Wireless Sensor Network, 2009, 2, 61-121
doi:10.4236/wsn.2009.12016 lished Online July 2009 (http://www.SciRP.org/journal/wsn/).
Copyright © 2009 SciRes. Wireless Sensor Network, 2009, 2, 61-121
Centralized Quasi-Static Channel Assignment for
Multi-Radio Multi-Channel Wireless Mesh Networks
Juan REN, Zhengding QIU
Institute of Information Science, Beijing Jiaoton g Un iversity, Beijing , China
Email: {04112047, zdqiu}@bjtu.edu.cn
Received February 19, 2009; rev i se d March 19, 2009; accepted March 20, 2009
Employing multiple channels in wireless multihop networks is regarded as an effective approach to increas-
ing network capacity. This paper presents a centralized quasi-static channel assignment for multi-radio
multi-channel Wireless Mesh Networks (WMNs). The proposed channel assignment can efficiently utilize
multiple channels with only 2 radios equipped on each mesh router. In the scheme, the network end-to-end
traffics are first modeled by probing data at wireless access points, and then the traffic load between each
pair of neighboring routers is further estimated using an interference-aware estimation algorithm. Having
knowledge of the expected link load, the scheme assigns channels to each radio with the objective of mini-
mizing network interference, which as a result greatly improves network capacity. The performance evalua-
tion shows that the proposed scheme is highly responsive to varying traffic conditions, and the network per-
formance under the channel assignment significantly outperforms the single-radio IEEE 802.11 network as
well as the 2-radio WMN with static 2 channels.
Keywords: Wireless Mesh Networks, Multihop Network, Channel Assignment, Multi-Radio
1. Introduction
WMN [1] is a promising wireless technology for nu-
merous applications, e.g., broadband home networking,
community and neighborhood networks, enterprise net-
working, building automation, etc. [2,3]. However, in-
terference among wireless links significantly impacts the
performance of WMNs. As a multi-hop wireless network,
the actual goodput available to WMN applications de-
creases a lot when forwarding or relaying packets over
multiple wireless hops.
Fortunately, the IEEE 802.11 PHY specification per-
mits simultaneous operation of multiple non-overlapping
channels. By deploying multi-radio routers in infrastruc-
ture-based networks and assigning radios to non-overlap-
ping channels, the routers can communicate simultane-
ously with little interference in spite of being in direct
interference range of each other. Therefore, the capacity
of wireless networks can be increased. While due to the
limited number of channels available, the interference
cannot be completely eliminated. In addition, the channel
assignment must be restricted to the number of radios on
each wireless node. So it’s a challenging problem de-
serving our research.
In equipping routers with multiple radios, a naive
strategy would be to equip each router with the number
of radios equal to the number of orthogonal channels.
However, this strategy is economically prohibitive due to
the significant number of non-overlapping channels. An-
other channel assignment strategy is to frequently change
channel on the interface, for instance, for each packet
transmission based on current state of the medium. Such
dynamic channel assignment approaches [4-6] require
channel switching at a very fast time scale (per packet or
a handful of packets). The fast-channel switching re-
quirement makes these approaches unsuitable for use
with commodity hardware, where channel switching de-
lays itself can be in the order of milliseconds [4]. Some
of the dynamic channel assignment approaches also re-
*This work is supported by National Basic Research Program of China
973 Pro
Copyright © 2009 SciRes. Wireless Sensor Network, 2009, 2, 61-121
quire specialized MAC protocols or extensions of 802.11
MAC layer, making them further unsuitable for use in
commodity 802.11 hardware.
In order to use multiple channels with commodity
hardware, several researches [7-9] focused on develop-
ing techniques that assign channels statically. Such static
assignments can be changed whenever there are signifi-
cant changes to traffic load or network topology. Since
WMN is an infrastructured network and aims to provide
reliable broadband services, such changes are infrequent
enough that the channel-switching delay and traffic
measurement overheads are insignificant. We refer to the
above as quasi-static channel assignments. However,
most of the existing quasi-static channel assignments are
performed offline and bound with routing.
In this paper, we address the problem of quasi-static
channel assignment independent of routing. A central-
ized quasi-static channel assignment algorithm is pro-
posed in the context of networks with multi-radio nodes.
