A Distributed Channel Assignment for 802.

11-based
Multi-Radio Wireless Mesh Networks
Deepesh Man Shrestha
Graduate School of Information
and Communication
Ajou University, Korea
Email: deepesh@ajou.ac.kr
Arun Ranjitkar
Graduate School of Information
and Communication
Ajou University, Korea
Email: arun@ajou.ac.kr
Young-Bae Ko
Graduate School of Information
and Communication
Ajou University, Korea
Email: youngko@ajou.ac.kr

Abstract—In this paper we propose a distributed channel
assignment protocol (DCAP) for topology formation in the
IEEE 802.11 based wireless mesh network. The non-overlapping
channels are assigned to the network interfaces based on the
simple heuristics (such as degree and node-ID) with the goals to
enable connectivity and reduce interference in the network. Ns-2
simulation results are presented and compared with the solutions
from integer linear programming (ILP) and the common channel
assignment (CCA) algorithm. We measured the number of
concurrent links, average throughput and end-to-end latency and
showed that DCAP produces optimally connected topology and
performs better in all cases compared to CCA.
I. I NTRODUCTION
Wireless multi-hop mesh networks based on the IEEE
802.11s is different from the traditional IEEE 802.11 WLAN.
The traditional WLAN architecture like cellular networks has
an access point (AP) directly connected to the gateway limiting
access to only those stations that are one hop away. In contrary,
the current IEEE 802.11s draft [1] defines a multi-hop WLAN
mesh of wireless routers (named mesh points or MPs) forming
a wireless infrastructure backbone. For increasing the network
capacity, nodes can be equipped with multiple interfaces and
use multiple channels. Each MP in such a WLAN mesh is also
expected to be self-configuring while forming the backbone
topology.
The problem of topology formation in the WLAN mesh
becomes non-trivial with the nodes having multiple interfaces
and channels. The network might be disconnected due to a
channel mismatch among neighbors or becomes affected by
some interference due to excessive use of same channels.
It might not be possible to fix one channel throughout the
life of network because of continuous topological changes.
Moreover, due to limiting constraints of resources, assigning
channels to a node interface might have a rippling effect [2]
over the entire topology. For example, when a node assigns
different channels, it might enforce its neighbors to change
their previously assigned channels to preserve connectivity.
This change might propagate further causing a ripple effect.
Thus, a proper channel assignment algorithm is a key require-
ment to generating a self-configuring network topology of high
performance.
We consider a static channel assignment for topology
formation where the channel assigned to a particular radio
interface is tuned for a significant time period. In general,
channel assignment can be classified into traffic dependent
and independent algorithms. Traffic dependent algorithms such
as [3], [4] use some predefined traffic profiles. [3] proposes
centralized interference-aware topology control and QoS rout-
ing in IEEE 802.11 based mesh networks. Their algorithm
produces k-connected network by assigning channels to min-
imize interference, upon which QoS routing is performed. [4]
considers availability of traffic profiles at each node before
assigning channels. However, it may not be feasible to predict
such profiles a priori due to a time varying traffic demand and
a wide range of communication pattern [5].
Probably the simplest traffic independent static channel as-
signment algorithm is common channel assignment (CCA) [5].
The radio interfaces of all nodes are assigned with the same set
of channels. For example, for all nodes having two interfaces
stick to channel 1 and 2. Clearly, CCA leads to the poor
channel utilization when the number of interfaces per node
is fewer than the number of channels.
The notion of traffic-independent static channel assignment
for generating topology is proposed in [5]. Its purpose is to
assign a pair of neighboring nodes a common default channel
to communicate with each other and enable network topology.
They propose a centralized adaptive priority based channel
assignment algorithm that uses depth first search (DFS) algo-
rithm to assign channels. [6] proposes a channel assignment
for topology formation with two radio access point. Each AP
is assigned two channels: one for inter-cluster and another for
the intra-cluster communication. Thus a network will always
have same channels similar to the common channel protocol.
Optimization model for static channel assignment in wireless
mesh networks with multiple radios is developed in [7]. They
propose an ILP based formulation to generate optimal channel
assignment solutions. The objective is to maximize the number
of concurrent links constraints on the number of interfaces
and channels. Maximizing concurrent links increases the traffic
carrying capacity of the network.
In this paper, we propose a new distributed channel assign-
