs
802.11g/a radio. For client access to the broad- were in the early 1990s; they are expensive and
band backbone, several options can be adopted. proprietary. To become commoditized, the eco-
The first option is called the ‘‘wired host route’’, nomic pressures are driving the standard pro-
in which a client can connect to the backbone via cesses. Furthermore, protocols for 802.11 ad hoc
an Ethernet connection. Any IP capable devices mode are insufficient for multi-hop and mesh net-
(e.g., a RFID reader, BACnet controller or data- working, because of lack of scalability in the MAC
base server) can be connected to the wireless net- protocol, resulting in poor network performance.
work this way. No software modification on the A working group within IEEE 802.11, called
client is required. The second option is called the 802.11s, has been formed recently to standardize
‘‘wireless host route’’, in which a client device con- the extended service set (ESS) [57]. 802.11s aims
nects to the wireless network via a wireless LAN to define MAC and PHY layers for mesh networks
interface such as an 802.11 b/g/a NIC. In this that extended coverage with no single point of fail-
arrangement, the client is ‘‘homed’’ on one of the ure. In such networks, 802.11 cellular wireless
routers in the network that satisfies the defined LAN access points relay information from one to
routing metric, such as signal strength. The client another in a router-like hop-by-hop fashion.
has the option to run KiyonÕs software or not. As users and access points are added, the capac-
With KiyonÕs software, a client becomes a full ity increases, as in the Internet, giving rise to a
ARTICLE IN PRESS
scalable and redundant architecture. Early discus- personal area networking (PAN) that has commu-
sions in this working group include definition of nication distances of around 10 m (or less), with
WMNs, usage cases, QoS, architecture specifica- applications in home networking space, with
tions, security, routing protocols, and develop- imminent wireless extensions to USB, IEEE
ment of new MAC protocols. Several task 1394, and with the capability to address the con-
groups have been formed to tackle these issues. vergence of PC, consumer electronics and IP mo-
Wi-Fi mesh networking can be implemented in bile phones. Vendors planning to produce
two basic modes: infrastructure and client mesh- 802.15.3a products have formed the WiMedia Alli-
ing. The former is an infrastructure ESS mesh, in ance [138], a branding and testing organization
which access points are interconnected through that will certify standards compliance.
wireless links that enable automatic topology UWB networks hold many advantages over
learning and dynamic path configuration. Clients other wireless networks, such as covert communi-
are associated with access points and need not be cations, low power and cost requirement, accurate
aware of the mesh. Infrastructure meshing creates location information, and extra high bandwidth.
wireless backhaul mesh among access points or However, the communication range is rather
wireless routers. This reduces system backhaul short. Mesh networks have been predicted to be
costs while increasing network coverage and reli- the killer application for UWB radio systems. A
ability. To provide an IEEE 802.11 ESS Mesh, new MAC proposed by MBOA, which deviates
802.11s will define an architecture and protocol substantially from the original IEEE 802.13a
based on the IEEE 802.11 MAC to create an IEEE MAC proposal, has added strong support for
802.11 wireless distribution system (WDS) that mesh networking and mobility, paving the way
supports both broadcast/multicast and unicast for UWB to enter the enterprise network. The
delivery at the MAC layer using radio-aware met- MBOA MAC uses piconet structure, combined
rics over self-configuring multi-hop topologies. with a decentralized resource-handling ability to
From the view of access points, the infrastructure allow for the reservation of timeslots for
meshing also forms an ad hoc network among ac- 802.15.3-like TDMA for high priority connections
cess points. The other mode of meshing, i.e., client requiring determinism while assigning contention-
meshing, is a layer-3 ad hoc IBSS (independent ba- based, best-effort access periods.
sic service set), in which all devices operate in ad IEEE 802.15.4 is intended for telemetry with low
hoc mode in a flat network, using IP routing. data rate, long battery life and low device cost
There is no distinction between access points and requirements. The ZigBee Alliance [148] is develop-
client. Client meshing enables wireless peer-to-peer ing higher-level protocols that will run over
networks to form between and among client de- 802.15.4 MAC and PHY layers that operate in
vices and does not require any network infrastruc- unlicensed bands worldwide. Raw data rates of
ture to be present. In this case, clients can hop 250 Kbps can be achieved at 2.4 GHz (16 chan-
through each other to reach other clients in the nels), 40 Kbps at 915 MHz (10 channels), and
network. To maximize the benefit that meshing 20 Kbps at 868 MHz (1 channel). The transmission
can offer, both modes should be supported simul- distance is expected to range from 10 to 75 m,
taneously and seamlessly in a single network. depending on power output and environmental
characteristics. The ZigBee network layer supports
17.2. IEEE 802.15 mesh networks multiple network topologies including star, cluster
tree, and mesh. In a mesh topology, a special node
IEEE 802.15.3a standard [65] is based on Mul- called coordinator is responsible for starting the
tiBand OFDM Alliance (MBOA)Õs physical layer network and for choosing key network parameters.
that uses ultra wide band (UWB) to reach up to The routing algorithm uses a request–response pro-
480 Mbps. A competing proposal of a Direct Se- tocol to eliminate sub-optimal routing.
quence-UWB (DS-UWB) claims support for up Recently a new working group, i.e., IEEE
to 1.3 Gbps. It is intended for high throughput 802.15.5, is established to determine the necessary
ARTICLE IN PRESS
mechanisms in the physical and MAC layers to en- • It is assumed no interference between nodes
able mesh networking [65] in wireless PANs. that are two hops away. Thus, the 802.16 mesh
suffers from the hidden terminal problem.