In the channel assignment, we use a novel scheme to
estimate the traffic load on each wireless link. The esti-
mation considers the traffic on the link itself as well as
the interfering traffics introduced by its neighbors. Hav-
ing knowledge of the expected load on each link, the
algorithm can intelligently select different channels for
each radio with the objective of minimizing network
interference, which as a result efficiently improves the
network capacity. To evaluate the algorithm performance,
a corresponding channel assignment protocol is imple-
mented in ns-2 simulations [10] and we incorporate the
well-known WCETT (Weighted Cumulative Expected
Transmission Time) path metric [11], which is tailored
for multi-radio multihop wireless networks, into the
AODV (Ad Hoc On Demand Distance Vector) routing
protocol [12] as our multi-radio routing protocol. The
performance evaluation shows that the proposed scheme
is highly responsive to varying traffic conditions, and the
network performance under the channel assignment sig-
nificantly outperforms the single-radio IEEE 802.11
network as well as the 2-radio WMN with static 2 chan-
The rest of the paper is organized as follows. Section 2
gives the system architecture of the proposed multi-radio
WMN. In Section 3, we describe the centralized
quasi-static channel assignment scheme. In Section 4, we
evaluate the performance of our channel assignment al-
gorithm using the ns-2 simulations. Section 5 concludes
the paper.
2. System Architecture
In this section, we formulate the interference problem
involved in wireless multihop networks and present the
architecture of multi-radio multi-channel WMNs to re-
solve this problem.
2.1. Interference Problem
Traditional 802.11-based wireless networks can’t trans-
mit data simultaneously as wired networks because of
the intra-path and inter-path interference. For example in
Figure 1, although the two flows transmit separately on
path 1 and path 2, nodes must compete with each other
for a common channel, which reduces network through-
put hardly. If node 3 is in transmission, all the nodes in
interference range of node 3 should keep silence, or a
collision will occur. In contrast, if we assign interfering
hops to different channels, then one collision domain can
be broken into several collision domains with each oper-
ating in a different frequency range. When the in-
gress-egress node pairs that originally pass through the
collision domain now take different paths to route their
traffic, hops using different channels can transmit simul-
taneously and the network throughput will increase.
2.2. Multi-Radio Multi-Channel WMN
Figure 2 gives the architecture of multi-radio multi-chan-
nel WMN. The wireless mesh backbone network consists
of mesh routers (MR), mesh access routers (MAR) and
the gateway. Mesh routers provide purely wireless rout-
ing services. Mesh access routers provide not only wire-
less routing services but also wireless access services.
Each WMN has at least one gateway, which can also be
served as the access point for wireless users. The integra-
tion of WMN with other networks such as the Internet,
cellular, IEEE 802.11, IEEE 802.15, IEEE 802.16, sen-
sor networks, etc., can be accomplished through the
gateway and bridging functions in the mesh routers.
In WMN only end users may frequently move and
mesh backbone facilities are almost static once been set-
tled. In addition, both the gateway and the mesh access
routers have aggregation capability. So we use them to
measure the ingress-egress network traffic in our load
estimation algorithm. Although in our network each
router is equipped with only 2 radios, the overall network
can utilize more channels with intelligent channel as-
signment to every link. This is the fundamental reason
for non-linear improvement in throughput with respect to
the increase in number of radios per node.
5 Path 1
Path 2
Interfere nce
Figure 1. Interference in wireless communication.
Copyright © 2009 SciRes. Wireless Sensor Network, 2009, 2, 61-121
1 5
5 Virtual Link
Operating on
Channel 1
Figure 2. Architecture of multi-radio multi-channel WMN.
3. Design of Channel Assignment
The goal of channel assignment in multi-radio WMNs is
to bind each radio to a channel in such a way that the
available bandwidth on each link is proportional to its
expected load. The link here means direct communica-
tion on a channel between the routing pair. In this section,
we describe the load estimation and channel assignment
3.1. Traffic Measurement
Optimizations of channel assignment using load estima-
tion require knowledge of network traffic information, so
we propose to measure the end-to-end traffics between
mesh access routers. The traffic measurement procedure
is described as follows:
At first, each mesh access router (including the gate-
way) measures its ingress-egress flows by probing data
periodically (the interval is set to 10s in simulations). For
convenient expression, we use the access router to indi-
cate both the mesh access router and the gateway in the
rest of this paper. Then each access router aggregates its
ingress-egress flows and sends the information to the
gateway. (The gateway is used as the computation center
since it owns the most powerful capacity.)