ment protocol, DCAP that induces the association between
neighboring nodes for generating a backbone topology. Our
scheme meets both ends of assigning common channels to the
neighboring MPs for enabling connectivity and reducing co-
 {CH2, CH1} {CH2, CH1}
PCL: <CH2, CH3>
maintaining a feasible network. Hence the problem is defined
1 4 1 4
OL -
as connectivity preserving interference bounded and proved to
OL 3 4 DL 1 2 PC be NP-complete [5].
DL - >2 L:
H
C
<C In this paper, we develop a distributed solution for channel
,3 H
H 1, assignment based on a heuristics such as node degree and
C C
:<L H node-ID. The degree of a node is a total number of its one-
3>
OL - C
DL 1 2
OL 3 4 P hop neighbors. Unlike CCA, we attempt to increase simulta-
DL -
neously operating links by utilizing available non-overlapping
3 2 3 2
PCL: <CH1, CH2> channels. Our scheme is independent of any specific traffic
{CH3, CH1} {CH3, CH1}
(a) (b) demand and can facilitate different adaptation modes including
Fig. 1: Example network with 4 nodes having 2 interfaces each the dynamic or hybrid channel switching schemes. Since the
and 3 channels {CH1,CH2,CH3}(a) Nodes with OL and DL focus is to generate a network topology which requires channel
values (b) Nodes with the highest priority transmit PCL to its assignment for a longer time period, other approaches such as
neighbors a dynamic channel assignment is considered beyond the scope
of this paper.
{CH2, CH1} {CH2, CH3} {CH2, CH1}
III. D ISTRIBUTED C HANNEL A SSIGNMENT P ROTOCOL
1 4 1 CH2 4
CH1 CH2 CH3 (DCAP)
PCFL 0 1 1
UCFL 0 1 1 Initially, degree and ID information of the neighbors, in-
CH3 CH1 cluding the number of interfaces are collected for ordering
nodes before the actual multi-channel assignment. For this
CH1 CH2 CH3
PCFL 2 0 0
purpose all nodes should have at least one of its interface
UCFL 0 1 1 assigned to a common channel for the entire network. In the
3 2 3 CH1 2
IEEE 802.11s based WLAN mesh network, this is achieved
{CH3, CH1} (a) {CH3, CH1} (b) {CH3, CH1} by using Simple Channel Unification Protocol [1]. Our multi-
Fig. 2: Example topology with 4 nodes in a grid for calculating channel assignment protocol (described below) is enabled in
the channel (a) Neighbor nodes with PCFL and UCFL data the next step after neighbor information is collected.
(b) Entire nodes set its candidate channels Exchanging neighbor information is performed by the
HELLO messages after a common channel is assigned to all
nodes in the initial step. After collecting node degree and
channel interference. We compare our scheme with the CCA ID information, ordering of neighbor nodes is performed in
and the results from [7]. In the following section we present the following manner. Each node maintains an ordered list
our network model. In section III we describe our algorithm (OL) that contains ID of its neighbors with higher degree
and present the performance evaluation results in section IV. compared to itself. It also stores the greater ID values with
In section V we conclude the paper with future works. the same degree. Rest of the neighbors with the lesser degree
and smaller ID with same degrees are stored in the dependent
II. N ETWORK M ODEL AND P ROBLEM D ESCRIPTION list (DL). Consider a network with 4 nodes each having two
We consider n wireless static nodes. Each node represents interfaces as shown in Fig. 1(a). Since all nodes have a same
MP in the WLAN mesh with multiple interfaces. We limit degree the OL is prepared by entering the node IDs that are
the number of interfaces to 2 for simplicity. Total number of greater than itself. The OL for Node 4 in the same figure
available channels is fixed. We consider 3 non-overlapping is empty because its neighbor’s ID are smaller than itself.
channels as in the IEEE 802.11b radios [8]. Transmission Similarly, the DL of Node 4 contains Node 1 and Node 2.
range is uniform for all nodes. It is defined as a range in which The node with an empty OL determine itself to be of the
a node can transmit a packet to a receiver. If other node pairs highest order and are considered as seed nodes.
that are within the transmission range and using same channel Each seed node first allocate candidate channels for its
initiate communication, they interfere with the communicating interfaces. A seed node constructs the candidate set by ran-
nodes which might cause a transmission failure. domly choosing I channels from their available channel set,
In a network with multi-interfaces, at least one interface of where I is the number of interfaces. Again referring to the
a node in a node-pair must be tuned to a common channel for Fig. 1(a) since the nodes have two interfaces, seed nodes 4
data transmission and reception. A channel assignment strat- and 3 randomly selects two channels from the channel set
egy should assign a common channel to all the neighboring {CH1, CH2, CH3}. Node 3 assigns CH1, CH2 and Node 4
pairs. Such a strategy would guarantee network connectivity assigns CH1 and CH2 in our running example. Seed nodes