17.3. IEEE 802.16 mesh networks
To enhance the 802.16 mesh, several proposals
While IEEE 802.11 networks fulfill the need for have been submitted to the standard committee
data services in a local area (i.e. last several hun- [19]. A group within 802.16, the Mesh Ad Hoc
dreds of feet), IEEE 802.16 aims at serving the committee, is investigating ways to improve the
broadband wireless access in metropolitan area performance of mesh networking. It is understood
networks [66] (i.e., last mile), supporting point- that only a small amount of meshing is required to
to-multipoint connection oriented QoS communi- see a large improvement in the coverage of a single
cations to extend fiber optic backbones. The origi- base station. More importantly, the following is-
nal 802.16 standard operates in the 10–66 GHz sues are considered in specifying the 802.16 mesh
frequency band and requires line-of-sight towers. MAC protocol:
The 802.16a extension, ratified in January 2003,
uses a lower frequency of 2–11 GHz, enabling non- • Avoiding hidden terminal collisions,
line-of-sight connections. With 802.16a, carriers • Selection of links,
will be able to connect more customers to a single • Synchronization,
tower and substantially reduce service costs. To al- • Power versus data rate tradeoffs, and
low consumers to connect to the Internet while • Greater routing-MAC interdependence.
moving at vehicular speeds, researchers are devel-
oping an extension to IEEE 802.16 standard called
802.16e. 18. Conclusion
WiMAX is the commercialization of the matur-
ing IEEE 802.16 standard. WiMAX antennas will The capability of self-organization in WMNs
be able to beam high-speed Internet connections to reduces the complexity of network deployment
homes and businesses miles away, eliminating the and maintenance, and thus, requires minimal up-
need for every building to be wired to the Internet. front investment. The backbone of WMNs pro-
WiMAX is ideal for many neighborhoods that are vides a viable solution for users to access the
too remote to receive Internet access via cable or Internet anywhere anytime. It can also enhance
DSL, and for anyplace where the cost of laying the reliability of the mobile ad hoc network of
or upgrading landlines to broadband capacity is mesh clients. WMNs enable the integration of
prohibitively expensive. In areas with cable or multiple wireless networks.
DSL access, WiMAX will provide consumers with WMNs can be built up based on existing tech-
an additional, and possibly cheaper, alternative. nologies. Some companies already have products
Along with the IEEE 802.16 standard, WiMAX for sale, while other companies have started to de-
is envisioned to provide long distance both line ploy WMNs in various application scenarios.
of sight and non-line of sight (NLOS) broadband However, field trials and experiments with existing
wireless access (BWA). WMNs prove that the performance of WMNs is
The 802.16 mesh in the current standard draft still far below what they are expected to be. As ex-
has several limitations: plained throughout this paper, many open re-
search issues need to be resolved:
• The 802.16 mesh has limited scalability. The
mesh can only support around 100 subscribers • Scalability. Based on existing MAC, routing,
due to centralized scheduling message structures. and transport protocols, the network perfor-
• The 802.16 mesh is based on a connectionless mance, indexed by throughput, end-to-end
MAC, so QoS of real-time services is difficult delay, and fairness, is not scalable with either
to guarantee [40]. the number of nodes or the number of hops in
ARTICLE IN PRESS
[12] S.M. Alamouti, A simple transmit diversity technique for [26] L. Buttyan, J.-P. Hubaux, Rational exchange—a formal
wireless communications, IEEE Journal on Selected model based on game theory, in: 2nd International
Areas in Communications 16 (8) (1998) 1451–1458. Workshop on Electronic Commerce, November 2001.
[13] J.B. Andersen, Array gain and capacity for known [27] F. Cali, M. Conti, E. Gregori, Dynamic tuning of the
random channels with multiple element arrays at both IEEE 802.11 protocol to achieve a theoretical throughput
ends, IEEE Journal on Selected Areas in Communica- limit, IEEE/ACM Transactions on Networking 8 (6)
tions 18 (11) (2000) 2172–2178. (2000) 785–799.