After receiving the flow information, the gateway
calculates the end-to-end traffic value between every pair
of access routers by further aggregate the flow informa-
tion. In this way, the real time value of the end-to-end
traffic between each pair of access routers is measured at
each echo interval. However, since what the quasi-static
algorithm needed is a long-term measured traffic, the
gateway performs an exponentially weighted moving
average (EWMA) of each end-to-end traffic load to get
an approximate long-term traffic. That is:
(,)(,) (1)(,)
TsdT sdTsd
 (1)
where denotes the end-to-end traffic between
access router pair s and d. In simulations, the smoothing
factor α = 0.7.
(, )Tsd
3.2. Initial Expected Load
Having the knowledge of the end-to-end traffics, the
gateway estimates the expected load on each wireless
link and assigns channels to links in order of the ex-
pected loads. The gateway is required to perform a new
estimation of the expected load either when it receives
the traffic information for the first time, or when the dif-
ference between the new traffic and the last one is large
To initial the load estimation, we assume that there is a
link between each pair of routers in direct communica-
tion range, and each end-to-end traffic load is equally
divided among all the paths with the least hops between
the pair of access routers. (Note that this link won’t
really exist if there is no common channel assigned to the
pair of routers on this link.)
If the number of all shortest paths between node s and
d is, and in those paths there are paths
passes link i, then the initial expected load for link i to
carry is calculated as follow:
(, )Psd (, )
(, )
()( ,)
(, )
Here we only count the shortest paths because the path
with less hops always have much better performance
compared with longer paths in multi-hop wireless net-
works if they all have enough bandwidth.
3.3. Channel Assignment
Having knowledge of the expected loads on all network
links, we start to assign channels to links as follows:
At first, all links are sorted by their expected loads.
Since links expected to carry higher traffic load should
be given more bandwidth, the link with the most ex-
pected load is prior to other links in choosing channels.
Assume every node is equipped with q radios and node a
and b is connected by link i. There are three conditions
for choosing a channel to link i:
1) If both of node a and b have used less than q radios,
then choose a channel with the least interference with
link i. (The chosen channel should also be different from
the used channels of a and b.) Meanwhile, update the
channel lists of node a and b.
2) If one of the two nodes (for example a) has used q
channels, then choose a channel from the channel as-
signment list of node a with the least interference with
link i. Meanwhile update the channel list of node b.
3) When both of node a and b have used q channels: if
there exists a common channel between node a and b,
then assign this channel to link i; else, choose one chan-
nel from the list of node a and b respectively and merge
the two channels to one. Meanwhile update the channel
Copyright © 2009 SciRes. Wireless Sensor Network, 2009, 2, 61-121
lists of all nodes which connect to node a or b using the
two channels directly or indirectly.
When link i has been assigned a channel, we remove it
from the unassigned link list. As a result, the link with
lower expected load now owns the priority to choose
channel next. This continues until all network links have
been visited, which is denoted as a cycle of channel as-
3.4. End or Feedback
After a cycle of channel assignment for all links, we need
to judge whether current channel assignment satisfies all
bandwidth requirements. If so, we terminate the whole
procedure and output the channel assignment results, or
we feedback the expected link loads under new channel
assignment to find a better channel assignment.
It’s easy to see that, if the available bandwidth on each
link is more than the traffic load it’s expected to carry,
no congestion will occur. So at first we need to estimate
the capacity for each link in the network. However, in
wireless networks, channels are shared by all links in the
same interference range. So when estimating the usable
capacity of a link, we should consider all traffic loads in
its interference range. According to the channel assign-
ment rules, the higher load a link is expected to carry, the
more bandwidth it should get. On the other side, the
higher loads its interfering links are expected to carry,
the less bandwidth it could obtain. Thus, the link capac-
ity should be proportional to its traffic load, and be in-
versely proportional to all other interfering loads. So the
capacity for a link i is given by:
() ()
Ci j
where B is the channel bandwidth and ()
ntf i stands
for the set of links in the interference range of i including
Then the residual capacity of link i can be obtained as
()() ()RC iC ii
 (4)
We use the minimal residual capacity of all links on a
path as the available bandwidth for this path. Then we
consider the bandwidth requirement of an end-to-end
traffic is satisfied if there exists a path with its available
bandwidth more than the required traffic value.
At the start of the traffic allocation, the initial
of each link is set to the expected capacity , and the
expected load
()RC i
of each link is set to 0. When a path
is chosen to carry an end-to-end traffic, all links on the
path reduce their residual capacities and increase their
expected loads by the allocated traffic value.
Assume there are totally N end-to-end traffics with
respectively the value of . We choose
path P with the maximal available bandwidth among all
the paths with both the least hop and enough available
bandwidth for traffic , which is because the WCETT
path metric used in multi-channel routing protocol pre-
fers to choose high throughput paths in multiple channels.