and generate a feasible network topology. In the other hand, then constructs a priority channel list (PCL) with connecting
assigning the same channel in the entire network decreases and the unused channels. Connecting channel (CC) is the one
spatial reuse and increases interference. Again a given strategy among the candidate channels alternatively assigned for each
should minimize interference by using more channels while node in the DL. Unused channel (UC) is the one that are not
 3 1 2
2 3 1 2
2 3 3 1
1 2 1 3
3 2 1 2
3 2 2 1
1 2 3
(a) Concurrent links in DCAP
Fig. 3: Maximum concurrent links for DCAP and optimal results obtained from ILP with two interface nodes and three
orthogonal channels
among the candidate channels of the seed node. As shown in
Fig. 1(b), node 4 sends PCL < CH2, CH3 > with CH2 as
the CC and CH3 as the UC to node 1. In the same PCL, it
sends CH1 as the CC and CH3 as the UC to node 2. Similarly,
node 3 also prepares its PCL with CH1 for node 2, CH3 for
node 1 and CH2 as an unused channel.
Nodes with the non-empty OL start a wait timer and expect
PCL from all nodes in OL before it expires. They have
two lists, namely a preferred and unused channel frequency
lists (PCFL and UCFL) that maintains the frequency of the
connecting and unused channels received from the PCLs. The
frequency of the channels that are received as the CC is
recorded in PCFL whereas the UC is recorded in UCFL. Now,
referring to the Fig. 2(a), since node 1 receives CH2 and
CH3 as the connecting channels from node 4 and node3, their
respective frequencies are updated to one in PCFL. Similarly
the count of unused channels, CH2 from node 2 and CH3 from
node 4 is updated in the UCFL. After the PCL from all OL
nodes are received, the candidate channels are selected based
on the following three cases:
Best case occurs when |NCC | < |Nif s |, where |NCC |
denotes the number of connecting channels and |Nif s | denotes
number of interfaces in a node. In this case we have a
flexibility of choosing unused channels for better spatial reuse,
including the connecting channels for preserving the connec-
tivity. Thus, both CC and UC are included in the candidate set
for maximizing channel utilization. In our example, best case
happens with the node 2. Refer to Fig. 2(a), in which node 2
receives CH1 from both nodes 3 and 4 as its CC. Thus it selects
CH1 to connect with its higher order nodes and assigns CH3
for its another interface. In the network with more channels,
we select an unused channel with higher frequency.
Good case occurs when |NCC | == |Nif s |. In this case, we
assign all the channels in PCLF with frequency more than 1 to
the candidate set balancing the connectivity and interference.
This case occurs in our example case of node 1 as depicted in
the Fig. 2(a). Since node 1 have two connecting channels CH2
and CH3 in its PCFL, both of these channels are assigned in
2 3 1
2
1 3 1
1 2
3 1 1 3
1 1
3 1 3 1
3 3 2
(b) Concurrent links in Optimal solution
its candidate set. Note that, we cannot use unused channels,
which might cause network partition.
Worst case is when |NCC | > |Nif s |. In this case the number
of connecting channels is greater than the number of interfaces.
In this case, a node searches for a common set of channels
from all the candidate sets that are assigned to its seed nodes
interfaces for preserving the connectivity. This is performed by
listing all the channels in PCL sent by the nodes in OL to its
neighbors. Since PCL is broadcasted, each node receiving this
message can know the CCs sent to another node. For example,
in Fig. 1(b) node 1 can decode the preferred channel for node
2 from the same PCL.
After selecting the candidate channels, a node constructs
and transmit PCL to its DL nodes if it is nonempty. The
PCL arriving from the nodes in its DL are ignored. If the
DL is empty, node need not further transmit the PCL. Thus
our algorithm terminate when such nodes finally receive the
PCL. Fig. 2(b) shows the final channel assignment to all of
the available interfaces of nodes.
IV. P ERFORMANCE E VALUATION
We implemented the proposed solution, DCAP and CCA
protocol in ns-2 simulator version 2.29 [9]. We enhanced ns-2
to support multiple interfaces and channels.
A. Simulation Environment
N static wireless nodes in grid and random topology are
placed in 1000x1000m2 . Grid topologies has a fixed 200m
distance between nodes and the network size is changed
by increasing number of nodes from 3x3 to 6x6 grid. In
random topologies, nodes are randomly distributed however,
forming a connected graph with 10 15, 20, 25 and 30 nodes.
The transmission range for all nodes is fixed to 250m and
interference range to 550m [4]. Based on the IEEE 802.11b
specifications, we set 3 channels and 2 interfaces at each node.
We generated traffic for 50s with different inter-arrival time
from 0.0001s to 0.1s. Data packet size is 1024 bytes. First,
we computed the number of concurrent links to compare with
the optimal solutions obtained from [7]. In the next set of
 TABLE I: Number of concurrent links 10