[14] P. Bahl, R. Chandra, J. Dunagan, SSCH: slotted seeded [28] S. Cen, P.C. Cosman, G.M. Voelker, End-to-end differ-
channel hopping for capacity improvement in IEEE entiation of congestion and wireless losses, IEEE/ACM
802.11 ad hoc wireless networks, in: ACM Annual Transactions on Networking 11 (5) (2003) 703–717.
International Conference on Mobile Computing and [29] K. Chandran, S. Raghunathan, S.R. Prakash, A feed-
Networking (MOBICOM), 2004, pp. 216–230. back-based scheme for improving TCP performance in ad
[15] H. Balakrishnan, V.N. Padmanabhan, R.H. Katz, Net- hoc wireless networks, IEEE Personal Communications 8
work asymmetry: the effects of asymmetry on TCP (1) (2001) 34–39.
performance, Mobile Networks and Applications 4 [30] W.-T. Chen, P.-Y. Chen, Group mobility management in
(1999) 219–241. wireless ad hoc networks, in: IEEE Vehicular Technology
[16] E.M. Belding-Royer, P.M. Melliar-Smith, L.E. Moser, Conference, Fall 2003, pp. 2202–2206.
An analysis of the optimum node density for ad hoc [31] P. Chevillat, J. Jelitto, A.N. Barreto, H.L. Truong, A
mobile networks, in: IEEE International Conference on dynamic link adaptation algorithm for IEEE 802.11a
Communications (ICC), vol. 3, June 2001, pp. 857–861. wireless LANs, in: IEEE International Conference on
[17] E.M. Belding-Royer, Multi-level hierarchies for scalable Communications (ICC), 2003, pp. 1141–1145.
ad hoc routing, ACM/Kluwer Wireless Networks 9 (5) [32] M. Chiang, To layer or not to layer: balancing transport
(2003) 461–478. and physical layers in wireless multihop networks, in:
[18] S. Bellofiore, J. Foutz, R. Govindaradjula, I. Bahceci, IEEE Annual Conference on Computer Communications
C.A. Balanis, A.S. Spanias, J.M. Capone, T.M. Duman, (INFOCOM), 2004, pp. 2525–2536.
Smart antenna system analysis, integration and perfor- [33] C.F. Chiasserini, R.R. Rao, A distributed power man-
mance for mobile ad hoc networks (MANETs), IEEE agement policy for wireless ad hoc networks, in: IEEE
Transactions on Antennas and Propagation 50 (5) (2002) Wireless Communications and Networking Conference
571–581. (WCNC), 2000, pp. 1209–1213.
[19] D. Beyer, N.V. Waes, K. Eklund, Tutorial: 802.16 MAC- [34] I. Chlamtac, M. Conti, J. Liu, Mobile ad hoc networking:
Layer Mesh Extensions, IEEE 802.16 Standard Group imperatives and challenges, Ad Hoc Networks 1 (1) (2003)
Discussions, February 2002. 13–64.
[20] R. Bhatia, M. Kodialam, On power efficient communica- [35] L.-U. Choi, K.B. Letaief, R.D. Murch, MISO CDMA
tion over multi-hop wireless networks: joint routing, transmission with simplified receiver for wireless commu-
scheduling, and power control, in: IEEE Annual Confer- nication handsets, IEEE Transactions on Communica-
ence on Computer Communications (INFOCOM), 2004, tions 49 (2001) 888–898.
pp. 1457–1466. [36] R.R. Choudhury, X. Yang, R. Ramanathan, N.H.
[21] S.D. Blostein, H. Leib, Multiple antenna systems: their Vaidya, Using directional antennas for medium access
role and impact in future wireless access, IEEE Commu- control in ad hoc networks, in: ACM Annual Interna-
nications Magazine (2003) 94–101. tional Conference on Mobile Computing and Networking
[22] N. Borisov, I. Goldberg, D. Wagner, Intercepting mobile (MOBICOM), 2002, pp. 59–70.
communications: the insecurity of 802.11, in: ACM [37] K. Sanzgiri, B. Dahill, B.N. Levine, C. Shields, E.M.
Annual International Conference on Mobile Computing Belding-Royer, A secure protocol for ad hoc networks, in:
and Networking (MOBICOM), September 2002, pp. 180– IEEE International Conference on Network Protocols
188. (ICNP), 2002, pp. 78–87.
[23] P. Bose et al., Routing with guaranteed delivery in ad hoc [38] S. Datta, I. Stojmenovic, J. Wu, Internal node and
wireless networks, in: 3rd ACM International Workshop shortcut based routing with guaranteed delivery in
on Discrete Algorithms and Methods for Mobile Com- wireless networks, in: Proceedings of the IEEE Interna-
puting and Communications, August 1999, pp. 48–55. tional Conference on Distributed Computing and Sys-
[24] BWN lab wireless mesh networks research project. tems; Wireless Networks and Mobile Computing
Available from: <http://www.ece.gatech.edu/research/ Workshop, Phoenix, AZ, April 2001, p. 46166.
labs/bwn/mesh/>. [39] D.S.J. De Couto, D. Aguayo, J. Bicket, R. Morris, A
[25] L. Buttyan, J.-P. Hubaux, Report on a working session high-throughput path metric for multi-hop wireless rout-
on security in wireless ad hoc networks, ACM Mobile ing, in: ACM Annual International Conference on
Computing and Communications Review 7 (1) (2002) 74– Mobile Computing and Networking (MOBICOM),
94. 2003, pp. 134–146.