When a path is chosen for traffic , there are two con-
ditions when allocating traffic to this path:
T (1...,)nN
1) If min(( ))
, we can allocate the whole
traffic to path P. So we decrease the residual capacity
and add the expected load of each link i on path P by .
We also decrease the total unallocated traffic by for
a later comparison of each iteration.
2) If, we can only allocate
traffic to path P. So we decrease the re-
sidual capacity and add the expected load of each link i
in path P by . We also decrease the total
unallocated traffic value by .
( ))RC i
min(( ))
iP RC i
min (
min(( ))
iP RC i
When all the N traffics have been checked, we termi-
nate the whole algorithm if the total unallocated traffic
equals to 0. Or, we compare the total unallocated traffic
of this cycle to the last one. If the unallocated traffic of
this cycle is no less than the last one, it means no im-
provement is made and we also terminate the whole
process. Otherwise, we feedback the new expected link
loads to the channel assignment for a better scheme.
After the whole algorithm ends, the gateway broad-
casts the channel assignment results. Then each mesh
router adjusts its radios to the assigned channels when
they receive the results.
If we combine the traffic measurement, load estima-
tion and channel assignment, the whole algorithm proc-
ess can be depicted as in Figure 3.
4. Simulation Analysis
To evaluate the performance of our channel assignment,
we run simulations using ns-2. We use the IEEE 802.11
MAC protocol with RTS/CTS enabled. The AODV pro-
tocol is used as the routing protocol for the single-radio
IEEE 802.11 network simulations. We modify the AODV
protocol using the WCETT metric, a prevailing path met-
ric, as our multi-radio routing protocol. The topology of
all networks in the simulations is a 25-node square grid.
To simplify and generalize the simulation, we configure
the center node as the gateway and all the other 24 nodes
as mesh access routers. The bandwidth of each channel is
2 Mbps. The ratio between interference and communica-
tion range is 2 and all nodes in multi-channel networks
are equipped with 2 radios.
Copyright © 2009 SciRes. Wireless Sensor Network, 2009, 2, 61-121
Channel Assignment
Traffic Measurement
Load Initialization
All traffics have been allocated
Or no improvement exists?
Traffics T
Link L oads
Capacity Estimation
Link Capaciti es C
Traffic Allocation
Load Estimation
Figure 3. The whole al gorithm proces s .
4.1. Static Traffic Simulation
4.1.1. 10 Flows Simulation
At first, we randomly choose 10 pair of nodes from the
25 nodes and assign each pair with a different CBR UDP
flow. The rate of each flow is chosen randomly between
0 and 0.8Mbps. The packet size is 1000 bytes and flows
run for 100 seconds. We test our channel assignment
algorithm in a 5-channel network with the 10 flows. To
have a good comparison, we also run the same flows in a
standard IEEE 802.11 network and a 2-radio 2-channel
network. The two networks both needn't any channel
assignment and the 2-channel network simulation is just
the original WCETT multi-radio routing simulation. We
evaluate the improvement of network performance by
comparing the aggregate throughputs, the average packet
delays and the average packet drop probabilities of the 3
The aggregate throughputs of the 3 networks are shown
in Figure 4. The standard 802.11 network throughput is
quite low and the 2-channel network throughput is about
twice of the standard 802.11 network. Although the im-
provement is significant, we can see both the two net-
work throughputs suffer sharp vibration. This is because
all network nodes using common channels is quite easy
to bring great interference. While in the 5-channel net-
work, we can efficiently limit interference to several
small areas using our channel assignment. So the net-
work throughput is quite stable and the value of the
throughput also increases a lot.
The average packet delay of each flow is shown in
Figure 5. We draw the delays of flows that haven’t re-
ceived any data packet successfully in the whole simula-
tion to the max value of the y axis. We can see the
1-channel network performs badly with 1 flow receiving
no packet. Although all flows in the 2-channel network
can transport data, the average packet delays for most
flows are quite large and exceed 0.5s. The maximal av-
erage packet delay of the 2-channel network is up to
1.83s. In the 5-channel network, the average packet de-
lays are much smaller. The maximal delay of all flows is
1.15s and there are 7 flow delays are below 0.5s. This
further proves that under our channel assignment, the
mesh network can efficiently utilize 5 or even more
channels with only 2 radios.