8
Topology Total links DCAP CCA Optimal [7]

Throughput (Mbps)
4x4 24 12 8 12 6
5x5 40 18 14 18
6x6 60 27 18 27 4

2 DCAP
CCA

simulations, we randomly selected a node as a source and 0

flood data packets in the network after assigning channels. 1x10-3 5x10-3 10x10-3 50x10-3 100x10-3

Inter-arrival time (s)

We performed each simulation 5 times to measure throughput
and end-to-end latency when the network is saturated with Fig. 4: Average throughput in 6x6 grid
the traffic. Throughput is computed at each receiving node
by summing the total number of bytes received during the 4

simulation time. Similarly, we measure latency by reducing a

received time of a packet at a particular node by its generated 3

End-to-end Latency (s)

time at source. The results are compared with the CCA
protocol. CCA as explained earlier has a same pair of channels 2
assigned to each node interfaces.
B. Simulation Results 1

We first present the results obtained on the number of DCAP

CCA

concurrent links in the grid topologies. The total number 0

of available links are 24, 40 and 60 in the 4x4, 5x5 and
6x6 grid respectively. Table I shows the results obtained
from [7], CCA and our proposed scheme. Since our scheme
and [7] uses all three available channels for the network, more
concurrent links can co-exist for making multiple transmission
as compared to CCA. Even though CCA do not require extra
overhead for assigning channels, achieving higher network
capacity by using optimal channel assignment protocol such
as DCAP has long term benefit. Fig. 3 shows the topology for
DCAP and the [7] in 4x4 grid. The instance of the concurrent
links are depicted by the bold lines, in which 50% of links
can simultaneously operate. Similarly, 45% of links can be
concurrently active in 5x5 and 6x6 grid. We observe that this
reduction is due to the limited number of available channels for
the larger networks. Moreover, when the number of interfaces
in all nodes are equal to the number of channels, our protocol
generate same topology as CCA. Thus, one additional channel
actually increases the concurrent links by 14% in average.
We next describe the performance of the DCAP and CCA
with respect to the traffic flow in grid topologies. In grid
topology links are likely to be more uniform, thus there is
lesser influence of interference whereas in random topology,
nodes with more neighbors faces more interference due to
excessive channel reuse. Such links are thus a main bottleneck
for the flow in the network. With highest inter-arrival time we
show the results for random topology in the next part of this
section.
Following observations are made from our simulation re-
sults. Our DCAP scheme performs best in all cases. With re-
spect to inter-arrival time, DCAP shows 27% more throughput
gain than CCA as shown in Fig. 6 in 6x6 grid. This is because
our scheme uses more concurrent links to carry traffic. As the
inter-arrival rate increases, more number of packets drop in
1x10-3 5x10-3 10x10-3 50x10-3 100x10-3
Inter-arrival time (s)
Fig. 5: Average end-to-end latency in 6x6 grid
CCA due to the saturation of available links for forwarding
packets in the network.
Fig. 5 shows the end-to-end latency measured in 6x6 grid
topology with respect to traffic inter-arrival rate. DCAP shows
20% decrease in average latency compared to CCA. Due to the
repeated use of same channels in CCA that causes interference,
packets generated at a source might take longer route to reach
distant destinations. Increase in latency with more traffic in
the network is expected in both schemes. As the concurrent
link gets saturated, the packet gets forwarded to the destination
nodes through longer paths.
Rest of the simulation results we describe are measured
across the increasing number of nodes (network size) for grid
and random topologies. Network size increases the density
of nodes and links. New links using new channels are con-
structive and increase throughput whereas the ones that uses
same channels are destructive as they increase the collision
probability of packets. We use maximum traffic inter-arrival
rate of 0.0001s in all the following cases.
Fig. 6 and Fig. 7 shows the throughput result across the
network size in the grid and random networks respectively.
We observe DCAP shows 28% average increase in first and
30% in second over CCA. With the increasing network size
throughput gradually decreases as a result of increasing co-
channel interference. Channel limitation causes more destruc-
tive links as have been described earlier. Because of some level
of uniformity, throughput achieved in grid topology is higher
than the random topology.
 20 4
DCAP
DCAP
CCA
CCA