ARTICLE IN PRESS
[40] J. Dickman, K. Rath, L. Kotecha, Proposal for 802.16 [57] J. Hauser, Draft PAR for IEEE 802.11 ESS Mesh, IEEE
Connection Oriented Mesh, IEEE 802.16 Standard Pro- Document Number: IEEE 802.11-03/759r2.
posal, March 2003. [58] M. Heissenbuttel, T. Braun, BLR: beacon-less routing
[41] R. Draves, J. Padhye, B. Zill, Comparisons of routing algorithm for mobile ad hoc networks, Computer Com-
metrics for static multi-hop wireless networks, in: ACM munications 27 (11) (2004) 1076–1086.
Annual Conference of the Special Interest Group on Data [59] G. Holland, N. Vaidya, Link failure and congestion:
Communication (SIGCOMM), August 2004, pp. 133– analysis of TCP performance over mobile ad hoc
144. networks, in: ACM Annual International Conference on
[42] R. Draves, J. Padhye, B. Zill, Routing in multi-radio, Mobile Computing and Networking (MOBICOM), 1999,
multi-hop wireless mesh networks, in: ACM Annual pp. 219–230.
International Conference on Mobile Computing and [60] Y. Hu, A. Perrig, D. Johnson, Ariadne: a secure on-
Networking (MOBICOM), 2004, pp. 114–128. demand routing protocol for ad hoc networks, in: ACM
[43] A. Dutta, K.D. Wong, J. Burns, R. Jain, A. McAuley, K. Annual International Conference on Mobile Computing
Young, H. Schulzrinne, Realization of integrated mobility and Networking (MOBICOM), September 2002, pp. 12–
management protocol for ad hoc networks, in: IEEE 23.
Military Communication Conference (MILCOM), 2002, [61] Y. Hu, A. Perrig, D. Johnson, Packet leashes: a defense
pp. 448–454. against wormhole attacks in wireless networks, in: IEEE
[44] Engim Inc., Multiple Channel 802.11 Chipset. Available Annual Conference on Computer Communications
from: <http://www.engim.com/products_en3000.html>. (INFOCOM), 2003, pp. 1976–1986.
[45] FCC Cognitive Radios. Available from: <http:// [62] L. Huang, T. Lai, On the scalability of IEEE 802.11 ad
www.fcc.gov/oet/cognitiveradio>. hoc networks, in: ACM International Symposium on
[46] B. Fette, SDR Technology Implementation for the Mobile Ad Hoc Networking and Computing (MOBI-
Cognitive Radio, FCC Workshop on Cognitive Radios, HOC), 2002, pp. 173–182.
May 2003. [63] J.-P. Hubaux, L. Butttan, S. Capkun, The quest for
[47] Firetide Networks. Available from: <www.firetide.com>. security in mobile ad hoc networks, in: ACM Interna-
[48] H. Frey, Scalable geographic routing algorithms for tional Symposium on Mobile Ad Hoc Networking and
wireless ad hoc networks, IEEE Network Magazine 1 Computing (MOBIHOC), 2001, pp. 146–155.
(2004) 18–22. [64] IEEE 802.11 Standard Group Web Site. Available from:
[49] Z. Fu, X. Meng, S. Lu, A transport protocol for <http://www.ieee802.org/11/>.
supporting multimedia streaming in mobile ad hoc [65] IEEE 802.15 Standard Group Web Site. Available from:
networks, IEEE Journal on Selected Areas in Communi- <http://www.ieee802.org/15/>.
cations 21 (10) (2003) 1615–1626. [66] IEEE 802.16 Standard Group Web Site. Available from:
[50] Y. Ganjali, A. Keshavarzian, Load balancing in ad hoc <http://www.ieee802.org/16/>.
networks: single-path routing vs. multi-path routing, in: [67] IEEE 802 Standard Working Group, Wireless LAN
IEEE Annual Conference on Computer Communications medium access control (MAC) and physical layer (PHY)
(INFOCOM), March 2004, pp. 1120–1125. specifications: high-speed physical layer in the 5 GHz
[51] G.D. Golden, et al., Detection algorithm and initial Band, IEEE 802.11a Standard, 1999.
laboratory results using V-BLAST space–time communi- [68] IEEE 802 Standard Working Group, Wireless LAN
cation architecture, IEE Electronics Letters 35 (1) (1999) medium access control (MAC) and physical layer (PHY)
14–16. specifications: further higher data rate extension in
[52] A. Grilo, M. Nunes, Link adaptation and transmit power 2.4 GHZ Band, IEEE 802.11g Standard, 2003.
control for unitcast and multicast in IEEE 802.11a/h/e [69] Intel Inc., Multi-Hop Mesh Networks—a new kind of Wi-
WLANs, in: IEEE International Conferences on Local Fi network.