010 2030405060 70 8090100
Time ( s)
Network Aggregat e Throughput (Mbps)
A ODV wi th 1 ch annel
W CE TT wit h 2 channels
W CE TT wit h 5 channels
Figure 4. The aggre gate ne twork throughput for 10 flows.
1 23 45678 910
0. 5
1. 5
2. 5
Flow Index
Average Pac ket Delay (s )
AODV wi th 1 channel
W CE TT wit h 2 c hannels
W CE TT wit h 5 c hannels
Figure 5. The average packet delay for 10 flows.
Copyright © 2009 SciRes. Wireless Sensor Network, 2009, 2, 61-121
We also compare the average packet drop probability
of each flow in Figure 6. The probabilities for the previ-
ous two networks are all quite large. In the 5-channel
network, the probabilities of 4 flows are quite small with
values below 4%. Although the ones for the other flows
are large, but they have decreased a lot compared to the
previous two networks.
4.1.2. 20 Flows Simulation
To derive the network to saturation, we randomly choose
20 pair of nodes from the 25 nodes and run the same
simulations in the above 3 networks. The aggregate thro-
ughput of the 3 networks is shown in Figure 7. From
Figure 7 we can see the throughput gain for 2-channel
network is not significant under the heavy traffic. While
both the value and the stability of throughput for
5-channel network get further significant increase, which
is due to the sufficient bandwidth brought by efficient
utilization of multiple channels.
The average packet delay of each flow is shown in
Figure 8. We can see that in the 1-channel network, 4
flows received no packet in the whole simulation. In the
2-channel network, the average packet delays for most
flows are also quite large with the maximal average de-
lay up to 2.32s for the 19th flow. While in the 5-channel
network, the average packet delays for most flows are
below 0.5s and the maximal one is only 0.96s.
We compare the average packet drop probability of
each flow in Figure 9. The average packet drop prob-
abilities for all the three networks are all quite large be-
cause of the heavy traffic. While compared to the two
networks without channel assignment, the average packet
drop probability for the 5-channel network is still much
smaller and the average drop probability of 9th flow is
nearly 0, which means it has almost transmitted all the
data packets successfully.
1 23 45 67 8910
Fl ow Index
Average Pac ket Drop Probability (%)
A ODV wi t h 1 c h annel
WCETT wit h 2 channels
WCETT wit h 5 channels
Figure 6. The average packet drop probability for 10 flows.
010 2030405060 708090 100
Time ( s)
Network A ggregat e Throughput (Mbps)
A ODV with 1 channel
WCETT wit h 2 channel s
WCETT wit h 5 channel s
Figure 7. The aggre gate ne twork throughput for 20 flows.
12345678910 11 1213 14 15161718 19 20
Fl ow Index
A v e rage P a cket Del ay (s)
A ODV wit h 1 channel
WCE TT wi t h 2 channels
WCE TT wi t h 5 channels
Figure 8. The average packet delay for 20 flows.
12345678910 11 1213 14 1516 17 1819 20
Fl ow Index
Average Pack et Drop Probability (%)
AO DV with 1 channel
W CE TT wit h 2 channel s
W CE TT wit h 5 channel s
Figure 9. The average packet drop probability for 20 flows.
Copyright © 2009 SciRes. Wireless Sensor Network, 2009, 2, 61-121
Table 1. Comparison of the average aggregate throughput.
(Mbps) 1 channels 2 channels 3 chan-
4 chan-
5 chan-
10 flows 0.525 0.903 1.231 1.4371.608
20 flows 0.777 0.814 1.448 2.0802.323
020 40 60 80100 120140 160 180
0. 5
1. 5
2. 5
Ti me ( s)
Net work Aggregat e T hrough put (M bps)
Figure 10. Impact of traffic change on aggregate network
4.1.3. The Average Aggregate Throughput
Without loss of generality, we also run the 10 and 20
flows respectively in the networks with 3 and 4 channels,
using the proposed channel assignment. We list the av-
erage aggregate throughputs of the 5 networks with 2
different traffics in Table 1. We see that the network
throughput increases with more channels assigned to the
network. While the improvement can’t be achieved as
high times as the number of channels, due to the unne-
glectable management packets and probing packets for
both the WCETT path metric and traffic measurement.
4.2. Varying Traffic Simulation
To evaluate the impact of traffic change on network per-
formance, we randomly choose 3 flows in the 10-flow
scene and vary their sending rates in the simulation.