15 3

End-to-end Latency (s)

Throughput (Mbps)

10 2

5 1

0 0
3x3 4x4 5x5 6x6 3x3 4x4 5x5 6x6
Network size Network size

Fig. 6: Average throughput in grid topologies Fig. 8: Average end-to-end latency in grid topologies
6
10 DCAP
CCA DCAP
CCA
5
8

End-to-end Latency (s)

Throughput (Mbps)

6
3

4
2

2
1

0
0
10 15 20 25 30
10 15 20 25 30
Network size
Network size

Fig. 7: Average throughput in random topologies Fig. 9: Average end-to-end latency in random topologies

From Fig. 8 and Fig. 9 we see the consistent performance
of DCAP over CCA. Latency in the random topology (Fig. 9)
is relatively higher than in the grid topology. In average, the
latency of DCAP is 3.2s and CCA is 4.1s in random topology
whereas DCAP has 2.4s and CCA has 2.9s in grid topology.
In random topology, packet delivery is often unsuccessful due
to heavy interference. In case of CCA, this gets worst with
limited use of channels and thus takes longer path to reach
the destination. We also observe that latency steadily increase
with the network size. The results are as expected because of
lesser spatial reuse in larger networks.
V. C ONCLUSION
We present the distributed channel assignment protocol in
multi-interface and multichannel wireless mesh networks. Our
method is based on simple but effective heuristics such as
neighbor degree and node identification. We showed that the
priority based selection of the channels can be performed in
a distributed manner for forming the wireless mesh network
topology with minimum interference. We also showed that
our channel assignment algorithm produces similar number of
concurrent links as generated by the ILP based protocol [7].
Our results show 30% of increased network throughput
compared to CCA and also reduces per-packet latency by 3s in
best cases. In general, we conclude that the advantage of being
selective in channel selection offers better network capacity.
For the network in which channel switching can be costlier
in terms of delay, channel assignment such as this proves to
be better alternative for the network with multiple interfaces
and channels. In future, we will investigate the performance of
our protocol in the network with mixed number of interfaces.
We will also study other topology control aspects such as
transmission power control with the channel assignments.
R EFERENCES
[1] IEEE P802.11s/D1.01, Draft Amendment to Standard for Information
Technology - Telecommunications and Information Exchange Between
Systems - LAN/MAN Specific Requirements - Part 11: Wireless Medium
Access Control (MAC) and physical layer (PHY) specifications: Amend-
ment: ESS Mesh Networking, March 2007
[2] A. H. M. Rad and Vincent W.S., Wong, ”Logical Topology Design and
Interference Assignment for Multi-Channel Wireless Mesh Networks,”
IEEE Globecom, 2006
[3] Jian Tang, Guoling Xue and Weiyi Zhang, ”Interference Aware Topology
control and QoS Routing in Multi-channel Wireless Mesh Networks,”
ACM MobiHoc 2005
[4] A. Raniwala and T. Chiueh, ”Architechture and algorithms for an IEEE
802.11-based multi-channel wireless mesh network,” Proceedings of IEEE
INFOCOM’2005
[5] Mahesh K. Marina and Samir R. Das, ”A Topology Control Approach
for Utilizing Multiple Channels in Multi-Radio Wireless Mesh networks,”
IEEE Broadnets 2005
[6] Jing Zhu, and Sumit Roy, ”802.11 Mesh Networks with Two Radio
Access Points,” IEEE ICC 2005
[7] Arndam K. Das, Hamed M. K. Alazemi, Rajiv Vijaykumar, and Sumit
Roy, ”Optimization Models for Fixed Channel Assignment in Wireless
Mesh Networks with Multiple Radios,” IEEE SECON 2005
[8] IEEE Standard for Local and metropolitan area networks, IEEE Std
802.11, 1999.
[9] www.isi.edu/nsnam/ns/