Computer Networks, 2003, pp. 334–345. [70] Intel Exploratory Research, Heterogeneous sensor net-
[53] X. Guo, S. Roy, W.S. Conner, Spatial reuse in wireless ad works. Available from: <http://www.intel.com/research/
hoc networks, in: IEEE Vehicular Technology Confer- exploratory/heterogeneous.htm>.
ence, October 2003, pp. 1437–1442. [71] Intel Mesh Networks Demonstrations. Available from:
[54] P. Gupta, P.R. Kumar, The capacity of wireless networks, <http://www.intel.com/idf/us/fall2003/presentations/
IEEE Transactions on Information Theory 46 (2) (2000) F03USWNTS116_OS.pdf>.
388–404. [72] K. Jain, J. Padhye, V. Padmanabhan, L. Qiu, Impact of
[55] V. Gupta, S. Krishnamurthy, M. Faloutsos, Denial of interference on multi-hop wireless network performance,
service attacks at the MAC layer in wireless ad hoc in: ACM Annual International Conference on Mobile
networks, in: IEEE Military Communication Conference Computing and Networking (MOBICOM), September
(MILCOM), 2002, pp. 1118–1123. 2003, pp. 66–80.
[56] Z.J. Haas, B. Liang, Ad Hoc mobility management with [73] D.B. Johnson, D.A. Maltz, Y.-C. Hu, The dynamic
uniform quorum systems, IEEE/ACM Transactions on source routing protocol for mobile ad hoc networks
Networking 7 (2) (1999) 228–240. (DSR), IETF Internet-Draft: work in progress, July 2004.
ARTICLE IN PRESS
[74] T.-S. Jou, D.E. Eastlake, ESS MESH Network Study [89] B. Lane, Cognitive radio technologies in the commercial
Group Meeting Minutes, May 2004. arena, in: FCC Workshop on Cognitive Radios, May
[75] J. Jun, M.L. Sichitiu, The nominal capacity of wireless 2003.
mesh networks, IEEE Wireless Communications 10 (5) [90] J. Li, C. Blake, D.S.J. De Couto, H.I. Lee, R. Morris,
(2003) 8–14. Capacity of ad hoc wireless networks, in: ACM Annual
[76] J. Kajiya, Commodity software steerable antennas for International Conference on Mobile Computing and
mesh networks, Microsoft Mesh Networking Summit, Networking (MOBICOM), 2001, pp. 61–69.
June 2004. [91] L. Li, J.Y. Halpern, P. Bahl, Y-M. Wang, R. Wattenho-
[77] V. Kawadia, P.R. Kumar, A cautionary perspective on fer, A cone-based distributed topology-control algorithm
cross layer design, IEEE Wireless Communications, for wireless multi-hop networks, IEEE/ACM Transac-
submitted for publication. tions on Networking, in press.
[78] J.-W. Kim, N. Bambos, Power efficient MAC scheme [92] J. Liu, S. Singh, ATCP: TCP for mobile ad hoc networks,
using channel probing in multirate wireless ad hoc IEEE Journals on Selected Areas in Communications 19
networks, in: IEEE Vehicular Technology Conference, (7) (2001) 1300–1315.
Fall 2002, pp. 2380–2384. [93] B. Liu, Z. Liu, D. Towsley, On the capacity of hybrid
[79] P. Kinney, IEEE 802.15 General Interest in Mesh wireless networks, in: IEEE Annual Conference on
Networking, IEEE 802.15 Request for Information of a Computer Communications (INFOCOM), San Fran-
Mesh Network Study Group, presentation slides, Novem- cisco, CA, 2003, pp. 1543–1552.
ber 2003. [94] A. Lozano, F.R. Farrokhi, R.A. Valenzuela, Lifting the
[80] Kiyon Autonomous Networks. Available from: <http:// limits on high-speed wireless data access using antenna
www.kiyon.com>. arrays, IEEE Communications Magazines 39 (2001) 156–
[81] L. Kleinrock, J. Silvester, Optimum transmission radii for 162.
packet radio networks or why six is a magic number, in: [95] D.A. Maltz, J. Broch, D.B. Johnson, Lessons from a full-
Proceedings of the IEEE National Telecommunications scale multihop wireless ad hoc network testbed, IEEE
Conference, Birmingham, Alabama, December 1978, pp. Personal Communications 8 (2001) 8–15.