Flows run for 180 seconds. At the 60th second, we in-
crease the sending rate of a flow by 0.2Mbps. Then at the
80th second, we increase the sending rate of another flow
by 0.4Mbps. At the 100th second, we decrease the send-
ing rate of the third flow by 0.4Mbps. After this, the traf-
fic changes are large enough for the gateway to reassign
channels. Because the traffic measurement interval is 10
seconds, the gateway should detect the traffic change and
start to reassign channels at the 110th second. Besides,
each node need to fresh its routing table since the net-
work channel assignment has changed. So we purge the
history information of the WCETT path metric in each
node after every new channel assignment.
Figure 10 shows the serial changes of aggregate net-
work throughput by varying the flow sending rates. We
can see from the 60th second to the 110th second, the ag-
gregate network throughput changes slightly after each
variation of flow sending rate and at last stays around
1.85Mbps. Then at the 110th second, the aggregate
throughput suddenly decreases to 0, which means the
network begins to reassign channels. After about 15
seconds vibration, the aggregate throughput comes back
to near 2.4Mbps which means the new channel assign-
ment as well as the WCETT path metric routing have
Compared to the last 1.85Mbps throughput before
channel reassignment, the network now owns a much
better bandwidth assignment. In addition, the channel
reassignment and routing together cost only about 15
seconds. So we can say our channel assignment combin-
ing with the WCETT path metric routing provides a good
choice for the multi-radio WMNs.
5. Conclusions
This paper formulated the channel assignment problem
in multi-radio multi-channel WMNs. Since the backbone
of WMN is an infrastructured network, we assume the
mesh routers are stationary and each of them is equipped
with 2 radios. Based on network traffic measurement and
the load estimation of wireless links, we present a cen-
tralized quasi-static channel assignment with the objec-
tive of minimizing network interference, which as a re-
sult greatly improves network capacity. Extensive simu-
lations show that the proposed scheme is highly respon-
sive to varying traffic conditions, and the network per-
formance under the channel assignment significantly
outperforms the single-radio IEEE 802.11 network as
well as the 2-radio WMN with static 2 channels.
6. References
[1] Mesh Networks Inc. http://www.meshnetworks.com.
[2] I. F. Akyildiz, X. D. Wang, and W. L. Wang, “Wireless
mesh networks: A survey,” Computer Networks, Vol. 47,
No. 4, pp. 445–487, 2005.
[3] Mesh Networking Forum, “Building the business case for
implementation of wireless mesh networks,” Mesh Net-
working Forum 2004, San Francisco, CA, October 2004.
Copyright © 2009 SciRes. Wireless Sensor Network, 2009, 2, 61-121
[4] R. Chandra and P. Bahl, “MultiNet: Connecting to multi-
ple IEEE 802.11 networks using a single wireless card,”
INFOCOM, Vol. 2, pp. 882–893, 2004.
[5] I. Wormsbecker and C. Williamson, “On channel selec-
tion strategies for multi-channel MAC protocols in wire-
less ad hoc networks,” IEEE Conference on Wireless and
Mobile Computing, Networking and Communications
(WiMob’2006), pp. 212–220, 2006.
[6] J. So and N. Vaidya, “Multi-channel MAC for ad hoc
networks: Handling multi-channel hidden terminals using
a single transceiver,” MobiHoc’04, May 24–26, 2004.
[7] A. Raniwala, K. Gopalan, and T. Chiueh, “Centralized
channel assignment and routing algorithms for multi-
channel wireless mesh networks,” ACM Mobile Com-
puting and Communications Review, Vol. 8, No. 2, pp.
50–65, 2004.
[8] J. Tang, G. Xue, and W. Zhang, “Interference-aware to-
pology control and QoS routing in multi-channel wireless
mesh networks,” ACM SIGMOBILE, Urbana-Champaign,
IL, pp. 68–77, 2005.
[9] A. Subramanian, H. Gupta, and S. R. Das, “Mini-
mum-interference channel assignment in multi-radio wire-
less mesh networks,” Proceedings of 4th Annual IEEE
Communications Society Conference on Sensor, Mesh and
Ad Hoc Communications and Networks ( SECON’'07), pp.
481–490, June 18–21, 2007.
[10] “ns-2 simulator,” http://www.isi.edu/nsnam/ns.
[11] R. Draves, J. Padhye, and B. Zill, “Routing in multi-radio,
multi-hop wireless mesh networks,” in Proceedings of
ACM MOBICOM, pp. 114–128, September 2004.
[12] C. Perkins, E. Royer, and S. Das, “Ad hoc on demand
distance vector (AODV) routing,” IETF Internet Draft,
draft-ietf-manet-aodv2-10.txt, January 24, 2002.