4.3.1–4.3.5. [96] P. Marshall, Beyond the outer limits: next generation
[82] Y.B. Ko, V. Shankarkumar, N.H. Vaidya, Medium access communications, FCC Workshop on Cognitive Radios,
control protocols using directional antennas in ad hoc May 2003.
networks, in: IEEE Annual Conference on Computer [97] M. McHenry, Frequency Agile Spectrum Access Tech-
Communications (INFOCOM), 2000, pp. 13–21. nologies, FCC Workshop on Cognitive Radios, May
[83] U.C. Kozat, L. Tassiulas, Throughput capacity of 2003.
random ad hoc networks with infrastructure support, in: [98] Mesh Networking Forum, Building the business case
ACM Annual International Conference on Mobile Com- for implementation of wireless mesh networks, Mesh
puting and Networking (MOBICOM), San Diego, CA, Networking Forum 2004, San Francisco, CA, October
2003, pp. 55–65. 2004.
[84] U.C. Kozat, I. Koutsopoulos, L. Tassiulas, A framework [99] MeshNetworks Inc. QDMA technology. Available
for cross-layer design of energy-efficient communication from: <http://www.meshnetworks.com/pages/technology/
with QoS provisioning in multi-hop wireless networks, in: qdma_overview.htm>.
IEEE Annual Conference on Computer Communications [100] MeshNetworks Inc. Scalable routing technology. Avai-
(INFOCOM), 2004, pp. 1446–1456. lablef from: <http://www.meshnetworks.com/pages/tech-
[85] L. Krishnamurthy, S. Conner, M. Yarvis, J. Chhabra, C. nology/msr_atp_overview.htm>.
Ellison, C. Brabenac, E. Tsui, Meeting the demands of [101] Microsoft Mesh Networks. Availablef from: <http://
the digital home with high-speed multi-hop wireless research.microsoft.com/mesh/>.
networks, Intel Technology Journal 6 (4) (2002) 57–68. [102] J. Mitola III, Software Radio Architecture: Object-
[86] L. Krishnamurthy, Making radios more like human ears: Oriented Approaches to Wireless System Engineering,
alternative MAC techniques and innovative platforms to Wiley Inter-Science, New York, 2000.
enable large-scale meshes, Microsoft Mesh Networking [103] S. Mueller, D. Ghosal, Multipath routing in mobile ad
Summit, June 2004. hoc networks: issues and challenges, in: M.C.i Calzarossa,
[87] T. Lai, D. Zhou, Efficient and scalable IEEE 802.11 ad E. Gelenbe (Eds.), Lecture Notes in Computer Science,
hoc mode timing synchronization function, in: 17th IEEE 2004.
International Conferences on Advanced Information [104] R.D. Murch, K. Ben Letaief, Antenna systems for
Networking and Applications, 2003. broadband wireless access, IEEE Communications Mag-
[88] M. Lampe, N. Rohling, W. Zirwas, Misunderstandings azine 40 (4) (2002) 76–83.
about link adaptation for frequency selective fading [105] A. Nasipuri, S. Ye, R.E. Hiromoto, A MAC protocol for
channels, in: IEEE International Symposium on Personal, mobile ad hoc networks using directional antennas, in:
Indoor, and Mobile Radio Communications (PIMRC), IEEE Wireless Communications and Networking Con-
2002, pp. 710–714. ference (WCNC), 2000, pp. 1214–1219.
ARTICLE IN PRESS
[106] Nortel Mesh Networks. Available from: <http:// [123] A. Spyropoulos, C.S. Raghavendra, Energy efficient
www.nortelnetworks.com/solutions/wrlsmesh/index. communications in ad hoc networks using directional
html>. antenna, in: IEEE Annual Conference on Computer
[107] PacketHop Networks. Available from: <http://www. Communications (INFOCOM), 2002, pp. 220–228.
packethop.com>. [124] A. Spyropoulos, C.S. Raghavendra, Asymptotic capacity
[108] R. Ogier, F. Templin, M. Lewis, Topology dissemination bounds for ad hoc networks revisited: the directional
based on reverse-path forwarding (TBRPF), IETF RFC and smart antenna cases, in: IEEE Global Telecommu-
3684, February 2004. nications Conference (GLOBECOM), 2003, pp. 1216–
[109] C. Perkins, E. Belding-Royer, S. Das, Ad hoc on-demand 1220.
distance vector (AODV) routing, IETF RFC 3561, July [125] J. Sucec, I. Marsic, Location management for hierarchi-
2003. cally organized mobile ad hoc networks, in: IEEE
[110] M. Petrovic, M. Aboelaze, Performance of TCP/UDP Wireless Communications and Networking Conference
over ad hoc IEEE 802.11, in: International Conference on (WCNC), 2002, pp. 603–607.
Telecommunications, 2003, pp. 700–708. [126] K. Sundaresan, R. Sivakumar, M.A. Ingram, T.-Y.
[111] P. Piggin, B. Lewis, P. Whitehead mesh networks in fixed Chang, A fair medium access control protocol for ad
broadband wireless access: multipoint enhancements for hoc networks with MIMO links, in: IEEE Annual
the 802.16 standard, IEEE 802.16 presentation slides, July Conference on Computer Communications (INFOCOM),
2003. 2004, pp. 2559–2570.
[112] N. Poojary, S.V. Krishnamurthy, S. Dao, Medium access [127] K. Sundaresan, V. Anantharaman, H.-Y. Hsieh, R.
control in a network of ad hoc mobile nodes with Sivakumar, ATP: a reliable transport protocol for ad-
heterogeneous power capabilities, in: IEEE International hoc networks, in: ACM International Symposium on
Conference on Communications (ICC), 2001, pp. 872–877. Mobile Ad Hoc Networking and Computing (MOBI-
[113] R. Poor, Wireless mesh networks, Sensors, February HOC), 2003, pp. 64–75.
2003. [128] S. Tierney, Mesh Networks, whitepaper of community-
[114] R. Poor, Wireless mesh links everyday devices, Electronic networking.org.
Engineering Times, 5 July 2004. [129] Tropos Networks. Available from: <http://www.tropos.
[115] D. Qiao, S. Choi, K.G. Shin, Goodput analysis and link com/technology/whitepaper.shtml>.
adaptation for IEEE802.11a wireless LANs, IEEE Trans- [130] D.N.C. Tse, M. Grossglauser, Mobility increases the
actions on Mobile Computing 1 (4) (2002) 278–292. capacity of ad hoc wireless networks, IEEE/ACM Trans-
[116] D. Qiao, K. Shin, UMAV: a simple enhancement to IEEE actions on Networking 10 (4) (2002) 477–486.
802.11 DCF, in: Hawaii International Conference on [131] Y.-C. Tseng, C.-S. Hsu, T.-Y. Hsieh, Power-saving
System Science, 2002. protocols for IEEE 802.11 based multi-hop ad hoc
[117] R. Ramanathan, On the performance of ad hoc networks networks, in: IEEE Annual Conference on Computer
with beamforming antennas, in: ACM International Communications (INFOCOM), 2002, pp. 200–209.
Symposium on Mobile Ad Hoc Networking and Com- [132] UCSD mesh networks testbed. Available from: <http://
puting (MOBIHOC), October 2001, pp. 95–105. www.calit2.net/>.
[118] R. Ramanathan, J. Redi, C. Santivanez, D. Wiggins, S. [133] J. Walker, Wi-Fi mesh networks, the path to mobile ad
Polit, Ad hoc networking with directional antennas: a hoc. Available from: <http://www.wi-fitechnology.com/
complete system solution, in: IEEE Wireless Communi- Wi-Fi_Reports_and_Papers/
cations and Networking Conference (WCNC), 2004, pp. Mesh_Networks_References.html>.
375–380. [134] X. Wang, W. Wang, M. Nova, A high performance
[119] A. Raniwala, K. Gopalan, T. Chiueh, Centralized channel single-channel IEEE 802.11 MAC with distributed
assignment and routing algorithms for multi-channel TDMA, Technical Report of Kiyon, Inc. (submitted for
wireless mesh networks, ACM Mobile Computing and patent application), October 2004.
Communications Review 8 (2) (2003). [135] P. Whitehead, P. Piggin, B. Lewis, S. Lynch Mesh
[120] A.K. Saha, D.B. Johnson, Self-organizing hierarchical extensions to IEEE 802.16 and 16a, IEEE 802.16
routing for scalable ad hoc networking, Technical Report, proposal, May 2003.
TR04-433, Department of Computer Science, Rice [136] The Wi-Fi Alliance. Available from: <http://www.wi-fi.
University. org/>.
[121] C.-C. Shen, C. Srisathapornphat, C. Jaikaeo, An adaptive [137] The WiMAX Forum. Available from: <http://www.wim-
management architecture for ad hoc networks, IEEE axforum.org/home>.
Communications Magazine 41 (2) (2003) 108–115. [138] The WiMedia Alliance. Available from: <http://www.
[122] J. So, N. Vaidya, Multi-channel MAC for ad hoc wimedia.org/>.
networks: handling multi-channel hidden terminals using [139] W. Xiang, T. Pratt, X. Wang, A software radio testbed
a single transceiver, in: ACM International Symposium for two-transmitter two-receiver space time coding wire-
on Mobile Ad Hoc Networking and Computing (MOBI- less LAN, IEEE Communications Magazine 42 (6) (2004)
HOC), May 2004, pp. 222–233. S20–S28.
ARTICLE IN PRESS
[140] K. Xu, X. Hong, M. Gerla, Landmark routing in ad hoc He received the 2002 IEEE Harry M. Goode Memorial award
networks with mobile backbones, Journal of Parallel and (IEEE Computer Society) with the citation ‘‘for significant and
pioneering contributions to advanced architectures and proto-
Distributed Computing, Special Issue on Ad Hoc Net- cols for wireless and satellite networking’’.
works 63 (2) (2002) 110–122. He received the 2003 IEEE Best Tutorial Award (IEEE Com-
[141] S. Xu, T. Saadawi, Does the IEEE 802.11 MAC proto- munication Society) for his paper entitled ‘‘A Survey on Sensor
col work well in multihop wireless ad hoc networks, Networks’’, published in IEEE Communication Magazine, in
August 2002.
IEEE Communications Magazine 39 (6) (2001) 130– He received the 2003 ACM SIGMOBILE award for his signifi-
137. cant contributions to mobile computing and wireless networking.
[142] G. Xylomenos, G.C. Polyzos, P. Mahonen, M. Saaranen, His current research interests are in wireless mesh networks,
TCP performance issues over wireless links, IEEE Com- sensor networks, InterPlaNetary Internet, wireless networks and
satellite networks.
munications Magazine 39 (4) (2001) 52–58.
[143] H. Yang, H. Luo, F. Ye, S. Lu, L. Zhang, Security in
mobile ad hoc networks: challenges and solutions, IEEE
Xudong Wang received his B.E. and
Wireless Communications 11 (1) (2004) 38–47. Ph.D. degrees from Shanghai Jiao
[144] T.-S. Yum, K.-W. Hung, Design algorithms for multihop Tong University, Shanghai, China, in
packet radio networks with multiple directional antennas 1992 and 1997, respectively. From
stations, IEEE Transactions on Communications 41 (11) 1998 to 2003, he was with the Broad-
band and Wireless Networking (BWN)
(1992) 1716–1724. Lab at Georgia Institute of Technol-
[145] M. Zapata, N. Asokan, Securing ad hoc routing proto- ogy. He also received the Ph.D. degree
cols, ACM Workshop on Wireless Secuirty (WiSe), from Georgia Institute of Technology
September 2002, pp. 1–10. in 2003. Currently, he is a senior
research engineer with Kiyon, Inc.,
[146] Y. Zhang, W. Lee, Intrusion detection in wireless ad hoc where he conducts research and
networks, in: ACM Annual International Conference on development of MAC, routing, and
Mobile Computing and Networking (MOBICOM), 2000, transport protocols for wireless mesh networks. His research
pp. 275–283. interests also include software radios, cross-layer design, and
communication protocols for cellular, mobile ad hoc, sensor,
[147] R. Zhong, R. Kravets, On demand power management and ultra-wideband networks.
for ad hoc networks, in: IEEE Annual Conference on He is a technical committee member of IEEE/ACM Broad-
Computer Communications (INFOCOM), 2003, pp. 481– WISE 2004, SANPA 2004, IEEE ITRE 2005, EWSN 2005, and
491. ConWiN 2005. He has been a technical reviewer for numerous
international journals and conferences. He has two patents
[148] The ZigBee Alliance. Available from: <http://www.zig- pending in wireless mesh networks. He is a member of IEEE,
bee.org/>. ACM, and ACM SIGMOBILE.
Ian F. Akyildiz received his BS, MS, Weilin Wang received his BS degree in
and PhD degrees in Computer Engi- radio-electronics from Zhongshan
neering from the University of Erlan- University, MS in computer science
gen-Nuernberg, Germany, in 1978, from New York University, and Ph.D.
1981 and 1984, respectively. in electrical engineering from City
Currently, he is the Ken Byers Distin- College of New York. He is the head
guished Chair Professor with the of KiyonÕs autonomic networks R&D
School of Electrical and Computer group. His research interests are in the
Engineering, Georgia Institute of areas of broadband wireless mesh
Technology and Director of Broad- network architecture and protocols.
band and Wireless Networking Labo- Prior to Kiyon, he was with GravitonÕs
ratory. wireless technology division, co-devel-
oped a multichannel MAC and multi-
He is the Editor-in-Chief of Computer Networks (Elsevier) hop routing wireless sensor networking system for homeland
Journal and of Ad Hoc Networks (Elsevier) Journal. security applications. He was previously chief network architect
He is an IEEE FELLOW (1995), an ACM FELLOW (1996). at OMM, working on high speed switching subsystems for
He served as a National Lecturer for ACM from 1989 until wavelength switching, routing, and mesh restoration. Earlier he
1998 and received the ACM Outstanding Distinguished Lec- was a senior advisor and architect at Nortel Networks
turer Award for 1994. advanced technology lab and broadband networks division,
He received the 1997 IEEE Leonard G. Abraham Prize award and contributed to NortelÕs intellectual property portfolio
(IEEE Communications Society) for his paper entitled ‘‘Mul- development, OPS terabit router QoS and MPLS capabilities,
timedia Group Synchronization Protocols for Integrated Ser- and the success of NortelÕs dynamic routing controller product
vices Architectures’’ published in the IEEE Journal of Selected line. He has a number of issued patents and several pending in
Areas in Communications (JSAC) in January 1996. the areas of wireless mesh networks.